In Vitro and In Vivo Correlation of Bone Morphogenetic Protein-2 Release Profiles from Complex Delivery Vehicles

Abstract

Local sustained delivery of bioactive molecules from biomaterials is a promising strategy to enhance bone regeneration. To optimize delivery vehicles for bone formation, the design characteristics are tailored with consequential effect on bone morphogenetic protein-2 (BMP-2) release and bone regeneration. Complying with the 3R principles (Replacement, Reduction, and Refinement), the growth factor release is often investigated in vitro using several buffers to mimic the in vivo physiological environment. However, this remains an unmet need. Therefore, this study investigates the in vitro-in vivo correlation (IVIVC) of BMP-2 release from complex delivery vehicles in several commonly used in vitro buffers: cell culture model, phosphate buffered saline, and a strong desorption buffer. The results from this study showed that the release environment affected the BMP-2 release profiles, creating distinct relationships between release versus time and differences in extent of release. According to the guidance set by the U.S. Food and Drug Administration (FDA), IVIVC resulted in level A internal predictability for individual composites. Since the IVIVC was influenced by the BMP-2 loading method and composite surface chemistry, the external predictive value of the IVIVCs was limited. These results show that the IVIVCs can be used for predicting the release of an individual composite. However, the models cannot be used for predicting in vivo release for different composite formulations since they lack external predictability. Potential confounding effects of drug type, delivery vehicle formulations, and application site should be added to the equation to develop one single IVIVC applicable for complex delivery vehicles. Altogether, these results imply that more sophisticated in vitro systems should be used in bone regeneration to accurately discriminate and predict in vivo BMP-2 release from different complex delivery vehicles.

Introduction

Local sustained delivery of bioactive molecules from biomaterials is a promising strategy to enhance bone regeneration. Many studies have shown enhancement of bone formation in ectopic and orthotopic locations by sustained release of various growth factors.¹ Despite these promising results, improvement of the local delivery vehicles and optimization of the growth factor release profile remains a challenge.

New technologies have enhanced tailoring of release profiles within the challenging topic of designing the appropriate delivery vehicle for clinical application in bone regeneration. The biomaterials fulfill both a scaffold and delivery role, and need to meet various physical, mechanical, biological, and chemical demands. Tailoring the design characteristics may influence the growth factor release^2–5 with consequent differential effect on bone regeneration. As such, extensive in vivo animal studies are needed to analyze the release profiles and efficacy of the various delivery vehicles.

In contemporary scientific practice the 3R principles (Replacement, Reduction, and Refinement) are warranted. Therefore, to estimate the release kinetics, growth factor release is often investigated in vitro using several buffers to mimic the in vivo physiological environment.^6–8 Nonetheless, various studies have showed that in vitro release cannot be extrapolated to in vivo release.^9–11 To the best of our knowledge, there are no studies available that have attempted to correlate the in vitro and in vivo release profiles in the field of bone tissue engineering, and as such the relevance of these in vitro profiles for future clinical applications remains unknown. Therefore, the aim of this study is to investigate the in vitro-in vivo correlation (IVIVC) of growth factor release in several commonly used in vitro buffers and to develop a predictive model providing a standardized method employing a commonly used growth factor for these purposes.

IVIVC is defined by the U.S. Food and Drug Administration (FDA) as “a predictive mathematical model describing the relationship between an in vitro property of a dosage form and relevant in vivo response.”¹² Several levels are described by the FDA protocol,¹³ including level A up to D. Level A correlation represents a point-to-point relationship between in vitro and in vivo profiles. Level A correlation is considered most informative and is recommended by the FDA. It is the only level that can be used to obtain biowaiver. Level B correlation is based on the principles of statistical moment analysis but is devoid of a point-to-point correlation and hence does not reflect the actual in vivo release profile. As such, this level may lack sufficient predictability. Level C correlation establishes a single point relationship between dissolution and a pharmacokinetic parameter. Since it is based on a single point analysis, it is does not reflect the complete shape of the plasma concentration time curve, which is critical to define in vivo performance of the studied drug. Multiple Level C correlation relates multiple dissolution time points to one or more pharmacokinetic parameter(s) and should be based on at least three dissolution time points covering the early, middle, and late stages of the dissolution profile. A multiple Level C correlation can be as useful as a Level A correlation. However, if a multiple Level C correlation is achieved, the development of a Level A correlation is feasible and preferred. Level D correlation is a rank order correlation comparing in vitro and in vivo release profiles. A level D correlation is only qualitative and is not adopted in the FDA IVIVC Guidance.

Given that level A IVIVC is the most informative, this study employs this correlation level to accurately predict the in vivo performance from the in vitro performance. Bone morphogenetic protein-2 (BMP-2), a promising bone formation inducing agent, was used and released from several biomaterial composites to investigate the internal and external predictability of the IVIVC. The IVIVC for locally delivered drugs, such as BMP-2, is challenging due to complex characteristics of the biomaterials and lack of a standardized in vitro model. Therefore, level A IVIVC was analyzed for composites with various characteristics releasing BMP-2 in several in vitro models as a first step toward determining the proper in vitro system and developing a predictive model for in vivo release of BMP-2. Furthermore, to make a qualitative assessment of the IVIVC, a rank order level D IVIVC between the different composites was performed.

Materials and Methods

Experimental design

To investigate the relationship between in vitro and in vivo BMP-2 release kinetics, various composites with predicted differential release profiles were studied in three in vitro models. The composites were based on 75% porous oligo(polyethelene glycol) fumarate (OPF) hydrogels (22.5% w/w) containing 2.5% (w/w) poly(lactic-co-glycolic acid) (PLGA) microspheres. The BMP-2 loading method and hydrogel chemical properties were modified to achieve distinct BMP-2 release kinetics. These results were described in previous articles.^14,15 BMP-2 was encapsulated in PLGA microspheres and/or adsorbed on the hydrogel to create hydrogels with different burst and sustained release of BMP-2. Apart from the different loading methods, hydrogel chemistry was modified to further tailor BMP-2 release by cross-linking sodium methacrylate (SMA; Sigma-Aldrich, St. Louis, MO), [2-(methacryloyloxy) ethyl]-trimethylammonium chloride (MAE; Sigma-Aldrich) or bis[2-(methacryloyloxy)ethyl] (BP) into the hydrogel to obtain negatively charged OPF (n-OPF), positively charged OPF (p-OPF) and phosphate modified hydrogels OPF (Ph-OPF), respectively. These modifications resulted in a total of 12 different composites (Table 1).

Table 1.

Summary of Composite Characteristics Used in the In Vitro and In Vivo Bone Morphogenetic Protein-2 Release Studies

	BMP-2 loading
Composite name	Hydrogel	Microspheres	Initial activity/implant (μCi)	BMP-2/implant (μg)	In vitro buffer
-OPF-Msp	0% Adsorbed	100% Loaded	3.7 ± 0.8	4.7 ± 1.0	CC
-OPF-Cmb	50% Adsorbed	50% Loaded	3.5 ± 0.2	4.6 ± 0.2	CC, PBS, SDB
-OPF-Ads	100% Adsorbed	0% Loaded	3.8 ± 0.2	4.9 ± 0.3	CC, PBS, SDB
n-OPF-Msp	0% Adsorbed	100% Loaded	3.0 ± 0.5	3.8 ± 0.6	CC
n-OPF-Cmb	50% Adsorbed	50% Loaded	2.7 ± 0.2	3.6 ± 0.2	CC, PBS, SDB
n-OPF-Ads	100% Adsorbed	0% Loaded	3.0 ± 0.1	4.0 ± 0.1	CC, PBS, SDB
p-OPF-Msp	0% Adsorbed	100% Loaded	2.4 ± 0.3	3.1 ± 0.4	CC
p-OPF-Cmb	50% Adsorbed	50% Loaded	2.3 ± 0.1	3.0 ± 0.2	CC, PBS, SDB
p-OPF-Ads	100% Adsorbed	0% Loaded	2.5 ± 0.1	3.2 ± 0.2	CC, PBS, SDB
Ph-OPF-Msp	0% Adsorbed	100% Loaded	3.1 ± 0.1	4.0 ± 0.1	CC
Ph-OPF-Cmb	50% Adsorbed	50% Loaded	3.0 ± 0.2	3.9 ± 0.3	CC
Ph-OPF-Ads	100% Adsorbed	0% Loaded	3.1 ± 0.6	4.0 ± 0.7	CC

Ads, adsorbed BMP-2; BMP-2, bone morphogenetic protein-2; CC, in vitro cell culture model; Cmb, combined microsphere encapsulated and adsorbed BMP-2; Msp, microsphere encapsulated BMP-2; n-OPF, negatively charged OPF; OPF, oligo(polyethelene glycol) fumarate; -OPF, unmodified OPF; PBS, phosphate buffered saline; Ph-OPF, phosphate modified OPF; p-OPF, positively charged OPF; SDB, strong desorption buffer.

The in vitro and in vivo BMP-2 release was evaluated by employing BMP-2 radiolabeled with ¹²⁵I. To simulate the cell-rich in vivo environment, in vitro release of all composites was investigated using a cell culture setup for 8 weeks. To analyze the influence of different commonly used in vitro buffers on in vitro BMP-2 release, a subset of composites was immerged in a cell-free environment in the presence of phosphate buffered saline (PBS) or a strong desorption buffer (SDB). The in vivo BMP-2 release was analyzed in a subcutaneous rat model with 8 weeks' follow-up. Subsequently, the IVIVC level A was investigated for all in vitro release systems.

BMP-2 radioiodination

Carrier-free Na¹²⁵I was obtained from PerkinElmer Life and Analytical Sciences (Boston, MA). To study the release profiles of BMP-2, a fraction of the incorporated BMP-2 was radiolabeled with ¹²⁵I, using the chloramine-T method as previously described.¹⁶ The radiolabeled BMP-2 was separated from the free ¹²⁵I by 24-h dialysis (10 kDa molecular weight cutoff [MWCO]; SpectraPor 7, Rancho Dominguez, CA) against 0.01 M phosphate buffered saline at pH 7.4 (Sigma-Aldrich). The ¹²⁵I-BMP-2 dialysate was concentrated in a Millipore device (10 kDa MWCO, Billerica, MA) and the purity was determined by trichloroacetic acid precipitation. The final ¹²⁵I-BMP-2 preparation contained 99.7% precipitable counts, which indicated the percentage of covalently bound ¹²⁵I to the BMP-2. Thereafter, ¹²⁵I-BMP-2 was mixed with nonlabeled BMP-2 (1:5.3 hot–cold ratio) and incorporated into the composite formulations.

Microsphere fabrication

PLGA 50:50 (M_w 52 kDa; Lakeshore Biomaterials) microspheres were fabricated using a double-emulsion-solvent-extraction (W1-O-W2) technique according to a previously described method.¹⁷ Briefly, an aqueous solution containing 130 μL (OPF-Msp), 65 μL (OPF-Cmb), or 0 μL (OPF-Ads) of 3.7 mg/mL ¹²⁵I-BMP-2/BMP-2 (1:5.3 hot:cold ratio) solution was emulsified with 250 mg PLGA 50:50 dissolved in 1.25 mL of dichloromethane using a vortex at 3050 rpm. The solution was re-emulsified in 2 mL of 2% (w/v) aqueous poly(vinyl alcohol) (PVA, 87–89% mole hydrolyzed, M_w = 13,000–23,000; Sigma-Aldrich) to create the double emulsion and added to 100 mL of a 0.3% (w/v) PVA solution and 100 mL of a 2% (w/v) aqueous isopropanol solution. After 1 h of slow stirring, the PLGA microspheres were collected by centrifugation at 2500 rpm for 3 min, washed three times with distilled deionized water (ddH₂O), and freeze dried to a free-flowing powder. The characteristics of the PLGA microspheres were reported in a previous study.¹⁴ The diameter of both unloaded and BMP-2-loaded microspheres was distributed between 0 and 100 μm. The PLGA microspheres used in this study lose ∼80% of their mass within 4 weeks after implantation.¹⁸

Fabrication of composites

OPF was fabricated using polyethylene glycol (PEG) with an initial molecular weight of 10 kDa according to previously described method.¹⁹ OPF (44% w/w), N-vinyl pyrrolidinone (13% w/w [NVP]; Sigma-Aldrich), Irgacure 2959 (0.2%; Ciba-Specialty Chemicals, Tarrytown, NY), and H₂O (42% w/w) were mixed with either SMA (200 mg, 8.2% w/w; Sigma-Aldrich), MAE (225 mg, 10.3% w/w; Sigma-Aldrich), or no additive to create hydrogels with a fixed negative (n-OPF), positive (p-OPF), or neutral (-OPF) charge, respectively. For Ph-OPF, OPF (41% w/w), NVP (29% w/w; Sigma-Aldrich), BP (8.2% w/w), and Irgacure 2959 (0.2% w/w; Ciba-Specialty Chemicals) were dissolved in deionized water (21.6% w/w).

To create the composites, the OPF/NVP, OPF/NVP/SMA, OPF/NVP/MAETAC, or OPF/NVP/BP paste (22.5% w/w) was mixed with NaCl salt particles (75% w/w, sieved to a maximal size of 300 μm) and PLGA microspheres (2.5% w/w). The resulting mixture was forced into a cylindrical mold with a diameter of 3.5 mm and exposed to ultraviolet light (365 nm at intensity of ∼8 mW/cm² black-Ray Model 100AP, Upland, CA) to cross-link the composites for 40 min in total. The composite implants were cut into 6 mm long rods, sterilized by ethanol evaporation for both the in vitro and in vivo experiments, and immersed in sterile ddH₂O to leach out the salt. After blot drying, additional BMP-2 was loaded on the composite matrix by adsorption for the OPF-Cmb and OPF-Ads scaffolds. By varying the BMP-2 loading method, three different composite implants were created consisting of 100% of the BMP-2 encapsulated in PLGA microspheres (OPF-Msp, sustained release), 50% of the BMP-2 encapsulated in PLGA microspheres, and 50% adsorbed on the composite (OPF-Cmb, combined burst and sustained release), and 100% adsorbed on the composite (OPF-Ads, mainly burst release) (Table 1).

The bioactivity of the released BMP-2 was reported previously and showed a similar bioactivity for the microsphere encapsulated and adsorbed growth factor after 9 weeks of release.¹⁴ Also, the released BMP-2 generated a similar biologic response compared to freshly added BMP-2 of corresponding dose in vitro. The degradation rate of the OPF hydrogel is reported previously and was slow with minimal in vitro degradation of cross-linked hydrogels with an OPF:NVP ratio >0.3 after 21 days in PBS.²⁰ Furthermore, histology shows a still visible porous structure of OPF after 9 weeks of implantation, as opposed to a fully resorbed Infuse^® absorbable collagen sponge (Medtronic, Minneapolis, MN).¹⁴

In vivo release measurements

Thirty-two 12-week-old male Harlan Sprague Dawley rats were used for this study according to an approved protocol by the local animal care and use committee. Animal studies were previously published.^14,15 Surgery was performed under sterile conditions and general anesthesia (ketamine/xylazine, 45/10 mg/kg). After shaving and disinfecting the surgical sites, subcutaneous pockets were created in each limb and filled with ¹²⁵I-BMP-2-loaded implants. Two subcutaneous pockets in the thoracolumbar region were used to implant the controls (unloaded implants). Acetaminophen (160 mg in 5 mL added to pint water bottle) was given as postoperative analgesia for the duration of 1 week. Four scintillation probes (model 44–3 low energy gamma scintillator; Ludlum Measurements, Inc.) connected to digital scalers (Model 1000 scaler; Ludlum Measurements, Inc.) as described previously,¹⁷ were used for determining in vivo ¹²⁵I-BMP-2 release kinetics. Directly after wound closure, the ¹²⁵I-BMP-2 activity was measured to determine the starting implanted dose. At each subsequent time point (biweekly the first week, weekly from week 1 onward), the rats were anesthetized using isoflurane (induction 4%, maintenance >1.5%) to measure the local ¹²⁵I-BMP-2 activity in duplicate over two 1-min periods. To determine the BMP-2 release, the ¹²⁵I-BMP-2 measurements were corrected for radioactive decay and background activity. The ¹²⁵I-BMP-2 activity was normalized to the starting implanted dose to determine the retained ¹²⁵I-BMP-2 dose and released amounts. After 9 weeks, the rats were euthanized by CO₂ asphyxiation.

In vitro BMP-2 release

BMP-2 release was analyzed using a W20-17 cell culture system. The composites were exposed to consecutive 7-day cell cultures (seeded at 20,000 cells per cm² in a 24-well plate) in 1 mL Dulbecco's Modified Eagle's Medium/Nutrient Mixture F-12 Ham 1:1 mixture (DMEM/F12; Sigma-Aldrich) supplemented with 10% fetal bovine serum and 1% penicillin at 37°C, 20% O₂, 5% CO₂. A subset of composites was analyzed for BMP-2 release in the presence of PBS (pH 7.4) or SDB (solution containing 0.5 M arginine, 0.5 M NaCl and 50 mM K₂HPO₄ in ddH₂O at pH of 7.5). The composites were placed in 1.0 mL PBS or SDB containing Eppendorf tubes and maintained at 37°C in the incubator. At weeks 0.5, 1, 2, 3, 4, 5, 6, 7, and 8, the culture medium was collected and replaced with fresh DMEM/F12, PBS or SDB. To determine the BMP-2 release, the collected culture medium was assayed for ¹²⁵I activity on a gamma counter. At the end of the study, the composites were collected to determine the remaining ¹²⁵I activity as a measure of retained BMP-2. All ¹²⁵I activity measurements were corrected for decay and normalized to the starting amount. The corresponding in vitro release profiles were determined by correlating the gamma-irradiation in counts/minute to the amount of BMP-2 released from the composites.

In vitro-in vivo correlation

The data acquired in the release studies were used to develop the IVIVC. The IVIVC was obtained by correlating the composites cumulative release in vitro with the cumulative in vivo release. Regression analysis was applied to the IVIVC plots and the corresponding equations described. To analyze the external predictive value of the in vitro models, superimposability of the IVIVCs was tested for composites with similarly expected release mechanisms. Therefore, OPF-Msp composites (expected to release BMP-2 mainly by polymer degradation and diffusion), OPF-Cmb composites (expected to release BMP-2 mainly by polymer degradation, diffusion, desorption, and ion exchange), and OPF-Ads composites (expected to release BMP-2 mainly by desorption and ion exchange) were analyzed for external predictability separately. To accurately predict the in vivo BMP-2 release profile, a point-to-point IVIVC is needed. Therefore, levels B and C were considered not suitable for developing a clinically relevant predictive model. To make a qualitative assessment of the IVIVCs, level D IVIVC was analyzed using the cumulative BMP-2 release of the different composites in the various release environments.

Statistical analysis

Statistical analysis was performed using Prism 7 (Graphpad Software, La Jolla, CA) and SPSS 22.0 software (SPSS, Inc., Chicago, IL). In vitro results (n = 3 per composite type studied) and in vivo results (n = 10 for -OPF, n-OPF, p-OPF, and Ph-OPF-Cmb; and n = 8 for Ph-OPF-Msp and Ph-OPF-Ads) of BMP-2 release are given as mean ± standard deviation. The BMP-2 release was analyzed point-to-point for a period of 8 weeks in vitro and in vivo. All datasets were tested for outliers using Hoaglin's outlier labeling rule,²¹ for normality of the residuals using the Shapiro-Wilk test and for homogeneity of variances using the Levene's test. Parametric data were analyzed with univariate analysis of variances and Benjamini Hochberg post hoc. Radar diagrams were used to illustrate the extent of cumulative BMP-2 release and ranking of the different composites in the various release environments (in vitro and in vivo). To analyze the predictive value of in vitro release for in vivo BMP-2 release different regression models were analyzed for fit using R².

Results

BMP-2 labeling and incorporation

Labeling of the BMP-2 with ¹²⁵I resulted in an activity per mass of 6.1 μCi/μg. The microspheres were loaded with either 2.9 μg BMP-2/mg PLGA (OPF-Msp) or 1.3 μg BMP-2/mg PLGA (OPF-Cmb). The composite scaffold characteristics are summarized in Table 1. The different loading methods resulted in comparable BMP-2 loading per scaffold. OPF modifications aiming at composite chemistry resulted in differences in BMP-2 loading (<1.9 μg BMP-2) within chemistry modifications due to differences in BMP-2 loss during the fabrication process (Table 1). To correct for this, the release kinetics were corrected for the starting amount of BMP-2 loading for each individual composite both in vitro and in vivo.

Animals

Five rats died 1 day after surgery, probably due to oversensitivity to xylazine, since no health problems were observed in the remainder of the rats after the xylazine was lowered. Nine implants were removed by the rats themselves from the subcutaneous pocket during the follow-up and were therefore discarded from further analysis. Detailed information on sample size used for each analysis is provided in Supplementary Table S1 (Supplementary Data are available online at www.liebertpub.com/tec).

In vivo and in vitro release profiles

As expected, the composites showed different in vivo BMP-2 release profiles. Most composites demonstrated a triphasic in vivo release profile with a large burst release phase (Phase 1, till week 0.5), high dose sustained release phase (Phase 2, weeks 0.5–5) and a low dose zero order release phase (Phase 3, week 5 onward) (Fig. 1A). After the large burst release, various release patterns were seen for the different OPF-composites in phase 2, including exponential growth, exponential decay and linear relationships of BMP-2 release versus time (Fig. 2). In Phase 3 all composites showed a linear relationship with different BMP-2 release rates. Ph-OPF-Msp and Ph-OPF-Cmb exhibited a tetraphasic in vivo release profile, with the high sustained release phase 2 divided into a lag phase and rapid release phase (Fig. 2).

FIG. 1.

Cumulative release of BMP-2 from various polymer composites in vivo in a rat subcutaneous implantation model (A) or [CC (B), PBS (C), SDB (D)]. The release profiles are divided into three phases, burst release (phase 1), high dose release sustained release (phase 2), and low dose sustained release (phase 3). Ads, adsorbed BMP-2; BMP-2, bone morphogenetic protein-2; CC, in vitro cell culture model; Cmb, combined microsphere encapsulated and adsorbed BMP-2; Msp, microsphere encapsulated BMP-2; n-OPF, negatively charged OPF; OPF, oligo(polyethelene glycol) fumarate; -OPF, unmodified OPF; PBS, in vitro phosphate buffered saline; Ph-OPF, phosphate modified OPF; p-OPF, positively charged OPF; SDB, in vitro strong desorption buffer model.

FIG. 2.

Relationship between the cumulative release of BMP-2 (%) in vivo in a rat subcutaneous model and time for the low dose sustained release phase (phase 2). Various release patterns were observed for the different OPF-composites in phase 2, including exponential growth, exponential decay, and linear relationships of BMP-2 release versus time.

Although different release profiles were observed in vitro compared to in vivo, the triphasic time frames were similar. In the cell culture set-up, the composites exhibited different burst releases in phase 1, exponential decay release profiles with large variability in phase 2, and zero order release in phase 3 (Fig. 1B). In PBS, a minimal burst release was observed for all composites, a near linear exponential decay release profile in phase 2, and a zero-order release in phase 3 (Fig. 1C). Minimal differences were observed between the composites in SDB. After the burst release with minimal variability, similar exponential decay patterns were seen for all composites during phases 2 and 3 (Fig. 1D). Since every phase showed different relationships for BMP-2 release versus time, the phases were analyzed for IVIVC level A separately. Phases 1 and 2 were considered most relevant since in vivo most composites released >80% BMP-2 till week 5 and low dose release was observed from week 5 onward for all composites.

In vitro-in vivo correlations

Level A IVIVC

All phases

For the cell culture model, OPF-Msp composites showed linear regression relationship (R² > 0.90) with residuals <19.6% (Fig. 3A). Similarly, OPF-Cmb composites showed linear regression relationships (R² > 0.91) with residuals <14.2% between in vitro and in vivo in the cell culture model (Fig. 3B). The OPF-Ads composites showed exponential decay regression relationships (R² > 0.99) with residuals <3.0% (Fig. 3C). For the cell-free PBS buffer, OPF-Cmb composites showed linear regression relationships (R² > 0.95) with residuals <9.4% (Fig. 3D). OPF-Ads composites showed an exponential decay relationship (R² > 0.99) with residuals <1.0% between in vitro and in vivo (Fig. 3E). For the cell-free SDB buffer, OPF-Cmb composites showed linear and exponential regression relationships (R² > 0.99) with residuals <13.1% (Fig. 3F). The OPF-Ads composites showed an exponential decay relationship (R² > 0.99) with residuals <1.8% between in vitro and in vivo (Fig. 3G).

FIG. 3.

Level A in vitro (CC, PBS, and SDB)-in vivo correlations (IVIVCs) of cumulative BMP-2 release (%) for the different OPF composites for all time points. The figure is horizontally divided into different in vitro models and vertically divided into different BMP-2 loading methods.

The confounding effect of the surface chemistry of the composites on the relationship, slope, span, and X- and Y-intercepts of the IVIVCs, influenced the regression models of all IVIVCs (Fig. 3). The acquired models were not superimposable and therefore lacked external predictability. Since a large area had to be extrapolated for the large burst release phase, the burst release phase and sustained release phase were analyzed for IVIVC separately.

Phase 1 (burst release phase)

In the cell culture setup, the burst release from the different composites showed a strong linear correlation (R² = 0.68) with the in vivo burst release and residuals <11% released BMP-2 (Fig. 4A, B). In the subset of composites the correlation between in vitro and in vivo BMP-2 burst release was very weak in PBS (R² = 0.06) and SDB (R² = 0.07), and moderate in the cell culture setup (R² = 0.45) (Fig. 4C, D).

FIG. 4.

Level A in vitro (CC, PBS, and SDB)-in vivo correlations (IVIVCs) of the BMP-2 burst release phase 1 for the all OPF composites (A, B) and a subset of composites (OPF-Cmb/Ads, n-OPF-Cmb/Ads, and p-OPF-Cmb/Ads) (C, D) with corresponding residuals.

Phases 2 and 3 (sustained release phase)

For the cell culture model (Table 2), the OPF-Msp composites showed a sigmoidal regression relationship (R² > 0.99) with residuals <2.2% (Fig. 5A). Similarly, the OPF-Cmb composites showed sigmoidal regression relationship (R² > 0.98) with residuals <4% (Fig. 5B). A linear regression relationship (R² > 0.97) was observed for OPF-Ads composites with residuals <1.8% (Fig. 5C). For the cell-free PBS in vitro model (Table 3), OPF-Cmb composites showed sigmoidal regression relationship (R² > 0.99) with residuals <3.1% (Fig. 5D). OPF-Ads composites showed exponential decay regression relationships (R² > 0.99) with residuals <0.6% (Fig. 5E). For the cell-free SDB in vitro model (Table 3), OPF-Cmb composites showed sigmoidal (R² > 0.99, -OPF, n-OPF) and exponential growth (R² > 0.94, p-OPF) regression relationships with <0.7% (-OPF, n-OPF) and <8.7% released BMP-2 (p-OPF) residuals (Fig. 5F). OPF-Ads composites showed an exponential decay relationship (R² > 0.99) with residuals <0.6% (Fig. 5G).

Table 2.

Internal Predictive In Vitro-In Vivo Correlation Level A Characteristics for the Sustained Release Phase in a Cell Culture Model

Composite name	Relationship	R2	Model
-OPF-Msp	Sigmoidal	0.998	Y = 51 + (98 − 51)/(1 + 10^[{49 − X} × 0.08])
-OPF-Cmb	Sigmoidal	0.998	Y = 65 + (98 − 65)/(1 + 10^[{58 − X} × 0.08])
-OPF-Ads	Linear	0.986	Y = 0.38 × X + 62.25
n-OPF-Msp	Sigmoidal	0.999	Y = 41 + (97 − 41)/(1 + 10^[{27 − X} × 0.06])
n-OPF-Cmb	Sigmoidal	0.999	Y = 54 + (99 − 54)/(1 + 10^[{44 − X} × 0.06])
n-OPF-Ads	Linear	0.989	Y = 0.19 × X + 81.17
p-OPF-Msp	Sigmoidal	0.997	Y = 35 + (95 − 35)/(1 + 10^[{34 − X} × 0.1])
p-OPF-Cmb	Sigmoidal	0.991	Y = 45 + (96 − 45)/(1 + 10^[{53 − X} × 0.2])
p-OPF-Ads	Linear	0.971	Y = 0.43 × X + 56.85
Ph-OPF-Msp	Sigmoidal	0.995	Y = 28 + (107 − 28)/(1 + 10^[{35 − X} × 0.09])
Ph-OPF-Cmb	Sigmoidal	0.981	Y = 52 + (160 − 52)/(1 + 10^[{57 − X} × 0.07])
Ph-OPF-Ads	Linear	0.998	Y = 0.28 × X + 75.15

Models were based on a sigmoidal function Y = bottom + (top − bottom)/(1 + 10^[{logEC50 − X} × hillslope]) and a linear function Y = slope × X + Y-intercept.

FIG. 5.

Level A in vitro (CC, PBS, and SDB)-in vivo correlations (IVIVCs) of cumulative BMP-2 release (%) for the different OPF composites for the sustained release phase (phase 2, 3). The figure is horizontally divided into different in vitro models and vertically divided into different BMP-2 loading methods.

Table 3.

Internal Predictive In Vitro-In Vivo Correlation Level A Characteristics for the Sustained Release Phase in Cell-Free In Vitro Buffers

Release buffer	Composite name	Relationship	R2	Model
PBS	-OPF-Cmb	Sigmoidal	0.999	Y = 64 + (97 − 64)/(1 + 10^[{23 − X} × 0.2])
	-OPF-Ads	Exponential decay	0.999	Y = (−64 − 96) × exp(−0.3 × X) + 96
	n-OPF-Cmb	Sigmoidal	0.999	Y = 54 + (96 − 54)/(1 + 10^[{18 − X} × 0.4])
	n-OPF-Ads	Exponential decay	0.992	Y = (17 − 98) × exp(−0.2 × X) + 98
	p-OPF-Cmb	Sigmoidal	0.992	Y = 44 + (98 − 44)/(1 + 10^[{9 − X} × 0.4])
	p-OPF-Ads	Exponential decay	0.996	Y = (−88 − 102) × exp(−0.2 × X) + 102
SDB	-OPF-Cmb	Sigmoidal	0.999	Y = 60 + (106 − 60)/(1 + 10^[{70 − X} × 0.03])
	-OPF-Ads	Exponential decay	0.999	Y = (33 − 99) × exp(−0.03 × X) + 99
	n-OPF-Cmb	Sigmoidal	0.999	Y = 44 + (114 − 44)/(1 + 10^[{70 − X} × 0.02])
	n-OPF-Ads	Exponential decay	0.998	Y = (74 − 100) × exp(−0.02 × X) + 100
	p-OPF-Cmb	Exponential growth	0.937	Y = 25 × exp(0.01 × X)
	p-OPF-Ads	Exponential decay	0.997	Y = (38 − 100) × exp(−0.04 × X) + 100

Models were based on a sigmoidal function Y = bottom + (top − bottom)/(1 + 10^[{logEC50 − X} × hillslope]), an exponential decay function Y = (Y0 − plateau) × exp(−K × X) + plateau, an exponential growth function Y = Y0 × exp(K × X), and a linear function Y = slope × X + Y-intercept.

The confounding effect of the surface chemistry of the composites on the relationship, slope, span, and X- and Y-intercepts of the IVIVCs, influenced the regression models of all IVIVCs (Fig. 5). The acquired models were not superimposable and therefore lacked external predictability.

Level D IVIVC

Level D IVIVC of BMP-2 release in the various conditions is shown per time-point in radar diagrams (Fig. 6) to analyze the rank order and extent of release of the different composites in the various release environments. The burst release varied in extent of release and ranking for the different composites. The variation in burst release between the different composites was minimal in the cell-free PBS and SDB, whereas large differences were observed between the composites in the cell culture model and in vivo. Furthermore, the ranking of the implants differed per release environment. For example, n-OPF-Ads had the highest burst release in vivo, but ranked fourth in cell culture set up, third in PBS, and fifth in SDB. Differences in ranking and extent of cumulative BMP-2 release were also observed for the different release environments in the sustained release phase up to 8 weeks.

FIG. 6.

Level D in vitro (CC, PBS, and SDB)-in vivo correlations (IVIVCs) of cumulative BMP-2 release (%) for a subset of OPF composites for all time points shown in radar diagrams per time point to analyze the rank order and extent of release of the different composites in the various release environments.

Discussion

The BMP-2 release profiles and IVIVC of various complex composites were investigated in several in vitro release environments. The latter affected the BMP-2 release profiles, creating distinct different relationships between release versus time and differences in extend of release. IVIVC resulted in level A internal predictability for individual composites. However, since the IVIVC was influenced by the BMP-2 loading method and composite surface chemistry, the external predictive value of in vivo release based on the in vitro and in vivo relationship was limited, stressing thereby the importance of including these cofounders in future IVIVC models.

The release environments influenced the BMP-2 release profiles differently. A large variation was seen in release profiles between the different composites in vivo and in the cell culture, whereas limited differences were seen in the cell-free in vitro environments. Although variation in cumulative release was observed in the cell culture model, the rank orders of the composites were different compared to in vivo. Despite the limited differences, the rank order also differed between PBS or SDB and in vivo. Furthermore, the in vivo relationship between release and time was not accurately predicted by the in vitro models. These results indicate that these in vitro models are not able to mimic the complex in vivo drug release conditions. Certain in vivo environmental differences could have explained differences in polymer erosion, drug diffusion, ion exchange, and desorption with subsequent differences in release profiles. For example, in PLGA foams, in vivo degradation was accelerated by an autocatalytic effect of the degradation products of the polymer.²² Furthermore, resident cells, proteins and ions could have influenced the polymer erosion, ion exchange, and desorption in vivo.^23–26 The superior in vivo reflection of the cell culture model against the cell-free in vitro models highlights the profound effect of biological influences on release profiles.

The various release phases affected the IVIVCs from the different composites. Since the in vivo burst release of the composites was >25% and no data points were available within this period, a large interval of the IVIVCs had to be extrapolated questioning the validity of the findings regarding this phase of release. Therefore, the burst and sustained release phase of the different composites was also analyzed separately. IVIVCs beyond the burst release phase showed models with better fit and internal predictability compared to all phases combined. In line with this, high burst releases are associated with lower correlating IVIVCs.²⁷ Rapid burst release phase from microencapsulated drugs may be related to the onset of bulk erosion of the polymer, providing additional pores for diffusion of the entrapped protein.²⁸ Since the rate limiting step of drug availability is drug permeability across the tissue barriers (a nonlinear kinetic process), IVIVC models fail to accurately predict the in vivo drug performance under burst release conditions. Notably, in the cell culture in vitro environment, a strong linear model (R² = 0.68) with good internal predictability (residuals within ∼10% BMP-2) was observed for the burst releases of the different composites. Since burst release is associated with a clinical relevant biological response and is often grossly underestimated in vitro,¹⁴ predicting the in vivo burst release using an in vitro model could offer a highly valuable tool.

Composite modifications influenced the external predictability of the IVIVC. To achieve a robust level A IVIVC, external predictability of multiple composite formulations with different release rates are recommended.¹² The external predictability of the achieved IVIVCs is complicated by confounding effects of the different composite formulations. To address this issue, we studied a set of biomaterials with tunable BMP-2 loading method and surface chemistry resulting into differential BMP-2 release profile. Indeed, both the BMP-2 loading method and the composite surface chemistry influenced the relationship, slope, and intercepts of the IVIVC regression models. Therefore, the models were not superimposable and could not accurately predict the in vivo release profiles of other formulations. Other studies did achieve a level A IVIVC with multiple formulations.^29–32 However, these studies used relatively simple delivery vehicles of PLGA microspheres with varying molecular weight and/or lactic to glycolic ratio. These microspheres have comparable chemistry and release is directed by similar erosion and diffusion processes, while complex composites used in this study have different chemistry and release is directed by different mechanisms including polymer erosion, diffusion, desorption, and ion exchange. The differences between these release mechanisms in vitro and in vivo could have influenced the IVIVC of the different composites. Altogether these results indicate that for complex delivery vehicles, a more sophisticated in vivo prediction model is needed. Various confounding factors should be analyzed and implemented as variables for successful predictive modeling of in vivo BMP-2 release and bone formation. Therefore, standards need to be set within the field, including the preferable in vitro environment to study BMP-2 release.

There is an unmet need for tuned in vitro release environments for proper prediction of IVIVC. The IVIVCs of different composites were investigated in commonly used in vitro systems (PBS and cell culture setup) and an SDB. Thus far, these models have been employed and grossly underestimate the in vivo release and did not explore the IVIVC.^3,9,11,33,34 In all buffers of this study, IVIVCs with good internal predictive value could be developed, in line with other observations. However, in line with previous observations,^29,35,36 all models lacked external predictive value due to the confounding effect of the composite formulations. This implies that advanced systems are needed to accurately discriminate between various release profiles, predict burst release, and imitate the complex in vivo release mechanisms. Thus far, there are a few limited alternatives. Level A IVIVC based on multiple delivery vehicle formulations has been reported using membrane dialysis in vitro systems.^30,31 A flow through dissolution apparatus (USP 4), incorporating Risperdal microspheres or dexamethasone microspheres in laminar flow cells with PBS 0.1% (w/v) sodium azide as circulation medium, demonstrated improved discrimination between different release profiles, better prediction of the in vivo release profiles, and better prediction of the in vivo burst release.^32,37 These studies indicate that creating more in vivo reflective in vitro models could improve the predictive value of in vitro release kinetics.

Conclusion

The large differences between in vitro and in vivo release force us to reconsider the in vitro BMP-2 release models used in bone tissue engineering. The cell-free in vitro buffers (PBS and SDB) used in this study represented gross under- or overestimation of the in vivo release, respectively, and were not able to discriminate between different in vivo release profiles. Although the cell culture model performed better in discriminating the different in vivo release profiles the cumulative release was ranked differently compared with the in vivo ranking. All in vitro buffers showed IVIVCs with good internal predictive value. However, due to the confounding effect of composite formulations, all models lacked external predictability. To develop a predictive in vitro model for in vivo release from complex delivery vehicles, in vitro models should imitate the in vivo environment. Potential confounding effects of drug type, delivery vehicle formulations, and application site could be analyzed to develop one single IVIVC for release from complex delivery vehicles.

Acknowledgments

We thank James L. Herrick and Carl T. Gustafson of Mayo Clinic for technical support. The authors wish to acknowledge the National Institutes of Health (R01 AR45871 and R01 EB03060), AO Foundation (AO startup grant S-15-46K), Dutch Arthritis Foundation (LLP22), and Anna-NOREF foundation for their financial support.

Disclosure Statement

No competing financial interests exist.