Journal of Controlled Release 229 (2016) 106–119

Contents lists available at ScienceDirect

Journal of Controlled Release

j ourna l homepage: www.e lsev ie r .com/ locate / jconre l

A novel in situ hydrophobic ion pairing (HIP) formulation strategy forclinical product selection of a nanoparticle drug delivery system

Young Ho Song a, Eyoung Shin a, Hong Wang a, Jim Nolan a, Susan Low a, Donald Parsons a, Stephen Zale a,Susan Ashton b, Marianne Ashford c, Mir Ali a, Daniel Thrasher a, Nicholas Boylan a, Greg Troiano a,⁎a BIND Therapeutics, 325 Vassar Street, Cambridge, MA 02139, USAb Oncology iMED, AstraZeneca, Macclesfield, Cheshire SK10 4TG, UKc Pharmaceutical Science, AstraZeneca, Macclesfield, Cheshire SK10 2NA, UK

⁎ Corresponding author.E-mail address: GregTroiano@bindtherapeutics.com (G

http://dx.doi.org/10.1016/j.jconrel.2016.03.0260168-3659/© 2016 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 3 December 2015Received in revised form 23 February 2016Accepted 16 March 2016Available online 18 March 2016

The present studies were aimed at formulating AZD2811-loaded polylactic acid–polyethylene glycol (PLA–PEG)nanoparticles with adjustable release rates without altering the chemical structures of the polymer or activepharmaceutical ingredient (API). This was accomplished through the use of a hydrophobic ion pairing approach.A series of AZD2811-containing nanoparticles with a variety of hydrophobic counterions including oleic acid, 1-hydroxy-2-naphthoic acid, cholic acid, deoxycholic acid, dioctylsulfosuccinic acid, and pamoic acid is described.The hydrophobicity of AZD2811 was increased through formation of ion pairs with these hydrophobic counter-ions, producing nanoparticles with exceptionally high drug loading—up to five fold higher encapsulation efficien-cy and drug loading compared to nanoparticles made without hydrophobic ion pairs. Furthermore, the rate atwhich the drug was released from the nanoparticles could be controlled by employing counterions with varioushydrophobicities and structures, resulting in release half-lives ranging from about 2 to 120 h using the samepoly-mer, nanoparticle size, and nanoemulsion process. Process recipe variables affecting drug load and release ratewere identified, including pH and molarity of quench buffer.Ion pair formation between AZD2811 and pamoic acid as a model counterion was investigated using solubilityenhancement as well as nuclear magnetic resonance spectroscopy to demonstrate solution-state interactions.Further evidence for an ion pairing mechanism of controlled release was provided through the measurementof API and counterion release profiles using high-performance liquid chromatography, which had stoichiometricrelationships. Finally, Raman spectra of an AZD2811-pamoate salt compared well with those of the formulatednanoparticles, while single components (AZD2811, pamoic acid) alone did not.A library of AZD2811 batches was created for analytical and preclinical characterization. Dramatically improvedpreclinical efficacy and tolerability data were generated for the pamoic acid lead formulation, which has been se-lected for evaluation in a Phase 1 clinical trial (ClinicalTrials.gov IdentifierNCT 02579226). Thiswork clearly dem-onstrates the importance of assessing awide range of drug release rates during formulation screening as a criticalstep for newdrug product development, and howutilizing hydrophobic ion pairing enabled this promising nano-particle formulation to proceed into clinical development.

© 2016 Elsevier B.V. All rights reserved.

Keywords:In situ hydrophobic ion pairingPLA–PEGNanoparticlesDrug release kineticsCounterionsAdjustable release rate

1. Introduction

Targeted nanoparticle drug delivery has the potential to open thetherapeutic window of pharmaceuticals [1–6]. Appropriate physicaland chemical engineering of the nanoparticles can enable optimal dis-tribution after systemic administration; that is, to achieve high local ac-tive pharmaceutical ingredient (API) concentrations in areas of disease,with limited distribution to healthy tissue and cells. The combination ofaltered biodistribution and controlled release of the drug payload canoverride the inherent disposition properties of the drug molecule and

. Troiano).

allow the particle characteristics to predominate, resulting in a drugconcentration–time profile at the site of action commensurate withthe drugmechanismof action. This profile is a key determinant of safetyand efficacy [7–9] and is dependent on the kinetics of drug release fromthe particle. As the ideal concentration–time profile will be different forevery drug, there is a critical need for delivery platforms to have adjust-able release kinetics.

For any matrix device used in controlled release, there are severalfactors governing drug release kinetics. These include drug solubilityand dissolution kinetics, desorption of surface bound or adsorbeddrug, drug diffusion through the matrix, matrix degradation and ero-sion, and a combination of such processes along with prodrugs[10–12]. For biodegradable nanoparticles, two of the most commonly

107Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

used approaches to fine-tune the drug release from the nanoparticlesare modification of matrix characteristics such as polymer composition[13,14] and encapsulation of prodrugs with various dissolution proper-ties [15–17]. The former approach suffers in that the other quality attri-butes of the particle can be impacted by thematrix change. For example,using a longer molecular weight polylactic acid block in a polylacticacid–polyethylene glycol (PLA–PEG) copolymer system will result inlower PEG coverage of nanoparticles, potentially enhancing particleclearance and reducing particle residence times in the plasma compart-ment. The prodrug approach results in the formation of a new chemicalentity, which adds chemical and regulatory complexity to the systemthat may result in suboptimal performance and a more arduous pathto demonstrating clinical utility [18].

To overcome these challenges and create a nanoparticle deliverysystem with adjustable release rates and high encapsulation efficien-cies, we have developed a technique called “in situ hydrophobic ionpairing.” By employing counterions of varying hydrophobicity andother properties, the nanoparticle release kinetics can be variedwithoutchanging the chemical structure of the polymer or API [19]. Additional-ly, the increased organic phase solubility of the hydrophobic ion paircompared to the API itself can result in increased drug loading and en-able administration of a smaller dose of particles in the clinic.

Here we describe the development and pharmaceutical characteri-zation of nanoparticle ACCURINS® of the aurora B kinase inhibitor,AZD2811. ACCURINS® are nanoparticles composed of block copolymersof poly-DL-lactide (PLA) and polyethylene glycol (PEG) in which thedrug payload is physically encapsulated. These particles overcome con-ventional drug delivery and distribution challenges for a variety of pay-loads with a wide range of targets [20]. ACCURINS® can be engineeredto distribute only to the vascular compartment initially, accumulate tohigh concentrations in diseased tissue, and bind to malignant cellswhen the particle surface is decorated with the appropriate ligand.AZD1152, a prodrug of AZD2811,was shown to be active in clinical trialsin a range of tumors, including hematological malignancies such asacute myeloid leukemia, but limited in its utility due to bone marrowtoxicity and a requirement for administration by 7-day continuous in-travenous infusion. Using the in situ hydrophobic ion pairing (HIP) ap-proach, we sought to optimize the AZD2811 drug loading and releasekinetics in ACCURINS® without varying other particle properties suchas particle size and polymeric components. The Quality Target ProductProfile defined during formulation development included sufficientlyhigh drug loading to enable practical clinical dosing and a near zero-order release profile lasting several days to mimic the continuous IV in-fusion employed in the prodrug clinical study, though the ideal releaseprofile was to be determined preclinically. A panel of ion pairing agentswas used to generate nanoparticle formulations spanning a range of re-lease rates, and the formulations were characterized with respect to bi-ological performance. Based on data from these results a clinicalcandidate has been identified and moved into clinical trials.

Hydrophobic ion pairing has been used in various applications as analternative approach to alter the solubility of biomolecules [21] or to in-crease the encapsulation efficiency of hydrophilic APIs into drug deliv-ery systems [22,23] by increasing log P and organic solubility.However, to the best of our knowledge, this is the first work thatshows how in situ HIP could be used to alter the release rate of theAPI, presumably through an increase in molar volume of the ion pairin comparisonwith theAPI alone and a consequent decrease in diffusionrate.We believe it allows HIP to become a novel formulation strategy toprovide a wide range of release profiles without impacting chemical at-tributes of the API or physicochemical attributes of the delivery system.This is an important distinction compared to ion pairing to create saltforms that are more easily entrapped into delivery systems.

Ion pairs are formed by electrostatic interactions, and the complexwill easily dissociate in the presence of an excess of oppositely chargedions [24]. Fig. 1 is a schematic representation of the HIP formation pro-cess, HIP encapsulated in nanoparticles, and HIP release from

nanoparticles. The HIP complex is generally lipophilic and partitionsinto the hydrophobic core of the polymer matrix during an encapsula-tion process, thus enhancing encapsulation efficiency [25]. This ap-proach has been employed for the delivery of various peptides andproteins, all relatively hydrophilic agents where the primary purposeof the ion pairing was to increase log P and enable encapsulation[26–29]. To prepare HIPs with both the best encapsulation efficiencyand release properties possible, it is necessary to screen ion pairingagents, whose rational selection depends on physicochemical proper-ties of the API being encapsulated.

Because the method can be applied to a variety of ionizable mole-cules, incorporating HIP into a nanoparticle delivery platform such asACCURINS® presents a broadly applicable formulation strategy forpharmaceuticals considering the overall high proportion of ionizablecompounds among druglike substances [30] and the existence of abroad range of pharmaceutically acceptable counterions with varyingphysicochemical properties. Thus, our nanoparticle platform has under-gone a remarkable increase in the number and types of APIs that can po-tentially be encapsulated as well as providing release rates that may bemodulated by a novel formulation strategy, enabling the selection of theoptimal ion pairing agent for clinical development.

2. Materials and methods

2.1. Drugs and chemicals

AZD2811 (formerly designated AZD1152 hydroxy-quinazolinepyrazole anilide (AZD1152hQPA, 2-[3-[[7-[3-[ethyl(2-hydroxyethyl)amino]propoxy]quinazolin-4-yl]amino]-1H-pyrazol-5-yl]-N-(3-fluorophenyl)acetamide) as an anhydrous free base andAZD1152 (barasertib, prodrug of AZD2811) were provided byAstraZeneca Pharmaceuticals (Macclesfield, UK). PLA–PEGwas receivedfrom Evonik (part #100 DL mPEG 5000 3.5CE), which has a number-average molecular weight of approximately 16 kDa for PLA and approx-imately 5 kDa for PEG (determined via nuclear magnetic resonance,NMR), with a polydispersity index of approximately 1.25 (determinedvia gel permeation chromatography). Ethyl acetate (EA), benzyl alcohol(BA), trifluoroacetic acid (TFA), and dimethylsulfoxide (DMSO) werefrom EMD Millipore (Billerica, MA, USA). For NMR studies, benzylalcohol-d8 and DMSO-d6 were purchased from Cambridge Isotope Lab-oratories (Tewksbury, MA, USA). Oleic acid, 1-hydroxy-2-naphthoicacid, cholic acid, deoxycholic acid, dioctyl sulfosuccinate sodium salt(docusate sodium salt), pamoic acid, citric acid, acetic acid, boric acid,phosphoric acid, sodium dihydrogen phosphate, di-sodium hydrogenphosphate dihydrate, and phosphate buffered saline (PBS) were pur-chased from Sigma-Aldrich Ltd. (St. Louis, MO, USA). All reagents wereanalytical or high-performance liquid chromatography (HPLC) gradeand were used without further purification.

2.2. API and counterion characterization

The acid dissociation constants (pKa) of AZD2811 and counterionsweremeasured via potentiometric titration using a Sirius T3 instrument(Sirius Analytical Instruments Ltd., Forest Row, E. Sussex, UK). Theoctanol–water partition coefficients (log P) for AZD2811were alsomea-sured by performingpotentiometric titrations in the presence of varyingratios of octanol and water. The pKa and log P data for AZD2811 wereused to generate a lipophilicity profile (plot of distribution coefficientslog D versus pH). In addition, log D was measured using a conventionalshake-flask measurement. Briefly, an organic phase comprising a mix-ture of EA and BA at a 79:21 weight ratio and an aqueous phase of10 mM citrate and 10 mM phosphate were pre-equilibrated overnightat room temperature with gentle mixing. The EA/BA solvent mixturewas used for the organic phase because it is more representative ofthe organic phase used during nanoparticle preparation. A stock solu-tion of API in DMSO was prepared, diluted into the two-phase system,

Fig. 1. A schematic representation of the HIP formation, HIP encapsulated in nanoparticles, and HIP release from nanoparticles. (A) AZD2811 drug only baseline, (B) AZD2811–oleic acid,and (C) AZD2811–pamoic acid.

108 Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

andmixed overnight at room temperature. The two phases were subse-quently separated, and an equal volume of each phase was transferredto a centrifuge tube and the API concentration in both phaseswas deter-mined by ultra-performance liquid chromatography–UV (UPLC-UV)analysis.

Solubility of AZD2811 was determined in water, various aqueousmedia, several neat organic solvents, and solvents containing organicacids by visual inspection followedbypreparation of saturated solutionsand 0.2-μm PTFE disc syringe filtration prior to analysis by HPLC. Someof the organic acids screened in BAwere TFA, deoxycholic acid, octanoicacid, oleic acid, 1-hydroxy-2-naphthoic acid, dodecylbenzensulfonicacid, and pamoic acid.

The chemical stability of AZD2811 was assessed in acid, base, andperoxide-containing solvent systems with and without the organiccounterions using a stability-indicating UPLCmethod. AZD2811was de-termined by UPLC analysis using a C18 reverse phase column (WatersCSH C18) and a mobile phase gradient of 15% to 85% acetonitrile with0.08% TFA. The eluent absorbancewasmonitored at 238 nm. Counterion

Table 1Summary of method conditions for UPLC and CAD analyses.

Compound Column Columntemp(°C)

Flow rate(mL/min)

Detectionmode

Mp

AZD2811 Waters CSHC18

30 0.3 UV, 238nm

0T

Cholic and deoxycholic acids WatersX-Bridge C8

30 0.5 CAD 0T

Oleic acid Waters BEH C8 80 0.3 CAD 0T

Dioctylsulfosuccinic acid WatersX-Bridge C8

60 0.5 CAD 0T

Pamoic acid and1-hydroxy-2-naphthoic acid

Waters HSS T3 30 0.3 UV, 238nm

0T

ACN = acetonitrile.

quantitation was assessed using UPLC conditions as described inTable 1; reverse phase chromatography with acetonitrile/water gradi-ents were utilized to separate the organic acids from API and other ex-cipients. For counterions without UV chromophores, corona arraydischarge (CAD) detection was utilized for quantitation.

2.3. 13C NMR spectroscopy

Carbon spectra were acquired using a 100-MHz Bruker instrument,80° to 90° pulse, 5-second delay, and 4096 scans. All spectra were ac-quired with the proton decoupler active both during pulse durationand acquisition (delay) to enhance signal-to-noise ratio.

2.4. Preparation of AZD2811-encapsulated nanoparticles

Nanoparticles were created via a nanoemulsion process using themodified oil in water (o/w) emulsification solvent extraction methodas described by Hrkach et al. [20]. Briefly, an organic phase composed

obilehase A

Mobilephase B

Gradient program

.1%FA/water

0.08%TFA/ACN

15% to 20% B over 4 min; 20% to 50% B over 1 min; 50% to85% B over 1 min

.02%FA/water

0.02%TFA/ACN

40% to 65% B over 5 min

.02%FA/water

0.02%TFA/ACN

40% to 65% B over 7.5 min

.02%FA/water

0.02%TFA/ACN

50% to 89% B over 8 min

.1%FA/water

0.08%TFA/ACN

50% to 80% B over 3 min

109Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

of AZD2811, appropriate counterions, and 16–5 PLA–PEG in a EA/BA sol-vent mixture was rapidly mixed and dispersed with an immiscibleaqueous phase containing the optimal concentration of surfactant forparticle size control. First, a coarse emulsion is formed by a handheldrotor/stator homogenizer. This coarse emulsionwas subsequently proc-essed by a high-pressure microfluidizer (Microfluidics, Inc., Westwood,MA, USA) to form a fine nanoemulsion. The emulsions were quenchedby addition into cold water or buffer solutions, resulting in extractionof solvent from the organic phase and hardening of nanoparticles. Sub-sequent processing included tangential flow filtration, which removesunencapsulated API and processing aids, and addition of sucrose for sta-bilization and cryopreservation of the nanoparticle suspension. Thenanoemulsion process of ACCURINS® is shown schematically in Supple-mentary Fig. 1.

As cholic acid has relatively low solubility in the organic phase (EA/BA), the cholic acid formulations were created by dissolving sodiumcholate in the aqueous phase while using TFA to solubilize the API inthe organic phase. Dioctylsulfosuccinic acid was prepared by a two-phase immiscible liquid extraction method using dioctyl sodiumsulfosuccinate. Briefly, sodium dioctylsulfosuccinate was dissolved inBA, and then concentrated hydrochloric acid solution and water wereadded to acidify the solution. The mixture was vortexed and afterphase separation, BA containing protonated dioctylsulfosuccininc acidon the top layer was aspirated and used as it is. Due to the low solubilityof pamoic acid in the organic phase, pamoic acid formulationswere cre-ated using DMSO as a cosolvent in the organic phase and TFA as a drugsolubilizer.

2.5. Variation of quench buffer pH

The pH of the quench phase was varied from pH 3 to 11 using citricacid–phosphate buffer solutions (0.1 M citric acid and 0.2 M Na2HPO4),a phosphate buffer solution (0.2 M Na2HPO4, 0.2 M NaH2PO4), and aBritton-Robinson buffer solution (0.1 M acetic acid, 0.1 M boric acid,and 0.1 M phosphoric acid) during the preparation of deoxycholicacid, dioctylsulfosuccinic acid, and pamoic acid with AZD2811 in nano-particles to study the effect of quenching the emulsion in different pHvalues.

2.6. Variation of quench buffer molarity

The effect of quench buffer molarity was also evaluated using phos-phate buffers of 1, 10, 100, 200, and 500 mM (pH 6.5) to determine thebuffering capacity required aswell as impacts of quench buffer molarityon formulation properties.

2.7. Nanoparticle characterization

AZD2811 nanoparticle formulations were characterized with re-spect to particle size, drug load, and in vitro release (IVR) kinetics de-scribed in Hrkach et al. [20] and briefly here.

2.7.1. Particle size and zeta potential analysisParticle size was determined by dynamic light scattering (DLS) of a

dilute aqueous suspension at 25 °C on a Brookhaven ZetaPALS instru-ment with a 660-nm laser scattered at 90°. DLS data were analyzedusing the cumulants method. Three measurements are performed oneach sample and a mean and standard deviation are reported as wellas the polydispersity index (PDI). Zeta potential is also measured onthis instrument for select samples, but because all ACCURINS® createdby this PLA–PEG copolymer demonstrate a near neutral surface (0 to−10mV, data not shown), this data does not play any role in the selec-tion or performance of the formulations.

2.7.2. AZD2811 load and encapsulation efficiency (EE)The AZD2811 load was calculated from the ratio of AZD2811 to the

gravimetric dry weight of the sample prior to adding sucrose to the sus-pension.

Drug Loading %ð Þ ¼ Drug content in nanoparticlesTotal nanoparticle weight� 100

Drug Encapsulation Efficiency %ð Þ ¼ Drug loadingTheoretical drug load� 100

2.7.3. AZD2811 IVR kineticsAZD2811 release kinetics were determined in vitro under physiolog-

ical sink conditions. Nanoparticles were suspended in 10% polysorbate20 in 0.01-M phosphate buffered saline (pH 7.4) and incubated withmild agitation in a 37 °C water bath. Periodically, an aliquot of the sus-pension was removed and ultracentrifuged at 264,000g for 30 min.Samples of the supernatant and the suspension prior to ultracentrifuga-tion were analyzed by UPLC, and the percent release was calculated bycomparing the released AZD2811 concentration in the supernatantwith the total concentration in the uncentrifuged sample.

2.7.4. Counterion (pamoic acid) IVR kineticsPamoic acid release kinetics were also determined in vitro under

physiological sink conditions using the samples generated by theAZD2811 IVR test. Pamoic acid from the supernatant and the suspensionprior to ultracentrifugation were analyzed by HPLC, and the percent re-lease was calculated by comparing the released pamoic acid concentra-tion in the supernatant with the total pamoic concentration in theuncentrifuged sample.

2.7.5. Raman spectroscopyFourier transform (FT) Raman spectra were acquired on a Nicolet

model 6700 spectrometer interfaced to a Nexus Raman accessory mod-ule. The instrument is configured with a Nd:YAG laser operating at1024 nm, a CaF2 beam splitter, and a indium gallium arsenide detector.OMNIC 8.1 software was used for control of data acquisition and pro-cessing of the spectra. Samples were packed into 3-inch glass NMRtubes for analysis. Each spectrum consisted of 512 scans at 2 cm−1 res-olution. The OMNIC 8.2 software package (Thermo Scientific, Waltham,MA, USA) was used to acquire, process, and evaluate the spectral data.

2.8. Pharmacokinetics

The pharmacokinetics of AZD2811-encapsulated nanoparticles werecompared to that of AZD2811. Male Wistar Han rats (four/dose group)were administered at either 5mg/kg or 0.5mg/kg AZD2811 doses intra-venously through a lateral tail vein. Blood samples were collected fromindwelling jugular vein catheters at various time points through 72 hpost-dose. Blood samples were collected into lithium heparin tubesand processed to plasma. Total AZD2811 plasma concentrations werequantified using a supported liquid extraction method followed by liq-uid chromatography–mass spectroscopy (LC–MS/MS). The LC–MS/MSconditions are shown in a previous report [19].

2.9. Preclinical pharmacodynamic and efficacy studies

2.9.1. Tumor pharmacodynamic & bone marrow toxicity studiesMale nude rats bearing SW620 tumors were randomized, based on

tumor size, into vehicle or treatment groups with five rats per group.Animals were dosed intravenously with placebo nanoparticles orAZD1152 for 4 consecutive days (days 1–4) or AZD2811 nanoparticlesfor two doses 48 h apart (days 1 and 3). Tumors and femurs were ex-cised postmortem at specified time points after the first dose, andfixed in 10% buffered formalin for 24 to 48 h. Femurswere subsequently

Table 2Solubility of AZD2811 in aqueous media and organic solvent.

Aqueous mediaa AZD2811solubility(mg/mL)

Organic solventa AZD2811solubility(mg/mL)

Deionized water 0.013 BA 16.73PBS 0.017 EA 0.08310% polysorbate 20 in PBS(pH 7.4)

0.582 DMSO N142

Acid/BA5% TFA/BA 225 12% oleic/BA 22910% deoxycholic/BA 148 9%

dodecylbenzenesulfonic/BA150

110 Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

decalcified in 10% formic acid, and then both tumor and femur sampleswere processed to paraffin blocks. [19]. Formalin fixed paraffin blockswere subsequently sectioned (4 μm), de-paraffinized and endogenousperoxidase blocked with 3% hydrogen peroxide. For pHH3, serum-freeprotein block (Dako) was applied prior to incubation with primary an-tibody (Upstate Biotechnology 1/1000 dilution) then samples devel-oped in liquid 3,3-diaminobenzidine (DAB; Dako) and counterstainedwith Carazzi's hematoxylin. Tumor and femur were scored visually bya pathologist for levels of pHH3 or a reduction in bonemarrow cellular-ity using a scoring systemwhere 0=no change, 1=minimal, 2=mild,3 = moderate, 4 = severe change compared to placebo nanoparticletreated animals.

3% octanoic/BA 108 9% pamoic/BA 2643% decanoic/BA 100 3% 1-hydroxy-2-naphthoic

acid/BA102

a % = % (w/w).

2.9.2. Tumor xenograft studies

Fragments of the small cell lung cancer (SCLC) patient-derived ex-plant SC61 were transplanted in athymic mice 6 to 9 weeks of age.Tumor growth was monitored by calipers and animals randomizedinto groups based on tumor size. Mice were then dosed intravenouslywith placebo nanoparticles or AZD1152 for 4 consecutive days (days1–4) or AZD2811 nanoparticles for two doses 48 h apart (days 1 and3). Tumor growth wasmonitored at least twice weekly for the durationof the study, whichwas 28 days. Statistical analysis of tumor growth in-hibition studies uses a comparison of LogRTV for each group of animals(where RTV is the geometricmean calculated usingfinal tumor volume/initial tumor volume for individual animals), with a one-sided Student'st-test, pooled variability across all groups and uses a 5% significancelevel.

3. Results

3.1. API and counterion characterization

The chemical structure of AZD2811 is depicted in Fig. 2(B) and thetwo basic amine groups are highlighted in blue. Using the Sirius T3 in-strument, the basic pKa values were determined to be 4.5 and 8.7 asmeasured via potentiometric titration using methanol as a cosolvent.However, as a first step before experimental measurements, a predic-tion approach was performed, using the ACD/pKa program (AdvancedChemistry Development, Inc., Toronto, ON, CAN) to estimate the pKaof the compound. This approach resulted in two alkaline pKa values of5.0 and 8.8, attributed to the two nitrogen molecules. There was goodsimilarity between observed and calculated pKa values.

Fig. 2(C) shows the lipophilicity profile for AZD2811. At pH ≥ 8.7,AZD2811 is neutral with a measured log Poctanol of 3.3. At intermediatepHs (4.5 b pH b 8.7), the stronger basic amine is ionized, thereby de-creasing the lipophilicity (log DpH 7.5 = 2.2). Under acidic conditions(pH b 4.5), the lipophilicity decreases further due to ionization of thesecond basic amine group. The log Dmeasurement from a conventional“shake-flask”method using (BA:EA)/buffer shows good similarity to li-pophilicity generated from the SiriusT3 instrument.

Fig. 2. (A) Chemical structure of AZD1152, prodrug of AZD2811 (B) Chemical structure of A

The solubility of AZD2811 in various media is shown in Table 2.AZD2811 was chemically stable in most of the solvent systems both inthe presence and absence of organic acids (data not shown).

AZD2811 solubility in the organic phase was dramatically increasedby the addition of organic acids, caused by the charge shielding resultingfrom ion pairing and the increase in lipophilicity.

Fig. 3 shows the molar concentration of AZD2811 dissolved as afunction of the molar concentration of each acid present in BA solution.It is clear that the molar solubility increase realized is commensuratewith the acid molarity, following a stoichiometric ratio of 1:1 with theexception of the diacidic pamoic acid, which follows a 1:2 ratio.

3.2. 13C NMR spectroscopy

Protonation of the amino functionality of AZD2811 by organic acidssuch as pamoic acid and 1-hydroxy-2-naphthoic acidwas studied by so-lution NMR spectroscopy in a 1:1 (v/v) mixture of BA and DMSO. Con-trol spectra of base (amino-functional AZD2811, Fig. 4(A) and (B))and acids (Fig. 4(C) and (D)) were acquired with 20-mg/mL solutions.Salt spectra (solutions with both the drug and acid present, Fig.5(A) and (B)) were acquired with 1.0- and 0.5-M equivalents of 1-hydroxy-2-naphthoic acid (mono-acid) and pamoic acid (di-acid), re-spectively, present in solution. AZD2811 drug (Fig. 4(A) and (B)),pamoic acid (Fig. 4(C)), and 1-hydroxy-2-naphthoic acid (Fig. 4(D))spectra were assigned considering proximity to heteroatoms, ring cur-rent effects from aromaticity, splitting or peak broadening due to cou-pling to nitrogen-14 (when possible), and guidance from NMR Predictsoftware (Mestralab Research, Santiago de Compostela, Spain). Reso-nances from carbons alpha, beta, and gamma to tertiary aliphatic nitro-gen (atom #4, pKa ~ 9) of AZD2811 (atom numbers 2, 3, 5, 6, 7, 36, 37)are observable in the 10- to 70-ppm region (Fig. 4(A)) of the spectrum.As shown in Fig. 5(A), these resonances shift upfield in the spectrum ofAZD2811–pamoate salt due to increased electron density resulting fromproximity to carboxylate anion (or solvent counterions solvating the

ZD2811 and (C) AZD2811 lipophilicity profile using octanol/water and (BA:EA)/buffer.

Fig. 3. Increase inmolar concentration of AZD2811 in BAwith the increase inmolar concentration of hydrophobic organic acids in BA. Note the grouping of data usingmonoacids along the1:1 M ratio line, while the diacidic pamoic acid data points align well with the 1:2 M ratio line.

Fig. 4. 13C NMR spectra of free AZD2811 (A) 2.5- to 75-ppm region and (B) 95- to 175-ppm region, (C) pamoic acid, and (D) 1-hydroxy-2-naphthoic acid with assignment of resonancesobserved.

111Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

Fig. 5. (A) The resonances shift upfield due in the spectrumof AZD2811–pamoate salt compared to free AZD2811, and (B) similar upfield shift observed in these resonances fromAZD2811-1-hydroxy-2-naphthoic acid salt spectrum. Absence of peaks representing amine-free base suggests a high degree of amine protonation in both salt samples, indicating interaction ofAZD2811 with those counterions in solution state.

112 Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

ammonium salt). A similar upfield shift is observable in these reso-nances in the AZD2811 with 1-hydroxy-2-naphthoic acid salt spectrum,shown in Fig. 5(B). Notably, in both salt spectra, residual resonances as-signable to amine-free base are not observed. While these spectra arenot strictly quantitative (due to relative short delay time and NuclearOverhauser Effect resulting from use of continuous proton decoupling),the absence of peaks representing amine-free base suggests a high de-gree of amine protonation in both salt samples.

3.3. Preparation and characterization of AZD2811-encapsulatednanoparticles

Hydrophobic counterions having various acidity, molecular weight,shape/bulkiness, and lipophilicity were characterized for pKa, Log P,and Log D7.5 and incorporated into AZD2811 nanoparticles for charac-terization of release rate. The physicochemical properties of the repre-sentative counterions measured by SiriusT3 are summarized inSupplementary Table 1.

Table 3Hydrophobic ion pairing formulation of AZD2811 as lead formulations selected for in vivo stud

Counterions used AZDload(%)

Encapsulationefficiency (%)

Nanoparticlediameter (nm)±STDa

Nanoppolydindex± STD

Baseline 3.2 15 129 ± 1.22 0.170Oleic acid 9.7 30 113 ± 0.32 0.1091-Hydroxy-2-naphthoic acidb 4.0 16 101 ± 1.59 0.1521-Hydroxy-2-naphthoic acid/cholicacidc

4.7 18 118 ± 0.92 0.198

Cholic acid 7 23 109 ± 0.89 0.166Deoxycholic acid 9.5 37 107 ± 0.95 0.124Dioctylsulfosuccinic acid 12 45 101 ± 0.12 0.078Pamoic acid 17 71 88 ± 0.06 0.096

a STD is the standard deviation based on three independent measurements of the same samb Brij 100 (non-ionic surfactant) was used as a surfactant to control the size, and thus countc Sodium cholate was used as a surfactant to control the size and counterion detected in nand As cholic acid has relatively low solubility in the organic phase (EA/BA), the cholic acid for

while using TFA as a drug solubilizer in BA.

For the initial counterion screening, a series of AZD2811-counterionnanoparticles were prepared using a design of experiments (DOE) ap-proach. This allowed us to determine if organic phase compositionplayed a role along with counterion selection in influencing the loadand release kinetics of the AZD2811 nanoparticles. The overall conclu-sion from the DOE (data not shown) was that the counterion is themost significant factor influencing loading and release. A standardleast-squared regression model was fit for drug loading and encapsula-tion efficiency. The counterion usedwas the only statistically significantinput variable (p b 0.0001) for both drug encapsulation and all releaserate attributes. Drug input (i.e., theoretical drug loading, p = 0.001)and organic phase solids concentration (p=0.01) were also significantfactors forfinal drug loading levels. The numerical prediction expressionfor drug loading, ignoring interaction factors, is: x + 0.149 ∗ theoreticalloading + 0.205 ∗ organic phase solids + 0.00665 ∗ counterionconcentration + 0.135 ∗ organic phase EA + 0.000417 ∗ size − 0.138,where x is a constant defined by the counterion used. Based on thesescreening experiments, the results of optimized HIP formulations of

ies.

articleispersity

a

Initialin vitrorelease(%)

Time to50%release(h)

Input [acid:drug]molar ratio infeedsolution

Output [acid:drug]molarratio in nanoparticlesmeasured

± 0.01 19 3.3 n.a. n.a.± 0.02 3.8 14 1 1± 0.02 1.2 15 2 1± 0.01 2.8 35 2 1.8c

± 0.05 3.5 66 2.1d 1± 0.01 1.8 48 1 1.3± 0.02 1.7 N72 1 1± 0.02 3.5 120 0.8 0.6

ple.erion detected in nanoparticles included only 1-hydroxy-2-naphthoic acid.oparticles included 1-hydroxy-2-naphthoic acid as well as cholic acid.mulations were created by dissolving sodium cholate in the aqueous phase as counterion

Fig. 6. (A) Cumulative AZD2811 release profiles from HIP formulations. (B) AZD2811 release rate (%/day) profiles from HIP formulations. Data points represent means and error barsrepresent standard deviations of the sample measurements.

113Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

AZD2811 selected as leads for in vivo studies are summarized in Table 3.The “baseline” formulations for AZD2811 are essentially polymer anddrug only with no additional excipients. Additionally we have foundthat all the quality attributes of the nanoparticles remain unchangedwhen stored in frozen (−20 °C) conditions. Stability data is availableup to one year at the writing of this manuscript but is beyond thescope of this paper.

The cumulative IVR rate profiles from six different HIP formulationsalong with baseline (no acidic counterions) as a comparison aredisplayed in Fig. 6(A). In vitro release kinetics of nanoparticles contain-ing organic acids varied considerably depending on the quantity andtype of acid employed, with progressively decreasing release rates ob-served with oleic acid, 1-hydroxy-2-naphthoic acid, cholic acid,dioctylsulfosuccinic acid, and pamoic acid, respectively. It is significantto mention here that an extension of the time to 50% release in vitrofrom less than 2 h to as long as 120 h was observed with very littleburst release (b 4%, desirable for sustained release drug delivery sys-tem) from all the HIP formulations compared to baseline nanoparticleswith 19% burst. Another manner in which to view the IVR data is as re-lease rate, shown in Fig. 6(B). In Fig. 6(A), it is apparent that release rateof formulations was widely varied, spanning from typical biphasic withan initial rapid release followed by a slower phase to more constant/steady release like dioctylsulfosuccinic and pamoic formulations. As

Fig. 7. Impact of buffered quench pH on drug loading of (A) AZD2811–deoxycholic acid, AZD2(B) AZD2811–pamoic acid nanoparticle formulations in contrast to the drug and counterion pKleast one amine protonation on AZD2811 and pH higher than the counterion pKa. Single samp

seen in Fig. 6(B), these two ion paired formulations showed well-controlled release rates.

Various factors affecting the drug load and release of nanoparticleswere studied, including quench pH and quench buffer molarity, to pro-vide leads in Table 3.

3.4. Key formulation variables tested: quench pH

Following emulsification, hardened nanoparticles are produced bysolvent extraction by diluting emulsion in a quench buffer. It was hy-pothesized that the pH of this quench buffer could influence the forma-tion of the encapsulated HIP and consequently, drug loading. Theloading of nanoparticle batches made with varying quench buffer pHemploying deoxycholic acid, dioctylsulfosuccininc acid, and pamoicacid as counterions is displayed in Fig. 7(A). As expected, the resultsshow quench pH is a significant factor for drug loading, and the drugload profile exhibits a pH-responsive nature for all three acids tested,supporting the hypothesis that drug load is maximized under pH condi-tions that favor the formation of an ion pair with the drug. Based on thisload versus pH profile data, an optimal pH was chosen for each system.

Fig. 7(B) shows drug loading versus quench pH for pamoic acid HIPformulations, with drug and counterion pKa values indicated. Improperquench pH resulted in low drug loading because of improper ionic

811–dioctylsulfosuccinic acid, and AZD2811–pamoic acid nanoparticle formulations anda values. In all cases, maximum drug loading occurs in a pH range that corresponds to atle measurements were made at each data point.

Fig. 8. Impact of quench pH on drug release of AZD2811–deoxycholic acid, AZD2811–dioctylsulfosuccinic acid, and AZD2811–pamoic acid nanoparticle formulations. Singlesample measurements were made at each data point.

114 Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

forms of either the counterion (neutral at low acidic pH) or AZD2811(neutral at high basic pH). Fig. 8 displays the impact of quench pH ondrug release from the nanoparticle. Faster drug release was observedas pH approached the AZD2811 pKa value of 8.7.

3.5. Key formulation variables tested: quench buffer molarity

The objective of this experiment was to determine the buffering ca-pacity required to ensure high drug loading, and to evaluate the effect ofquench buffer molarity on drug load and drug release by varying buffermolarity (ionic strength) at a fixed pH of 6.75. Fig. 9 shows the relation-ship between drug loading (A) and cumulative drug release at 24 h(B) versus quench buffer molarity.

The ionic strength of the quench buffer impacted drug loading dra-matically, with a trend of higher molarity of buffer yielding higherdrug loading up to 0.2 M, where a plateau was reached. In the case ofvery low ionic strength phosphate buffer (0.001 M), the low drug loadis attributed to low buffering capacity, as the very low molarity bufferscould not maintain the pH upon the influx of acid during emulsion ad-ditionwith a pH drop to 4.0 from an initial pH 6.75.When ionic strengthwas between 0.01 and 0.10M, buffer strengthwas adequate tomaintainpH between 6.50 and 6.75. However, the drug loading was lower thanthat of nanoparticles prepared using 0.2 M or greater quench buffer.This observation is consistent with the impact of ionic strength on thepartitioning behavior of the ionized species in the emulsion. The mostprofound impact, however, was the dependence of drug release on

Fig. 9. Impact of quench buffer ionic strength on (A) drug loading and post-quench pH and (B)that efficient encapsulation of the HIP cannot be achieved with very low buffering capacity andvery high buffer concentration. Single sample measurements were made at each data point.

ionic strength. While a mechanistic explanation of this observationhas not yet been demonstrated, it is speculated that small buffer ionsmay disrupt the ionic interaction of pamoate and drug. In addition, asshown by Lengsfeld et al. [24], during the dissolution of HIPs in an aque-ous electrolyte solution, reverse ion exchange can occur as small hydro-philic counterions are substituted for the hydrophobic organic ions,resulting in the re-formation of the parent drug, with a concomitant in-crease in aqueous drug solubility. Therefore, we hypothesize that thehigh ionic strength buffers provide a significant anion concentrationthat may be encapsulated, providing an unfavorable environment forHIP formation. Using elemental analysis (data not shown), we observeda trend of higher phosphate content in nanoparticles made using thehigher ionic strength quench buffers, supporting this theory. Ionicstrength used in the system would have a pronounced effect on theion pairing phenomenon. The significant impact of quench phase ionicstrength on drug load as well as drug release supports the ionic interac-tion of counterionswith API in the nanoparticle as a driver of these char-acteristics. Based on drug load and drug release versus ionic strength ofquenched buffer, an optimal ionic strength was chosen for each system.

Further insight into the role of the AZD2811–pamoic acid HIP inmoderating drug release rate may be gained from examination of thecumulative release curves of the pamoic acid and AZD2811. Nanoparti-cle formulations were prepared with each HIP component individually(PAM-only nanoparticle and AZD-only nanoparticle) and with the twocomponents together (AZD-PAM nanoparticle). Release kinetics ofeach component from these three formulations are plotted in Fig.10(A). Pamoic acid–only nanoparticles exhibited high burst of ~55%and complete release of pamoic acid at 1 h. Similarly, AZD-only nano-particles exhibited high burst of ~20% and N80% release of AZD2811 at24 h (Fig. 10(A), open symbols). When pamoic acid was encapsulatedalong with AZD2811, both HIP components exhibited significantlyprolonged release kinetics;moreover release rateswere nearly identicalon a cumulative percentage basis (Fig. 10(A), closed symbols). Themolar ratio of the released components is shown in Fig. 10(B); the ob-servation that the HIP components release from the particle in a con-trolled and stoichiometric fashion when formulated together, butrelease very quickly when encapsulated alone, provides further evi-dence of the ion pairing effect and of the association between acidicpamoic acid and basic AZD2811 during both encapsulation and diffu-sional release from the nanoparticle.

3.6. Spectral identification AZD2811-PAM supporting ion pair formation innanoparticles

Raman spectra were collected of AZD2811, pamoic acid, the 2:1AZD2811:pamoic acid salt, nanoparticles formulated with AZD2811

AZD2811 release of AZD2811–pamoic acid nanoparticle formulations. These data suggestthat the HIP association within the nanoparticle may be disrupted after encapsulation at

Fig. 10. (A) Release kinetics of pamoic acid from pamoate-only particles and AZD2811–pamoic acid–encapsulated nanoparticles, as well as of AZD 2811 fromAZD2811-only particles andAZD2811–pamoic acid–encapsulated nanoparticles. (B) Molar ratio of released pamoic acid and AZD2811. These data support co-diffusion and co-release of counterion and drug. Datapoints represent means and error bars represent standard deviations of the sample measurements.

115Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

and pamoic acid together, and nanoparticles formulated without eitherHIP component (placebo). As can be seen clearly in Fig. 11, Raman spec-tra of nanoparticles formulated with both AZD2811 and pamoic acidmatch most closely with the spectra of the AZD2811–pamoic acid salt,while spectral features of either free API or counterion are not evident.This suggests that an intermolecular interaction between the API andcounterion similar to that present in the solid salt is also present inthe nanoparticle-encapsulated materials.

3.7. Pharmacokinetics

The pharmacokinetics of the nanoparticle formulations in ratsdemonstrate considerably higher and more prolonged circulatingdrug levels compared to the same dose of AZD2811 solution (Fig.12). These data represent total plasma drug levels, and as such donot distinguish between encapsulated and released AZD2811.

Fig. 11. Raman spectra of single components (AZD2811, pamoic acid), AZD2811-pamoate2:1 salt, placebo nanoparticles, and nanoparticles containing AZD2811 and pamoic acid.The spectra of the nanoparticles containing AZD2811and pamoic acid (AZD-PAM NPs)are comparable to that of the preformed salt (AZD2811 pamoate 2:1 salt).

Enhanced bioanalytical techniques have been shown to separate“free” non-protein bound API from encapsulated API in plasma[31]. The large differentiation between the PK curves of parentAZD2811 and those of the nanoparticle formulations are consistentwith long circulating particles that maintain their controlled releaseattributes in vivo.

Baseline nanoparticles differ in pharmacokinetics including volumeof distribution (73.6 mL/kg), half-life (13.7 h) and clearance (4.3 mL/h/kg) compared to the other HIP nanoparticles (volume of distribu-tion = 46.7 mL/kg; half-life = 18.2 h and clearance = 1.9 mL/h/kg forthe cholic acid nanoparticle for example).

With the exception of the baseline formulation that was excludedbased on low drug loading and fast IVR, all initial lead formulationswere screened in biological assays including pharmacodynamics andtissue distribution studies [19]. Select formulationswith themost prom-ising screening results were progressed into bone marrow toxicity andtumor growth inhibition studies.

3.8. Bone marrow toxicity and tumor mechanism of action

As demonstrated previously [19], nanoparticle formulationsdemonstrating sustained release cause minimal bone marrow toxicity(Fig. 13). On day 5, compared to vehicle-treated animals, a markedhypocellularity was observed in bone marrow from AZD1152-treatedanimals. AZD2811–oleic nanoparticles caused a less severehypocellularity than AZD1152, while AZD2811–cholic nanoparticleshad little effect on bone marrow cellularity and appeared similar tovehicle-treated animals. On day 9, bone marrow cellularity was similarin all treatment groups indicating reversibility of bone marrow toxicityeffects. Additional data using the wider panel of formulations producedsuggest that all formulations with release kinetics at least as slow as thecholic acid formulation (~35-hour release half-life) demonstrate re-duced bone marrow toxicity compared to the parent drug (Fig. 14).

SW620 tumors stained ex-vivo with pHH3 show misalignment ofchromosomes and large polyploid cells can be detected inH& E sections5 days after treatment with AZD1152, AZD2811–oleic or AZD2811–cholic nanoparticles, both indicative of the mechanism of action of anAur B kinase inhibitor (Fig. 13).

No adverse effects attributable to counter-ions was observed in pre-clinical studies described. In addition, many of the counter-ions used inthese studies are approved for parenteral use. For example, olanzapinepamoate is a depot formulation of olanzapine administered intramuscu-larly (Zyprexa) and ethanolamine oleate (Ethamolin) is administered asa local IV (http://www.accessdata.fda.gov/drugsatfda_docs/nda/2009/022173_zyprexa_relprevv_toc.cfm).

Fig. 12. API pharmacokinetics for various nanoparticle formulations compared to parent API at 5 mg/kg dose (A) and for pamoic acid and cholic acid formulations at 0.5 mg/kg (B). Fornanoparticle formulations, plasma concentrations represent total drug levels, and for all formulations aside from “baseline,” plasma half-life is primarily dictated by particle clearancemechanisms (particle plasma half-life of ~15 h). Parent AZD2811 is a solution formulation of the API.

116 Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

3.9. Tumor growth inhibition

Sustained release nanoparticles have been evaluated in immuno-compromised rats or mice bearing a range of human tumor xenograftsin pharmacodynamic and efficacy endpoint studies [19]. In general,the nanoparticle formulations exhibit equivalent or enhanced tumorgrowth inhibition compared to AZD1152, even when administered athalf the dose intensity of the parent drug. In models where more thanone nanoparticle were evaluated, the slower-releasing formulation

Fig. 13. Changes in tumor and bone marrow in nude rats following treatment with AZD1152nanoparticle. Animals were administered placebo nanoparticle, AZD1152 (25 mg/kg oncenanoparticle (25 mg/kg dosed on days 1 and 3; total dose 50 mg/kg). Tumors taken 5 days aH&E staining; bone marrow samples taken 5 and 9 days after the first dose were analyzed by s

always performed comparably or superior to the faster-releasing for-mulation. In some models the slowest release nanoparticle (pamoicacid) demonstrated tumor regression that was more durable than theresponse seen with AZD1152 (Fig. 15).

4. Discussion

Despite promising clinical activity in elderly acutemyeloid leukemiapatients, the broader utility of the aurora B kinase inhibitor AZD2811

(prodrug of AZD2811) drug solution, AZD2811–oleic nanoparticle and AZD2811–cholica day for days 1–4; total dose 100 mg/kg), and AZD2811–oleic and AZD2811–cholicfter first dose were analyzed for pHH3 and polyploidy; as well as tumor cell integrity bytaining with H&E.

Fig. 14. Bone marrow pathology, demonstrating the toxicity observed for AZD1152(prodrug of AZD2811) compared to placebo and nanoparticle formulations with variousrelease kinetics. 0 represents intact bone marrow, 3 represents significant impact onbone marrow. All nanoparticle formulations with release kinetics at least as slow as thecholic acid formulation are effective at reducing the bone marrow toxicity compared tothe parent drug.

117Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

(dosed as prodrug AZD1152)was limited by toxicity and a requirementfor a 7-day infusion [19]. The formulation and preclinical work de-scribed here demonstrate how nanoparticles with the appropriate re-lease kinetics of AZD2811 can widen the therapeutic window toincrease efficacy and reduce toxicity using superior dosing regimens.Empirical determination of the ideal release profile was enabled by pro-duction of nanoparticle formulations with a variety of release profileswithout impacting other nanoparticle attributes that might confoundthe biological results. The data here suggest that formulations with theslowest release kinetics performed best, particularly with regard tobonemarrow toxicitymitigation. The selection of our clinical candidate,the pamoic acid formulation, from the handful of qualifying candidateswith sufficiently slow release kinetics, was based on a number of chem-istry, manufacturing, and control (CMC) considerations, including en-capsulation efficiency, drug load, and process robustness.

Though AZD2811 was the only molecule studied in this body ofwork, it can be considered a model compound for the broad applicabil-ity of HIP in a nanoparticle platform. During formulation screening,

Fig. 15. In vivo tumor growth inhibition in the human SC61 small cell lung cancer primaryexplant model in immunocompromised mice. Animals were administered placebonanoparticle, AZD1152 (25 mg/kg once a day for days 1–4; total dose 100 mg/kg) andAZD2811–pamoic nanoparticle (25 mg/kg dosed on days 1 and 3; total dose 50 mg/kg).Student's t-test statistical analysis was applied comparing Placebo NP with AZD1152and AZD2811-PAM NP groups **p b 0.01 (black symbols) and AZD1152 compared toAZD2811-PAM NP groups (day 4 p b 0.001; red symbols, day 11 p b 0.01; red symbols,all points beyond day 11 are p b 0.001).

counterion candidateswill be determined by the physicochemical prop-erties of the API. For example, extremely weak bases with a pKa of b5will require counterions with acidic groups that are strongenough—typically at least two pKa units below the base. Likewise, aweak acidic drugwill require strongly basic counterions. For encapsula-tion purposes, more hydrophilic drugs (negative log P) will requiremore hydrophobic counterions to allow theHIP to be sufficiently hydro-phobic to allow for encapsulation. Naturally, for developing therapeu-tics candidates, some consideration must also be given to the safetyand pharmaceutical precedence of the counterion itself. The counter-ions used here were chosen based on the following criteria: high lipo-philicity, strong versus weak acidity, and precedent human clinicalexperience as salt formers and/or prodrug ester formers.

We have leveraged some of the general rules from salt formation[32–35] for our nanoparticle ion pairing work such as ensuring a deltapKa (basic drug–acidic counterions) over two pH units apart for coun-terion selection. It should be noted that our nanoemulsion process is amixed solvent system and the solvent effects on pKa for acid and basecan be profound [36–38]. As ionic interactions are the driving forcesfor the HIP formation, these can be dramatically influenced by pH,ionic strength, and the relative concentrations of ions. Thus, the influ-ence of the emulsion and quench pH, ionic strength of quench media,and stoichiometric relationship between drug and counterion must allbe considered using this approach. Solution pH is one of the most im-portant factors in ion formation by nature.

The resulting changes in nanoparticle loading and release kineticscan be approximately predicted by the physicochemical properties ofthe HIP. As shown in Table 3 (from results) and exemplified in Fig.16(A), the drug loading generally increases with counterion hydropho-bicity. Additionally, because these systems exhibit primarily diffusion-driven release, the IVR rate generally decreases with increasing HIPmolar volume as calculated by ACD/ChemSketch Freeware (AdvancedChemistry Development, Inc., Toronto, ON, CAN). For each HIP molarvolume, molar of various counterions (Supplementary Table 1) wasadded to AZD2811 molar volume calculated (≈373 cm3/mol), exceptpamoic acid HIP inwhich the ratio of 2:1 drug to acidwas used to calcu-late. Fig. 16(B) is a plot of HIPmolar volume of [AZD2811+ counterion]versus release, showing a strong inverse correlation.

HIP dissolution kinetics may also play a role in determining the re-lease profile, but we believe that this is less critical for these nanoparti-cle systems, in which the encapsulated components are molecularlydispersed in the polymer matrix. A detailed description of the thermo-dynamic determinants of HIP solubility is beyond the scope of the cur-rent paper. Data (not shown) from HIP nanoparticle batches wherewe have intentionally varied particle Tg by altering polymer molecularweight demonstrate a corresponding change in drug release kinetics.Because HIP solubility or dissolution kinetics in the release media isnot expected to be influenced by the polymer molecular weight withinwhich it is encapsulated, this impact of particle Tg on release kineticssuggests diffusional control of release. Because the HIP must diffusethrough the polymer matrix prior to release, counterion propertiesthat influence theHIP diffusion coefficient such asmolar volume are an-ticipated to impact release rate.

In addition to enabling control of drug loading and release rate, theHIP approach provides advantages to the nanoparticle production pro-cess itself. For example, the higher solubility of the HIP in the organicphase in comparison with the free API enables higher concentrationsof drug to be used for processing, which in turn allows for higher theo-retical drug loading and/or higher solids concentration in the organicphase. Both of these changes typically result in higher encapsulation ef-ficiency and overall process efficiency.

In this study we performed a deeper investigation into the interac-tion of AZD2811 and pamoic acid as amodel HIP system using a numberof orthogonal methods. Solubility enhancement aswell as nuclearmag-netic resonance spectroscopy demonstrated solution-state interactions.Stoichiometric ratios of both encapsulated and released pamoic acid to

Fig. 16. (A)Hydrophobicity of counterion versus drug loading, and (B)molar volumeof (AZD2811+counterion) versus drug release. For pamoic acid formulation, ratios of 2:1 drug to acidwere used for molar volume calculation.

118 Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

API, measured by high-performance liquid chromatography, providedfurther evidence for interaction both through particle formation anddrug release. Raman spectra of the encapsulated HIP compared well topreformed AZD2811-pamoate salt but not to the spectra of the freeforms of the API and counterion.

Although theHIP approach has enabled the selection of nanoparticleformulations that have shown promising preclinical data, we acknowl-edge the historical lack of translation from preclinical results to clinicalproducts in the nano-medicine field [39–41]. Though a number of thesefailures can be attributed to the lack of appropriate pharmaceutical de-velopment, it still must be emphasized that formulation and preclinicalsuccess is no guarantee of positive clinical data. That said, the develop-ment efforts outlined here go beyond producing an academic tool andinclude pharmaceutical development considerations required forbench-to-bedside translation. We have now begun dosing patients ina Phase 1 clinical trial with the AZD2811-pamoate formulation devel-oped using the HIP approach, and hence are one step closer tocommercialization.

5. Conclusions

In summary, the HIP approach to formulating nanoparticles with ad-justable release rateswas used to generate a library of AZD2811 batchesfor analytical and preclinical characterization. This library was used toquickly screen and optimize release kinetics from the nanoparticleproduct. Batches with release half-lives ranging from 2 to N100 hwere produced using the same polymer, nanoparticle size, andnanoemulsion process. Batches with the slowest release kinetics gener-ally outperformed the faster-releasing formulations, particularly withregard to bone marrow toxicity. Vastly improved preclinical efficacyand tolerability data were generated for the pamoic acid lead formula-tion, which has been selected for clinical evaluation (ClinicalTrials.govIdentifier NCT 02579226).

With this formulation tool, release kinetics can be decoupled fromother nanoparticle physicochemical properties, enabling lead candidateselection to occur between batches with comparable particlebiodistribution performance. This approach can provide formulationscientists with a tool for developing themost effective nanoparticle for-mulations for targeted drug delivery and we hope will ultimately resultin products with superior efficacy and toxicity profiles.

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.jconrel.2016.03.026.

Acknowledgments

We would like to thank Drs. Jeff Hrkach, Allen Horhota, and KevinMcDonnell, Jim Murray, and Colin Howes for the helpful scientific dis-cussion and suggestions, Peter Hall for the pathology and JessChmielecki, Jeanne Tran, Ujjwal Joshi, Maria Figueiredo, Zach Lovatt,

Max Mahoney, Nicola Derbyshire, and Paula Taylor for the technicalsupport. We would also like to thank Dr. Robert Prud'homme for thehelpful scientific consultation and advice.

This work was funded by AstraZeneca.

References

[1] P.V. Devarajan, Targeted Drug Delivery: Concepts and Design, Springer, New York,2014.

[2] R.I. Mahato, A.S. Narang, Targeted Delivery of Small and Macromolecular Drugs, CRCPress/Taylor & Francis, Boca Raton, FL, 2010.

[3] D.F. Emerich, C.G. Thanos, Targeted nanoparticle-based drug delivery and diagnosis,J. Drug Target. 15 (2007) 163–183.

[4] R. Singh, J.W. Lillard Jr., Nanoparticle-based targeted drug delivery, Exp. Mol. Pathol.86 (2009) 215–223.

[5] X. Yu, M.V. Pishko, Nanoparticle-based biocompatible and targeted drug delivery:characterization and in vitro studies, Biomacromolecules 12 (2011) 3205–3212.

[6] I.K. Kwon, S.C. Lee, B. Han, K. Park, Analysis on the current status of targeted drugdelivery to tumors, J. Control. Release 164 (2012) 108–114.

[7] T.M. Allen, P.R. Cullis, Drug delivery systems: entering the mainstream, Science 303(2004) 1818–1822.

[8] A. Budhian, S.J. Siegel, K.I. Winey, Controlling the in vitro release profiles for a sys-tem of haloperidol-loaded PLGA nanoparticles, Int. J. Pharm. 346 (2008) 151–159.

[9] S.D. Li, L. Huang, Pharmacokinetics and biodistribution of nanoparticles, Mol. Pharm.5 (2008) 496–504.

[10] D.Y. Arifin, L.Y. Lee, C.H. Wang, Mathematical modeling and simulation of drug re-lease from microspheres: implications to drug delivery systems, Adv. Drug Deliv.Rev. 58 (2006) 1274–1325.

[11] A. Kumari, S.K. Yadav, S.C. Yadav, Biodegradable polymeric nanoparticles based drugdelivery systems, Colloids Surf. B: Biointerfaces 75 (2010) 1–18.

[12] M. Polakovic, T. Gorner, R. Gref, E. Dellacherie, Lidocaine loaded biodegradablenanospheres. II. Modelling of drug release, J. Control. Release 60 (1999) 169–177.

[13] F. Alexis, Factors affecting the degradation and drug-release mechanism ofpoly(lactic acid) and poly[(lactic acid)-co-(glycolic acid)], Polym. Int. 54 (2005)36–46.

[14] G. Mittal, D.K. Sahana, V. Bhardwaj, M.N. Ravi Kumar, Estradiol loaded PLGA nano-particles for oral administration: effect of polymer molecular weight and copolymercomposition on release behavior in vitro and in vivo, J. Control. Release 119 (2007)77–85.

[15] S.M. Ansell, S.A. Johnstone, P.G. Tardi, L. Lo, S. Xie, Y. Shu, T.O. Harasym, N.L.Harasym, L. Williams, D. Bermudes, B.D. Liboiron, W. Saad, R.K. Prud'homme, L.D.Mayer, Modulating the therapeutic activity of nanoparticle delivered paclitaxel bymanipulating the hydrophobicity of prodrug conjugates, J. Med. Chem. 51 (2008)3288–3296.

[16] C. Luo, J. Sun, B. Sun, Z. He, Prodrug-based nanoparticulate drug delivery strategiesfor cancer therapy, Trends Pharmacol. Sci. 35 (2014) 556–566.

[17] J.J. Wang, K.S. Liu, K.C. Sung, C.Y. Tsai, J.Y. Fang, Lipid nanoparticles with different oil/fatty ester ratios as carriers of buprenorphine and its prodrugs for injection, Eur. J.Pharm. Sci. 38 (2009) 138–146.

[18] S.D. Clas, R.I. Sanchez, R. Nofsinger, Chemistry-enabled drug delivery (prodrugs): re-cent progress and challenges, Drug Discov. Today 19 (2014) 79–87.

[19] S. Ashton, Y.H. Song, J. Nolan, E. Cadogan, J. Murray, R. Odedra, J. Foster, P.A. Hall, S.Low, P. Taylor, R. Ellston, U.M. Polanska, J. Wilson, C. Howes, A. Smith, R.J.A.Goodwin, J.G. Swales, N. Stritmatter, Z. Takáts, A. Nilsson, P. Andren, D. Trueman,M. Walker, C.L. Reimer, G. Troiano, D. Parsons, D. De Witt, J. Ashford, J. Hrkach, S.Zale, P.J. Jewsbury, S.T. Barry, Aurora kinase inhibitor nanoparticles target tumorswith favorable therapeutic index in vivo, Sci. Transl. Med. 8 (2016), 325ra17.

[20] J. Hrkach, D. Von Hoff, M.M. Ali, E. Andrianova, J. Auer, T. Campbell, D. De Witt, M.Figa, M. Figueiredo, A. Horhota, S. Low, K. McDonnell, E. Peeke, B. Retnarajan, A.Sabnis, E. Schnipper, J.J. Song, Y.H. Song, J. Summa, D. Tompsett, G. Troiano, T. VanGeen Hoven, J. Wright, P. LoRusso, P.W. Kantoff, N.H. Bander, C. Sweeney, O.C.Farokhzad, R. Langer, S. Zale, Preclinical development and clinical translation of a

119Y.H. Song et al. / Journal of Controlled Release 229 (2016) 106–119

PSMA-targeted docetaxel nanoparticle with a differentiated pharmacological pro-file, Sci. Transl. Med. 4 (2012), 128ra139.

[21] J.D.Meyer, M.C.Manning, Hydrophobic ion pairing: altering the solubility propertiesof biomolecules, Pharm. Res. 15 (1998) 188–193.

[22] R. Gaudana, A. Parenky, R. Vaishya, S.K. Samanta, A.K. Mitra, Development and char-acterization of nanoparticulate formulation of a water soluble prodrug of dexameth-asone by HIP complexation, J. Microencapsul. 28 (2011) 10–20.

[23] N.M. Pinkerton, A. Grandeury, A. Fisch, J. Brozio, B.U. Riebesehl, R.K. Prud'homme,Formation of stable nanocarriers by in situ ion pairing during block-copolymer-directed rapid precipitation, Mol. Pharm. 10 (2013) 319–328.

[24] C.S. Lengsfeld, D. Pitera, M. Manning, T.W. Randolph, Dissolution andpartitioning behavior of hydrophobic ion-paired compounds, Pharm. Res. 19(2002) 1572–1576.

[25] S.H. Choi, T.G. Park, Hydrophobic ion pair formation between leuprolide and sodiumoleate for sustained release from biodegradable polymeric microspheres, Int. J.Pharm. 203 (2000) 193–202.

[26] W.G. Dai, L.C. Dong, Characterization of physiochemical and biological properties ofan insulin/lauryl sulfate complex formed by hydrophobic ion pairing, Int. J. Pharm.336 (2007) 58–66.

[27] A. Patel, R. Gaudana, A.K. Mitra, A novel approach for antibody nanocarriers devel-opment through hydrophobic ion-pairing complexation, J. Microencapsul. 31(2014) 542–550.

[28] S. Sun, N. Liang, H. Piao, H. Yamamoto, Y. Kawashima, F. Cui, Insulin-S.O (sodiumoleate) complex-loaded PLGA nanoparticles: formulation, characterization andin vivo evaluation, J. Microencapsul. 27 (2010) 471–478.

[29] L. Yang, F. Cui, K. Shi, D. Cun, R. Wang, Design of high payload PLGA nanoparticlescontaining melittin/sodium dodecyl sulfate complex by the hydrophobic ion-pairing technique, Drug Dev. Ind. Pharm. 35 (2009) 959–968.

[30] D.T. Manallack, The pK(a) distribution of drugs: application to drug discovery,Perspect Medicin Chem. 1 (2008) 25–38.

[31] V.V. Ambardekar, S.T. Stern, NBCD pharmacokinetics and bioanalytical methods tomeasure drug release, in: D.J.A. Crommelin, J.S.B. de Vlieger (Eds.), Non-BiologicalComplex Drugs, Springer International Publishing 2015, pp. 261–287.

[32] P.H. Stahl, C.G. Wermuth, Handbook of Pharmaceutical Salts: Properties, Selection,and Use, Verlag Helvetica Chimica Acta, Zürich, 2011.

[33] A.T. Serajuddin, Salt formation to improve drug solubility, Adv. Drug Deliv. Rev. 59(2007) 603–616.

[34] A.T. Serajuddin, P.C. Sheen, D. Mufson, D.F. Bernstein, M.A. Augustine, Physicochem-ical basis of increased bioavailability of a poorly water-soluble drug following oraladministration as organic solutions, J. Pharm. Sci. 77 (1988) 325–329.

[35] W.Q. Tong, G. Whitesell, In situ salt screening—a useful technique for discovery sup-port and preformulation studies, Pharm. Dev. Technol. 3 (1998) 215–223.

[36] R.J. Bastin, M.J. Bowker, B.J. Slater, Salt selection and optimisation procedures forpharmaceutical new chemical entities, Org. Process Res. Dev. 4 (2000) 427–435.

[37] G. Garrido, V. de Nogales, C. Ràfols, E. Bosch, Acidity of several polyprotic acids, ami-odarone and quetiapine hemifumarate in pure methanol, Talanta 73 (2007)115–120.

[38] H. Kim, C.R. Babu, D.J. Burgess, Quantification of protonation in organic solventsusing solution NMR spectroscopy: implication in salt formation, Int. J. Pharm. 448(2013) 123–131.

[39] P. Ruenraroengsak, J.M. Cook, A.T. Florence, Nanosystem drug targeting: facing up tocomplex realities, J. Control. Release 141 (2010) 265–276.

[40] Y.H. Bae, K. Park, Targeted drug delivery to tumors: myths, reality and possibility, J.Control. Release 153 (2011) 198–205.

[41] A. Kumar, L.A. Dailey, B. Forbes, Lost in translation: what is stopping inhalednanomedicines from realizing their potential? Ther. Deliv. 5 (2014) 757–761.