Powder Technology 258 (2014) 222–233

Contents lists available at ScienceDirect

Powder Technology
journal homepage: www.elsevier.com/locate/powtec

A control strategy for bioavailability enhancement by size reduction:

Effect of micronization conditions on the bulk, surface and blending
characteristics of an active pharmaceutical ingredient
Dolapo Olusanmi a,⁎, Dimuthu Jayawickrama a, Dongsheng Bu a, Gary McGeorge a, Helen Sailes c,
Joanne Kelleher c, John F. Gamble b, Umang V. Shah d, Mike Tobyn b
a
Bristol-Myers Squibb Pharmaceuticals, 1 Squibb Drive, New Brunswick, NJ 08903, USA
b
Bristol-Myers Squibb Pharmaceuticals, Reeds Lane, Moreton, Wirral CH46 1QW, UK
c
Bristol-Myers Squibb Pharmaceuticals, Watery Lane, Swords, Dublin, Ireland
d
Imperial College London, South Kensington, London SW7 2AZ, UK

a r t i c l e i n f o a b s t r a c t

Article history: In a Quality by Design (QbD) development environment the effect of early process parameters on downstream
Received 13 September 2013 manufacturing parameters, and the ultimate effect on drug product quality, need to be understood. For poorly
Received in revised form 7 March 2014 soluble drugs, size reduction is frequently employed to obtain consistent in-vivo exposures. As a result,
Accepted 8 March 2014
micronization is a key early stage processing step for many active pharmaceutical ingredients (APIs).
Available online 14 March 2014
This paper demonstrates the effect of varying micronization conditions on an API for which micronization is
Keywords:
deemed necessary to ensure consistent drug delivery after human administration. Material micronized to differ-
Micronization ent extents are confirmed as different by surface area, surface energy, particle size analysis, bulk density and sur-
Milling face adhesion measurements.
QbD These material characteristics can be correlated with the outcomes from a key processing step, blending. The evo-
Blending lution of the blending process is followed using PAT techniques, so that an overall understanding of the relation-
Surface energy ship between particle properties and blend uniformity can be demonstrated.
Surface area Execution of such a study during drug development can enable selection of the appropriate control strategy to
Cohesion
ensure production of API in the desired range where consistent optimal bioavailability and downstream process-
ability are achieved.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction
In the pharmaceutical industry a significant percentage of marketed
drugs [1] and new chemical entities [2] are classified as BCS Class II,
where bioavailability is controlled by the drug release from the product
matrix due to the inherent low solubility of the drug substance. In order
to improve the absorption of such drug substances which may be first in
line in terms of their therapeutic benefits pharmaceutical scientists
often employ alternative approaches to designing the active pharma-
ceutical ingredient (API) and also the finished product. These include
but are not limited to amorphous forms, salt formation of the API, co-
crystal formation, spray dried intermediates, deposition on inorganic
mesoporous materials and particle size control. The formation of salts
of ionisable drug entities, often weak salts [3], while increasing solubil-
ity will also alter properties such as hygroscopicity, electrostatics, melt-
ing point, polymorphism and mechanical properties; all of which can
impact chemical stability [4]. Additionally, not all drug entities are
⁎ Corresponding author. Tel.: +1 732 227 6131.
E-mail address: dolapo.olusanmi@bms.com (D. Olusanmi).
ionisable and in fact developing salt forms of APIs is not beneficial in
every case [5]. In some cases crystal size modification can be employed.
This can be achieved either through bottom-up (crystallization) or top-
down (size reduction) approaches to obtain the desired dissolution rate
enhancement.
Furthermore, for low dose formulations particle size control is need-
ed to ensure content uniformity (CU), a regulated quality standard of
oral solid dosage forms. A key aspect of the drug development process
is crystallization, and a primary challenge of crystallization is associated
with difficulties in controlling the size and shape distribution of crystals,
especially when complex crystallization processes are involved [6,7],
and with increasing complexity of molecules [8]. In cases where bio-
availability of drug entities can be improved by size and shape modifica-
tion the ideal scenario would be to “dial a particle size and shape
distribution” using robust bottom-up approaches from crystallization
that can be scaled up to commercial scale, while maintaining required
polymorphic form, crystallinity and stability. The use of supercritical
fluid technology to produce crystalline API of narrow size distribution
in the sub-micron range has been developed and discussed [9,10], how-
ever, instances where the process has been scaled up for commercial

http://dx.doi.org/10.1016/j.powtec.2014.03.032
0032-5910/© 2014 Elsevier B.V. All rights reserved.
 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233
scale manufacture have not been found, therefore as yet it is not a com-
mercially viable option.
Rohrs et al. [11] developed a method to estimate the particle size
limits needed to ensure that content uniformity criteria for a solid dosage
form are met based on median particle size, particle size distribution
spread and dose. The method assumes uniform mixing of the blend,
and a log-normal particle size distribution for the API, with key input fac-
tors such as d[0.5], ratio of the d[0.9]/d[0.5], dose and API particle density.
For a given dose the maximum value of d[0.5] that can be accommodated
without the risk of failing blend content uniformity is affected by the ratio
of the d[0.9]/d[0.5]. As the latter increases, the d[0.5] value has to be re-
duced in order to pass content uniformity. This is particularly applicable
to low dose formulations where the presence of large particles would
be expected to have a greater impact on CU. Further reduction of the
dose at a specified tablet weight would increase the risk of CU failure.
As a consequence of the aforementioned factors when particle size
reduction is required to enhance bioavailability and/or ensure content
uniformity milling, particularly micronization, is often widely employed.
The mechanical activation during milling may inadvertently cause vary-
ing and uncontrolled degrees of disruption to the particle surfaces [9], as
well as exposure of non-habit surfaces due to plastic deformation and
fracture. Therefore the principles of Quality by Design (QbD) should be
applied to this process.
The implementation of QbD, a systematic approach to drug develop-
ment, in the pharmaceutical industry is advocated by the regulatory
agencies, and the principles are given in the ICH Q8 guideline [12].
This practice involves determining the functional relationships that
link material attributes and processing parameters to product ‘critical
quality attributes’ (CQAs). The use of Process Analytical Technology
(PAT) during development enhances process understanding thus lead-
ing to improved control of the manufacturing process [13]. In addition
to drug product CQAs that impact the desired quality, safety and effica-
cy, CQAs can additionally include those properties that affect down-
stream processability [12]. A substantial effort is made to ensure the
former part because without it a product cannot make it onto the mar-
ket, however, the latter is also important. The design space, i.e. the link-
age between the process inputs and CQAs, should be determined for
each unit operation that forms part of the manufacturing process. For
micronization the optimal design space is achieving the powder proper-
ties required for the biopharmaceutics requirements, while ensuring
the best possible downstream processability. Downstream processabil-
ity can be defined in terms of ease of bulk powder handling, processing
time to ensure quality etc.; essentially all these can be summed up into
processing efficiency. Without establishing an optimal design space for
the micronization operation a wide range of materials can be produced
which meet the specifications for particle size, e.g. d[0.9] b 40 μm, but
which may differ widely in terms of processability. As discussed by
Olusanmi et al. [14] the 3-dimensional long-range order associated
with crystalline organic materials can result in asymmetrical properties
of the various crystal facets which can consequently impact their me-
chanical properties. The surface chemistry on each crystal facet may
vary as the various crystallographic planes dissect the unit cell (the fun-
damental building block of crystals repeated in space to form the crystal
structure) at different planar orientations and angles. Thus milling may
result in the exposure of differing chemical groups, and differing surface
concentration of those groups [15]. As a consequence, the surface ener-
getics of a micronized sample may be significantly different to the orig-
inal un-milled material.
The surface energetics of pharmaceutical powders are often charac-
terized in terms of their dispersive surface energy, with a higher disper-
sive surface energy often suggested to represent a more “reactive”
surface [16]. The total surface free energy of a solid can be split into dis-
persive (non polar) and specific (polar) components. The former are
non-specific interactions which are due to long-range London disper-
sive forces, while the latter are specific short-range directional chemical
interactions which involve charge redistribution and sharing [17].
223
Techniques used to characterize the surface energetics of pharma-
ceutical powders include IGC (Inverse Gas Chromatography) [9,
18–22], sessile drop contact angle methods on compacted powders,
and contact angle methods on macroscopic single crystals [15,23,24].
There are several drawbacks associated with the sessile drop contact
angle methods and these have been discussed extensively by a number
of authors [25–27]. Furthermore dispersive surface energy measure-
ments by liquid wetting angle techniques are difficult to implement re-
producibly on free-flowing powders [28]. For sessile drop contact angle
measurements on individual crystal faces large specially grown crystals
are required, but these give better indications of dispersive surface ener-
gy anisotropy resulting from surface chemistry differences, rather than
an averaged value obtained from powder samples [15,24]. Surface ener-
gy values determined by measurements at infinite dilution, as described
by Thielmann [13], are mostly biased towards the highest energy sites of
the powder sample [29] because the amount of probe injected only
covers a small percentage, b 0.1%, of the powder sample [17]. This has
the benefit of distinguishing subtle differences in the surface properties,
even in the case of nominally similar batches of the same material [9].
In some cases the distribution of active sites may be more relevant to
the practical properties of a material than the high energy sites [16]. This
is particularly pertinent in cases where the effect of milling on disper-
sive surface energy is of interest. In the case of materials with cleavage
planes, following exposure of such faces during processes such as mill-
ing or agitated drying, the properties of the cleavage plane may domi-
nate the surface energetics of the resulting product. Heng et al. [18]
observed an increase in the dispersive surface energy, measured by
IGC, of paracetamol form I powder with milling, and this was attributed
to an increase in the proportion of the (010) facet, the cleavage plane,
which exhibited the highest dispersive surface energy compared to
the other facets as determined by sessile drop contact angle measure-
ments on individual crystal faces of macroscopic crystals. Exposure of
non-habit surfaces due to fracture may subsequently introduce further
complexities in the interaction of organic particulates with different
chemical entities compared to unmilled material. Overall, processes
which involve size reduction may further compound the dispersive sur-
face energy heterogeneity of the product due to exposure of new sur-
faces coupled with the possible disordering of crystalline surfaces
caused by mechanical stress. Therefore when micronization is required
it is essential that during development, the effect of micronization ex-
tent within the acceptable range for the biopharmaceutical require-
ments on powder properties (both surface and bulk) and downstream
processability is understood.
Work has been reported in literature on the effect of milling on the
surface properties of APIs and particulate materials in general. While
some authors have reported decreases in surface energy with milling
[19,21,30], others have observed milling to increase the surface energy
of powders [18,31]. Where surface energy increases with milling extent
has been reported it has often been attributed to the exposure of higher
energy crystal facets and/or the creation of higher energetic surfaces on
the particles. Typically a number of these publications report changes to
the degree of crystallinity of the milled material either due to localized
lattice disorder and/or reduced crystallinity [20,30,32], changes to the
surface chemistry of milled material [18,31]. An important question,
which has not been addressed, is what is the effect of micronization
and associated change in surface energetics on the bulk handling and
processing behavior of pharmaceutical API during development and
manufacture? Publications where the effect of these milling-induced
changes are demonstrated are limited e.g. Vippagunta et al. [33]. To en-
able better implementation of quality by design and more efficient drug
development it is important to understand the impact of such changes
to other downstream processing operations, and this subject would be
the focus of this present work.
The objective of this milling study is to establish a control strategy
for producing micronized API with consistent powder properties in
the desired target product profile for dissolution enhancement.
224 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233
Furthermore the effect of milling extent of a commercially available ac-
tive pharmaceutical ingredient, compound X, on its powder properties
and subsequent downstream processing behavior is determined. Both
the surface and bulk properties of the milled material are characterized,
with the former conducted using B.E.T. surface area and dispersive sur-
face energy heterogeneity analysis. Subsequently the effect of the deter-
mined powder properties of the API batches during the blending of a
formulation of low (b5% w/w) drug loading is assessed. Batches with
a wide range of milling parameters and particle sizes were produced
to determine a suitable design space that would meet all CQA
requirements.
2. Materials and methods
2.1. Materials
The API used in this study is manufactured by Bristol–Myers Squibb
(BMS). Excipients typically used in a dry granulation process were uti-
lized in the study.
The probe liquids for contact angle measurements, ethylene glycol
(N99%), dimethyl sulfoxide (N99%), and glycerol (N 99%) were obtained
from Sigma-Aldrich. Formamide (N99.5%), and diiodomethane (N 99%)
were obtained from Acros Organics.
2.2. Milling
0.8 kg of a batch of the crystallized API, Batch A, was milled in 4
separate lots of 200 g to different extents, runs 1 to 4, using a 4-
inch jet mill. Different milling conditions were used with the aim
of producing materials with distinguishable bulk and surface prop-
erties to enable evaluation on the CQAs, and thus aid implementa-
tion of the appropriate control strategy. An additional batch, Batch
B, of the same API was milled to three extents as part of a milling
cross trial analysis i.e. low pressure/high feed rate, high pressure/
low feed rate and mid-point between the two, with no limits set
on particle size distribution. In total 8 batches milled to different ex-
tents were assessed in this study.
2.3. Content uniformity (CU) modeling
A proprietary implementation of the Rohrs [11] CU model was used
to assess the potential effect of particle size on CU for a low drug load,
b5%, and low unit dose b 10 mg.
2.4. Surface area
The surface area of the milled samples was obtained using a Gemini
2390a surface area analyser supplied by Micromeritics, USA, with Nitro-
gen as the adsorbate. About 0.5–1 g of sample was analyzed for surface
area measurements. To ensure accurate surface area determination,
samples were outgassed prior to analysis for 24 h at 50 °C using nitrogen
gas to remove residual moisture adsorbed on the surface of the particles
which might affect the accuracy of the results. During analysis, the sam-
ples were equilibrated for 10 s followed by an evacuation rate of
66.6 kPa/min. Multi-point B.E.T. measurements were taken in the
range 0.05–0.3 p/p0. Two samples were analyzed for each batch and
the average determined.
2.5. Surface energy analysis
The dispersive and specific surface energy of the unmilled and
milled samples was characterized using a Surface Energy Analyser
(SEA), supplied by Surface Measurement Systems (Alperton, UK). The
SEA, a next generation Inverse Gas Chromatography (IGC) system, has
been used to successfully characterize the surface energy of pharmaceu-
tical materials at finite dilution, and has been previously described in an
earlier publication [31]. An advantage that the SEA has over the tradi-
tional IGC is that the former is designed to determine the surface en-
ergy at targeted surface coverages; increasing the concentration of
probe molecules will increase the number of lower energy (“less
active”) sites that are involved in the interaction with the probe
molecules as high energy sites are interacted with preferentially
[16,34].
Prior to analysis each milled sample was filled into silanised glass
capillary columns of 3 mm inside diameter and 300 mm length by ver-
tical tapping, with both ends of the columns stoppered with silanised
glass wool to prevent the movement of the sample bed during analysis.
The dispersive surface energy analysis method involved the injec-
tion of a series of dispersive n-alkane probes, namely decane, nonane,
octane and heptane, the volumes of which were controlled in order to
obtain specific surface coverages in the range of 0.15 up to 20%. For spe-
cific surface energy analysis two mono-polar probes, chloroform (Lewis
acid) and ethyl acetate (Lewis base), were also injected at equivalent
surface coverages as used in the dispersive surface energy method. Sur-
face adsorbed water and other impurities were removed through condi-
tioning of the columns using dry helium for 2 h at 30 °C/0% RH. For one
of the samples duplicate analysis of the same sample column was con-
ducted to ensure that these degassing conditions were sufficient to
clean the samples without impacting a physical change during analysis.
The measured results remained unchanged for the duplicate analyses. A
column gas flow rate of 10 cm3/min was used for all samples, and the
columns were maintained at 30 °C and 0% RH. To determine the column
dead time, methane gas was injected at a volume of 0.208 cm3. This ac-
counts for any differences in the packing behavior of the powder sam-
ples due to their morphology.
The dispersive surface energy data were analyzed using the Dorris
and Gray [35] approach at peak max. Voekel et al. [36] reported that dis-
persive surface energy values calculated from the Dorris and Gray [35]
method were comparable to those obtained using the Schultz [37]
method.
For the Dorris and Gray approach used here the dispersive compo-
nent of the surface energy is obtained from a plot of [RT ln VN] versus
the carbon number of the alkane probes, where R is the universal gas
constant (J/mol K), T is the temperature (K), and VN is the net retention
volume, a fundamental surface thermodynamic property of the solid–
vapor interaction, described in Eq. (1).


j T
VN ¼  F  ðtR −t0 Þ  ð1Þ
m 273:15
where j is the James–Martin correction, m is the mass of the sample in
the column (g), F is the carrier gas flow rate (cm3/min), tR is the reten-
tion time of the probe (minutes), t0 is the mobile phase hold-up time
(minutes) and T is the column temperature (K). The dispersive surface
energy can be calculated using the peak max and the peak center of
mass, the latter of which accounts for non-normal peaks in the elution
data. In this study tailing in the elution peaks was not observed there-
fore the peak max approach was used.
Determination of the specific component of the surface energy was
conducted using the approach described by Dong et al., [38] from a
plot of [RT ln VN] versus the molar deformation polarization of the
probes, PD, as described in Eq. (2).
0  1
2
MW  r −1
PD ¼ @   A ð2Þ
ρl  r2 þ 2
where MW, r, and ρl are the molar mass, reflective index, and liquid
density, respectively, of the probe.
 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233 225

−
The acid (Lewis acceptor), γ+
s and base (Lewis donor), γs numbers estimated 1.2 μm and 0.05 μm surface roughness as specified by the
of the solid based on the van Oss [39] concept were obtained from the manufacturer. The WFA test is proposed to determine the frictional ad-
specific free energy values using Eq. (3). hesion of powder samples towards a specific surface e.g. stainless steel.
 However the authors of this work are also interested to assess whether
  
þ = − = or not the set-up can be used to determine differences in adhesion, if
1 1
− þ
−ΔG ¼ NA  a  2 γl  γs þ γl  γs
2 2
ð3Þ
any, due to increased “reactivity” of the powder sample during
particle-surface interactions as a result of milling propensity. The tests
where ΔG is the specific free energy of adsorption, NA is Avogadro's were conducted in a 25 mm bore, 10 ml borosilicate glass vessel. This
number, a is the cross-sectional area of the probe molecule, and γ+
l and
set up has the advantage that a smaller amount of material is required
γ−
l represent the electron acceptor and donor parameters, respectively,
compared to the bulk density test described above. However the
of the probe molecule. The specific surface energy was calculated using amount of sample used is still greater than desired, particularly in the
− early stages of drug development where only small amounts of API
the geometric mean of γ+ s and γs as described in Eq. (4).
are available. Following a conditioning cycle (as described above) and
consolidation of the powder bed using a vented piston a step sequence
 1 of predetermined normal stress of 15 kPa was applied while the afore-
SP − þ =2
γS ¼ 2  γs  γs ð4Þ mentioned wall friction disc was rotated and the frictional torque mea-
sured over a range of stresses between 1 and 18 kPa. It is assumed that
where γSP
s is the specific surface energy of the sample. The total surface only the surface of the powder bed is sheared by the disc. The shearing
energy is the sum of the dispersive and specific surface energy values. stress is calculated from the measured frictional torque and the wall co-
efficient of friction is determined from the slope of the shear stress ver-
2.6. Particle size analysis sus normal stress data. This described test method was developed by
Freeman Technology, and the wall friction coefficient is defined below
Particle size characterization was carried out using two techniques, in Eq. (5);
laser light scattering and image based analysis.
shear stress
2.6.1. Laser light scattering (LLS) Wall friction coefficient; μ ¼ tanðKÞ ¼ ð5Þ
normal stress
Laser light scattering was carried out on a Malvern Mastersizer 2000
(Malvern Instruments, Malvern, UK), equipped with a liquid dispersion
system. The dispersion medium was an aqueous solution of Tween 80 where Ø is the friction angle.
saturated with the API and filtered. For particle size measurements The coupons were gently but thoroughly cleaned with acetone be-
50 mg of API were dispersed into 5 ml of the dispersion media and fore each analysis, and all tests on the FT4 rheometer were carried out
vortexed for 15 s. Measurements were taken over 10 s, at obscuration in environmentally controlled laboratory conditions.
levels between 10 and 25%. Ultrasonication was applied during the
measurements at 25% level over 10 s. 2.8. Sessile drop contact angle measurements on stainless steel

2.6.2. Image based analysis A Krüss Drop Shape Analyser DSA 10 (Krüss GmbH, Hamburg,
Image based particle size analysis was conducted using a Malvern Germany) was used for the advancing contact angle measurements.
Morphologi G3 particle characterisation system (Malvern Instruments, Ethylene glycol, formamide, diiodomethane and dimethyl sulfoxide
Malvern, UK). Samples were dry dispersed using the systems automat- were used as probe liquids, properties of which are reported in
ed sample dispersion unit onto a glass plate situated on an automated Table 1. Contact angle data using water could not be obtained due to
sample stage. Particle imaging was conducted using a 20× magnifica- the liquid completely wetting the stainless steel surface such that an
tion lens with vertical z-stacking enabled, taking an additional 5 planes angle could not be measured. Measurements were obtained in open
above the focal plane (equivalent to 27.2 μm in total), to account for 3- air at ambient conditions. Probe liquid droplet was monitored using
dimensionality within the samples. Morphological filtering of the raw CCD camera and the shape of the droplet was fitted with a tangent
data was conducted, in order to remove partially imaged/overlapping method to obtain contact angle data using the Drop Shape Analysis soft-
particles, on a sample by sample basis using a combination of convexity, ware (DSA version 1.0) (Krüss GmbH, Hamburg, Germany). A minimum
solidity and particle width filters. of 4 droplets on 3 different stainless steel coupons were measured. Be-
fore analysis with each probe liquid the coupons were placed in 50 ml
2.7. Bulk density and wall friction angle characterization of acetone and sonicated for 1 min at 30 °C. The contact angle data
was analyzed using the method reported by Owens and Wendt [41] to
Bulk density was characterized using the FT4 powder rheometer determine surface energy.
(Freeman Technology, Worcestershire, UK). The API sample is loaded
into a fixed volume vessel and “conditioned” before any analysis. During 2.9. Blending
the conditioning cycle the FT4 rheometer blade traverses downwards
into the powder bed and then upwards at conditions which effect slic- To determine the effect of the milling on the processing behavior of
ing rather than compacting of the powder bed. This conditioning cycle the material blending of the milled batches was conducted.
gently displaces the powder in order to loosen and slightly aerate it,
and therefore makes the result independent of operator's method of
loading powder into the sample vessel or excess air [40]. Subsequently
Table 1
excess powder is removed with the aid of a vessel ‘split’ so that the pow-
Properties of liquid probes used for sessile drop contact angle measurements.
der is level with the top of the vessel which has a fixed volume of 25 ml.
p
Bulk density is determined as the mass of the powder in this 25 mm Probe liquids Density (kg/m3) γLV (mJ/m2) γdL (mJ/m2) γL (mJ/m2)
bore, 25 ml vessel. Duplicate analysis of each milled batch was Diiodomethane 3325 50.8 50.8 0.0
conducted. Formamide 998 58.3 32.3 26
The wall friction angle (WFA) of the powders was also evaluated Ethylene glycol 1109 48.0 29.0 19.0
Dimethyl sulfoxide 1100 43.6 34.9 8.7
using the FT4 Powder Rheometer and 304 stainless steel discs of
226 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233

2.9.1. Sample preparation (a) Two NIR blending profiles. A: fast and B: slow
The API and excipients were charged into the bin prior to the blend
120
operation in an order such that the API was sandwiched in the powder

NIR predicted potency

bed. The batch size for each run was ~1.4 kg. 100

2.9.2. Blending and NIR monitoring 80

A 6 L size tote bin (A&M process equipment Ltd, Ontario, 60 A
Canada) was used for blending using a portable laboratory tumble
B
blender (Model ATS050LP, A&M process equipment Ltd, Ontario, 40
Canada). A 3-inch bin lid was custom modified with sapphire window
20
to house a non contact online NIR instrument. The NIR spectra of blend-
ing powder were acquired with a Brimrose 5030 near infrared spectro- 0
photometer (Brimrose Inc., Sparks, MD). The blender was operated at 0 100 200 300 400
15 RPM for all the blending steps. The total number of revolutions for Rotations
blending was set to 400. At each revolution NIR spectra were acquired
when the NIR is closest to the ground (i.e. at the bottom of the bin).
(b) Ln(RSDe) vs revolutions from
This position ensured complete coverage of the sapphire window on Figure 1(a) with linear fit.
the bin lid by the powder. The general acquisition parameters were: 4.5
wavelength range 1100–2300 nm, number of scans 15, and resolution 4
2 nm. All the data were acquired in reflection mode. A custom-made 3.5
software based on MatLab (Matlab Inc., Natick, MA, USA) and
3
PLS_Toolbox (Eigenvector Research Inc., Wenatchee, WA, USA) was de-

Ln RSDe
2.5 A
veloped in house to determine real time API potency during blending.
2 B

2.9.3. Calibration model 1.5

Linear (A)
A multivariate calibration model based on Partial Least Square (PLS) Linear (B)
1
was developed and validated to quantitatively determine the amount
API in the blend. The calibration model was developed using offline 0.5
samples encompassing the API potency range. The gravimetric amount 0
present in the calibration standard was used as the reference value for 0 20 40 60 80 100
calibration. Absorbance spectra were first pretreated with standard nor- Revolutions
mal variate (SNV), followed by second derivative (Svaitzky_Golay, 11
smoothing and 2nd order polynomial). API peak region of the second Fig. 1. (a): Two NIR blending profiles. A: fast and B: slow. (b): Ln(RSDe) vs revolutions
from Fig. 1 (a) with linear fit.
derivative spectra were used for the calibration model. A calibration
model based on Partial Least Square (PLS) was developed with one fac-
tor. The cross validation error for the model was ~3%.

2.9.4. Quantitative measurement of blending profiles

In this study it was necessary to develop a quantitative parameter in of revolutions (r) can be derived assuming a first order kinetics
order to compare different blending profiles. The kinetics and homoge- (Eq. (7)). Eq. (8) enables one to determine a rate, or blending, coeffi-
neity of blending can be described using the relative standard deviation cient (k) to quantify the blending process. As an example Fig. 1(a) illus-
(RSD) of the content of a component in a mixture. A novel approach to trates two blending profiles, one fast (A) and another slow (B)
quantify blending profiles has been described by Shi et.al. [42] using a generated with online NIR and multivariate calibration, and the deter-
term called the RSDE as defined in Eq. (6) and its relationship to first mined LnRSDE profiles are shown in Fig. 1(b). These plots show two dis-
order mixing kinetics. tinct linear relationships between LnRSDE and revolutions characterized
by their respective rate constants (slope of the linear regression plot).
0vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 The use of quantitative rate constants allows one to compare blending
uX 2
u ðxi −EÞ profiles using a single parameter, hence minimizing the subjectivity of
u
Bt C
B ðn−1Þ C the analysis.
RSDE ¼ B
B
C
C ð6Þ
@ E A

Table 2
dRSDE Particle size distribution of batches obtained from jet-milling process and milling param-
− ¼ ðkRSDE Þ ð7Þ
dr eters.

Batch number Milling parameters Particle size (μm)

−LnRSDE ¼ ðkr þ cÞ ð8Þ Feedrate (Kg/h) Pressure (Bar) d[0.1] d[0.5] d[0.9]

Batch-A Run 1 9.6 1.5 2.17 20.5 83.5

where xi is an individual spectrums determined concentration for one Batch-A Run 2 9.6 2.0 1.94 15.9 59.9
Batch-A Run 3 8.4 3.0 1.29 7.92 33.6
revolution, E is the expected target mean concentration and n is the Batch-A Run 4 7.2 3.0 1.02 5.40 19.6
number of measurements. Consequently, the RSDE term is a measure Batch-B Unmilled 51.0 127.0 350.4
of the relative deviation for the target composition, as opposed to the Batch-B Cross trial 1 7.2 5.0 1.13 6.31 19.9
traditional RSD metric that looks at the difference from the average of Batch-B Cross trial 2 3.6 9.0 0.69 2.43 5.97
Batch-B Cross trial 3 5.5 7.0 1.09 4.50 11.7
the metric. A linear relationship (Eq. (8)) between RSDE and the number
 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233 227

Fig. 2. Geometric (volume based) particle size distribution profiles of milled batches, measured using Malvern Morphologi G3.

3. Results and discussion
3.1. Particle size distribution and content uniformity
The Malven LLS particle size distributions attained for the milled
samples and the mill settings used to achieve them are shown in
Table 2.
The attained d[0.9] and d[0.5] values cover a wide range; an overlay
of the Morphologi G3 determined PSD profiles of all the batches shown
in Fig. 2. It is noted that while the general trends are similar in both cases
the median sizes obtained from the Morphologi are generally higher
than those obtained from the wet dispersed LLS method, possibly due
to differences in the lower particle size resolution range of the two sys-
tems and different sample preparation approaches. From Fig. 2 it can be
seen that with increased milling extent the population of coarse parti-
cles gradually decreases while the population of fine particles increases,
with some batches having skewed distributions.
A graph, based on the method developed by Rohrs et al. [11], that
gives an estimate of the effect of d[0.9]/d[0.5] ratio and d[0.5] on content
uniformity for the relevant drug load and unit dose is shown in Fig. 3.
For b 5% drug loading and b 10 mg unit dose, application of the method,
which assumes uniform blending, indicates that provided that the d
[0.9]/d[0.5] ratio is not greater than 2, d[0.5] values up to 90 μm are ac-
ceptable to ensure acceptable content uniformity. Further decreasing
the dose would of course increase the chances of failing CU with these
particle size specifications. The actual obtained d[0.9]/d[0.5] values are
in the range of 2.5 to 4. This ratio is greater than the desired target for
a narrow size distribution, according to the method of Rohrs et al.
[11], however, all the batches are predicted to pass tablet content uni-
formity since the d[0.5] values are considerably smaller than 90 μm.
As expected by increasing the d[0.9]/d[0.5] ratio from 2 to 4 there is a
significant reduction in the predicted maximum acceptable d[0.9]
from 90 to 25 μm.
An analysis of the mill settings against the PSD and surface area
values showed that for the mill used the feed rate provides the best cor-
relation to the surface area of the milled product, with an R2 value of
0.94, compared to R2 of 0.77 for mill pressure. Lower feed rate gives
higher surface area, while the reverse is the case for mill pressure.

Fig. 3. Graphical estimation of effect of particle size distribution on content uniformity.

228 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233

Table 3
Summary of surface energy results, arranged in order of surface area.

Batch number Average B.E.T. SSA m2/g n = 2 Dispersive surface energy at 5% Free energy of absorption at Specific surface energy 5% Total surface energy 5%
surface coverage mJ/m2 5% surface coverage surface coverage mJ/m2 surface coverage mJ/m2
kJ/mol

Ethyl acetate Chloroform

Batch-B-unmilled 0.117 35.1 6.52 7.74 53.6 88.6

Batch-A Run 2 0.954 43.9 7.03 8.90 68.6 112.5
Batch-A Run 1 1.08 48.1 7.24 9.22 72.8 120.9
Batch-A Run 3 1.89 50.2 7.37 9.33 76.5 126.7
Batch-B Cross trial 1 2.34 48.6 7.39 9.20 75.2 123.7
Batch-B Cross trial 1 2.34 48.7 7.27 9.36 75.2 123.8
Batch-A Run 4 3.42 53.2 7.54 9.58 79.3 132.5
Batch-B Cross trial 3 3.98 48.8 7.25 9.29 74.7 123.6
Batch-B Cross trial 2 5.00 50.8 7.54 9.51 78.3 129.2

3.2. Surface area and surface energy
The surface area and surface energy of all the samples are shown in
Table 3. The surface area results are the mean of two measurements,
with % RSD generally 3% or less. Surface area provides a good description
of powder samples compared to d[0.9] alone, as the former takes into
account surface contributions from all the particles, both coarse and
fine, present in the sample.
For the surface energy measurements, one of the samples, Batch-B
Cross trial 1 highlighted in bold in Table 3, was analyzed in duplicate
using two separate samples to determine reproducibility. As the repro-
ducibility was observed to be acceptable, and with the analysis time for
the measurement up to three days per sample, replicate analysis for the
remaining samples was not conducted.
The reproducibility of the two samples analyzed from Batch-B Cross
trial 1 is within 1 mJ/m2 across the measured and calculated parameters.
Therefore for comparison between the batches any differences in the
dispersive surface energy values above 1 mJ/m2 can be said to be real.
Cross trial samples 1–3 were milled using Batch-B, while runs 1 to 4
were from Batch-A. When these sets of samples are grouped according
to batch history the dispersive surface energies at 5% coverage are ob-
served to increase with increasing surface area and milling extent. The
increase in specific surface energy of the batches with milling is concur-
rent with that of the dispersive surface energy (shown in Appendix A,
Figs. 1A & 2A, respectively). A combination of these two factors leads
to an overall increase in the total surface energy with milling extent,
as shown in Fig. 4. However, despite Batch-B Cross trial 2 having the
highest milling extent and surface area of all the batches, its dispersive
surface energy is lower than Batch-A Run 4 which has a lower milling
extent and surface area.
Comparison of the surface energy heterogeneity profiles of Batch-B
Cross trial 2 and Batch-A Run 4, shown in Fig. 5, indicates that Batch-A
Run 4 has a higher dispersive surface energy at closer to 0% surface
coverage, which represents the surface energy of the “highest” energy
sites. The reason for this observation is unclear and may be partly attrib-
uted to differences in batch history; unfortunately, a sample of the un-
milled API that was used to obtain Batch-A was not available for analysis
in order to confirm or refute this hypothesis.
It is of interest to understand the reason for the observed surface en-
ergy increase with milling extent. Milling extent clearly and expectedly
increases the surface area of the product. Comparison of the surface en-
ergy heterogeneity profiles of runs 1 to 4 regressed to ~0% surface cov-
erage (Fig. 6), i.e. similar to infinite dilution conditions where high
sensitivity to “higher energy” surfaces is expected, shows that not only
are the surface energy values at 0% greater than at 5%, but also that
the extent of increase from runs 1 to 4 across the heterogeneity profiles
are similar. Therefore the data suggests that milling has created higher
energetic sites. It is possible that the process of milling, during impart-
ment of mechanical energy, could lead to the creation of more energetic
surfaces; such mechanical energy may result in the creation of localized
lattice disorder on the surface of the particles in the sample. In fact this
phenomenon has been reported in literature. With increased milling
energy the extent of creation of localized lattice disorder might increase.
In such a case, “higher” energetic sites would be created thus leading
to higher measured surface energy values at ~0% surface coverage, es-
sentially infinite dilution region. This hypothesis was explored by pow-
der x-ray diffraction (PXRD) analysis of some of the milled batches to
determine the presence of peak broadening, which can be caused by lat-
tice disorder. It is important to note that PXRD analysis of the samples
was conducted months after SEA analyses. Aging time post-milling
has been observed to affect the surface energetics and flowability of
Metformin HCl due to re-ordering of mechanically induced surface de-
fects [33]. Nevertheless no significant differences in peak widths, nor
presence of amorphous material, were observed for the batches.
Dispersive surface energy mJ/m2

61
Batch-B Cross-trial 2
140 59
57
Total Surface Energy

130 Batch-A Run 4

55
5 % coverage

120
53
mJ/m2

110 Runs 1-4 51

49
100 Cross-trials
47
90 45
0 0.02 0.04 0.06 0.08 0.1
80 Surface coverage n/nm
0.00 1.00 2.00 3.00 4.00 5.00 6.00
B.E.T. SSA (m2/g) Fig. 5. Surface energy comparison for batches Batch-B Cross trial 2 and Batch-A Run 4. The
profiles of both samples are well described by an exponential fit with R2 values of greater
Fig. 4. Variation of total surface energy of milled batches as a function of surface area. than 0.99 in both cases.
 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233 229

65 Batch A run 1
Table 4
Surface energy (mJ/m2) Batch A run 2 Raw data from advancing contact angle measurements on stainless steel coupons.
60
Batch A run 3
Stainless steel coupon 1
55 Batch A run 4
Probe 1 2 3 4 Average Stdev
50 Ethylene glycol 47.0 48.6 44.8 48.9 47.3 1.88
Formamide 47.6 48.3 48.7 51.5 49.0 1.71
45 Diiodomethane 42.2 43.9 44.8 43.8 43.7 1.08
Dimethyl sulfoxide 45.1 42.5 45.2 44.7 44.4 1.27
40 2
R for Owens–Wendt [41] plot = 0.75
Calculated surface energy values: dispersive = 34.0 mJ/m2, specific = 5.1 mJ/m2
35
0 0.02 0.04 0.06 0.08 0.1 Stainless steel coupon 2
Surface coverage (n/nm)
Probe 1 2 3 4 Average Stdev

Fig. 6. Surface energy heterogeneity profiles of runs 1 to 4. Ethylene glycol 47.3 45.5 48.5 45.1 46.6 1.59
Formamide 54.5 56.1 55.2 52.4 54.6 1.58
Diiodomethane 45.6 43.6 42.6 42.8 43.7 1.37
Dimethyl sulfoxide 42.3 41.9 43.3 41.3 42.2 0.84
2
R for Owens–Wendt [41] plot = 0.75
Therefore the available data suggests that although milling creates
Calculated surface energy values: dispersive = 33.5 mJ/m2, specific = 4.4 mJ/m2
“higher energetic” sites (based on surface energy data) these cannot
be attributed to lattice disorder. The consequences of the formation of Stainless steel coupon 3
these higher energetic sites by milling are explored further in the fol- Probe 1 2 3 4 Average Stdev
lowing section. Ethylene glycol 50.0 50.1 48.3 42.4 47.7 3.63
Using the surface energy values reported here the work of cohesion Formamide 49.4 53.5 53.6 52.1 52.2 1.96
and adhesion (mJ/m2), were calculated using Eqs. (9) and (10), respec- Diiodomethane 44.5 45.1 43.0 42.1 43.7 1.37
tively. Dimethyl sulfoxide 42.2 46.7 41.6 43.3 43.5 2.28
R2 for Owens–Wendt [41] plot = 0.80
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Calculated surface energy values: dispersive = 35.2 mJ/m2, specific = 4.1 mJ/m2
WCohesion ¼ 2 γDA  γDA þ 2 γSP A  γA
SP
ð9Þ

each use therefore the comparisons made between the theoretical

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
WAdhesion ¼ 2 γDA  γDSS þ 2 γSP A  γSS
SP
ð10Þ
In Eqs. (9) and (10), the subscripts A and SS refer to API and stainless
steel, respectively, while the superscripts D and SP refer to dispersive
and specific surface energy values, respectively.
Work of adhesion towards stainless steel was estimated using
dispersive and specific surface energy values of 34.2 ± 0.9 mJ/m2
and 4.6 ± 0.5 mJ/m2, respectively, determined from advancing
contact angle measurements on stainless steel coupons described in
Section 2.8. The raw data for the contact angle measurements and the
R2 values of the Owens–Wendt plots are provided in Table 4. The results
generally show good reproducibility for four measurements on each
stainless coupon, and across the three coupons. The slightly higher stan-
dard deviation for SS coupon 3 and the generally low R2 values of 0.75–
0.80 for the measurements could be due to residual organic surface con-
tamination. Mantel and Wightman [43] observed that the measured
surface energy values, particularly the polar surface energy, of stainless
steel surfaces determined by contact angle measurements using organic
probe molecules are highly dependent on the pre-measurement
cleaning treatment ranging from water to O2 plasma cleaning. Measure-
ments obtained from surfaces cleaned with water and acetone gave the
work of adhesion to stainless steel from surface energy measurements
and the adhesion experiments on the FT4 are valid and adequate for
the purpose of this manuscript.
The variation of work of adhesion and cohesion with surface area is
shown in Fig. 7. While work of adhesion does not change significantly,
the work of cohesion shows a more marked increase with milling ex-
tent. The latter effect is contributed to by the fact that both the specific
and dispersive surface energies of the samples increase with milling
extent.
3.3. FT4 rheometer characterization
The first set of wall friction angle (WFA) values was obtained using a
stainless steel coupon of 1.20 μm surface roughness. The WFA°, which is
determined from the slope of the shear stress versus applied normal
stress as shown in Fig. 8 for Batch-A Run 3, measures the angle of wall
friction between the powder surface and the rotating disc used. Repro-
ducibility analysis of the test using Batch-A Run 3 showed values to be
reproducible to within ±0.3°. From the values shown in Table 5 it can
be observed that for the 1.2 μm surface roughness disc, hereafter
Work of cohesion and adhesion (mJ)

lowest values, while those that were O2 plasma cleaned gave the highest
300
values of both dispersive and polar surface energy.
The coupon cleaning treatment used in this study is sonication in ac-
250
etone (described in Section 2.8), therefore it is expected that there are
residual organic contaminants that affect the accuracy of the results. Work of cohesion
200
Nevertheless routine cleaning of tablet press tooling and equipments
Work of adhesion
used for manufacture is generally done using organic solvents, for ex-
150
ample IPA or acetone, sometimes with sonication for tablet press
toolings. Therefore while the dispersive and specific surface energy
100
values reported here may not be accurate for a contaminant-free stain-
less steel surface they represent a realistic state of stainless steel tooling/
50
equipment surfaces that the API would come into contact with during 0.00 1.00 2.00 3.00 4.00 5.00 6.00
routine drug product manufacture. Furthermore the stainless steel cou- Surface area (m2/g)
pons used for the wall friction angle measurements on the FT4 rheom-
eter, described in Section 2.7, were also cleaned with acetone prior to Fig. 7. Variation of work of adhesion and cohesion with surface area.
230 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233

Wall Friction Angle °(1.2SRC)

14.0
Measured shear stress, kPa
41
12.0 Runs 1 - 4
39
10.0 37
Cross-trials
Batch-A Run 3 35
8.0
1.20 micron 33
6.0 coupon run 1
31
4.0 29
Batch-A Run 3
2.0 1.20 micron 27
coupon run 2
0.0 25
0.0 5.0 10.0 15.0 20.0 80 90 100 110 120 130 140
Applied normal stress, kPa Total surface energy (mJ/m2)

Fig. 8. Slope of shear vs. applied normal stress. Reproducibility of WFA at 1.2 μm using Fig. 9. Correlation between WFA° and total surface energy at 5% coverage which shows a
batch-A Run 3. linear incremental relationship.

referred to as 1.2SRC, the adhesion against the stainless steel coupon
surface increases with milling extent. The differences in the numerical
values are small but greater than the ±0.3° expected variability based
on the reproducibility analysis.
Possible reasons for the increase in WFA with milling extent include
(1) greater surface energy of the powder particles with milling extent
causing them to be more “reactive” towards the steel surface and/or
(2) increased surface area with milling extent causing a larger contact
area for “binding”, particularly in the grooves on the surface of the cou-
pon. This greater contact area would have the effect of increasing fric-
tion between the powder particles and the metal surface. To
determine the contribution of increased “reactivity” of powder samples
towards stainless steel due to increased surface energy with milling, the
wall friction tests were conducted on fresh samples of selected batches
using a coupon of 0.05 μm surface roughness (0.05SRC). The 0.05SRC
has a much smoother surface finish, and therefore is expected to exhibit
less frictional adhesion to the powder particles. In this case the differ-
ences in the WFA of the batches are much less pronounced, and the
trend observed for 1.2 μm surface roughness no longer holds. This ob-
served surface roughness effect is in agreement with the work of Bunker
et al. [44] who found that surface roughness played a significant role in
the observed adhesion between lactose particles and punch surfaces;
for the same lactose batch punches with rougher surfaces exhibited
greater adhesion during compression. In fact the batches tested with
the 0.05SRC here exhibit similar WFA° under these conditions ~ 35 ±
0.6°. Therefore option 1 of the proposed mechanisms above can be
ruled out. If the increase in adhesion of the samples is due to increased
“reactivity” one would expect this effect to hold regardless of the surface
roughness of the coupon used. This corroborates the SEA analysis in the
previous section where it was deduced that for this material although
increased milling extent creates higher energetic sites it does not lead
to a proportional increase in the work of adhesion towards the stainless
steel coupons used.
Interestingly there is a linear relationship between the WFA° values
at 1.2SRC and the total surface energy values measured by SEA (Fig. 9).
The wall friction values increase with surface energy and this trend is
grouped according to batch history. The observed correlation between
1.2SRC and surface energy may be solely attributed to a surface area ef-
fect; the frictional adhesion increases with surface area due to increas-
ing contact area between powder bed and steel surface, while the
surface energy also increases with surface area as a consequence of mill-
ing extent for reasons already discussed.
Another outcome of the increased milling extent is a reduction in the
bulk density of the batches. The reduction in bulk density is proportional
to surface area increases with milling extent, irrespective of the batch his-
tory. With reducing particle mass it is expected that the impact of electro-
static forces will increase in competition with gravity forces. Particulate
materials with small particle size and low bulk density provide ideal con-
ditions for tribo-charging [45]; the electrification of particulate materials
in pharmaceutical systems which can arise due to sliding, rolling and im-
pact during processing and handling operations [46]. Such charging,
which has been shown to affect adhesion and agglomeration of particu-
late materials [47], can potentially have detrimental effects on product
quality, and result in loss of powder through deposition, segregation
within mixed formulations etc. [45]. Furthermore bulk density is one of
the key factors which affect the flowability of bulk solids [48].
3.4. Effect of milling propensity and powder properties on blending
In this section the effect of milling degree and resulting powder
properties on the blending unit operation is described. This portion of
the study was carried out about ~4 months after milling these batches
therefore the API batches would have undergone a significant resting
time. This reflects a realistic manufacturing scenario where milled
batches may undergo resting times of weeks, and sometimes months,
during transport and storage before further processing. Fig. 10 illus-
trates blending profiles and the corresponding LnRSDE for some selected
blends. These profiles include blends manufactured with API of variable
size (d[0.9] PSD of 83.49 μm, 33.6 μm and 5.97 μm). Selecting these sizes
of particles provides the ability to evaluate the mixing behavior of large,

Table 5
FT4 measured data.

Batch number B.E.T. surface area (m2/g) Bulk density g/ml Wall friction angle (°) using a SS coupon of
(average of n = 2) (average of n = 2) estimated 1.2 μm surface roughness

Batch-A Run 2 0.95 0.462 31.6

Batch-A Run 1 1.08 0.438 32.3
Batch-A Run 3 1.89 0.367 35.5
Batch-A Run 4 3.42 0.281 37.8
Batch-B (unmilled) 0.117 0.546 26.0
Batch-B Cross trial 1 2.34 0.293 36.4
Batch-B Cross trial 2 5.00 0.240 38.8
Batch-B Cross trial 3 3.98 0.273 37.7

150
100
API Potency in blend
* * * * * D90 5.97
50
0 -0.025
-50
-100
-150 0.5
0 50 100 150 200 250 300 350 400
Number of rotations
Fig. 10. Online NIR blending profiles of pre-blends.
medium and small sized API in a low drug load formulation. The profiles
with large API particles (d[0.9] values of 83.49 μm and 33.6 μm) exhibit
relatively faster kinetics to reach blend uniformity. For both profiles the
macro mixing (convection type) is completed by ~100 revolutions and
diffusion type mixing is completed by ~175 revolutions. In comparison
to this, the blend manufactured with API with a d[0.9] of 5.97 μm dem-
onstrated completion of macro mixing within ~75 revolutions but diffu-
sion type of mixing is still not achieved after 400 revolutions. It seems
this slow but steady micro mixing (i.e. diffusion type) is happening
throughout the blending time. A similar behavior can easily be visual-
ized with corresponding LnRSDE plots. The slopes of the LnRSDE vs. rev-
olutions are clearly different for the three batches as shown in the
graph. The largest and smallest slopes correspond to the largest particle
size (d[0.9] values of 83.49 μm) and smallest particle size (d[0.9] values
of 5.97 μm) blends respectively.
Notably, the blending profile of 5.97 μm blend has high intensity API
peaks as denoted by asterisks. The presence of a significant number of
“API concentration spikes” is the result of areas of concentrated API
passing the window and being sampled during the measurement. The
spikes occur either because of the presence of agglomerates or aggre-
gates. When an agglomerate appears it carries more weight in contrib-
uting to the spectrum and results in intense peaks from time to time
during the blending profile. In general it is not surprising to observe
API concentration spikes at the beginning of blending profiles, however,
the continuous presence of such concentration spikes throughout the
blending profile is indicative of presence of aggregates/agglomerates
that are difficult to break with the shear applied during tumble blend-
ing. The observation of larger particles exhibiting faster blending kinet-
ics than smaller particles of the same chemical entity is in corroboration
with that of Bellamy et al. [49].
The theoretical work of cohesion from surface energy values,
displayed in Fig. 7, shows a general increase with surface area. However
a closer look at the data shows that the samples with d[0.9] of 33 μm and
5.97 μm have similar theoretical cohesion values despite the consider-
able differences in surface area, 1.89 m2/g and 5 m2/g, respectively.
The work of cohesion data is contrary to the clear differences in the
blending kinetics of these two batches that have been elucidated in
this current section.
Collectively the data presented in this study indicates that surface
energetics may not be the main factors governing the blending of partic-
ulate materials evaluated here. The surface area of the d[0.9] = 5.97 μm
batch is over 2.5 times that of the d[0.9] = 33.6 μm batch, and the bulk
density of the former is 50% lower. These differences in surface area
appear to have a greater impact on the cohesivity and “blendability”
observed here. Fig. 11 shows a linear correlation between the surface
area of the milled batches and rate constant obtained from the LnRSDE
D. Olusanmi et al. / Powder Technology 258 (2014) 222–233 231
0.000
D90 83.49 R² = 0.87
(blending kinetics)
Slope of LnRSDE
-0.005
D90 33.6 -0.010
-0.015
4.5 -0.020
D90 5.97 LnRSDE
3.5 D90 33.6 LnRSDE
-0.030
LnRSDE
D90 83.49 LnRSDE
0.0 1.0 2.0 3.0 4.0 5.0 6.0
2.5 B.E.T. Surface area (m2/g)
1.5 Fig. 11. Blending rate as a function of surface area.
data obtained using the procedure described in Fig. 1. As shown in this
study the API batch smallest particle size distribution also has the
highest surface area. From the PSD data shown in Table 2, large particles
were not present in batch B cross trial 2 immediately post-milling. For-
mation of agglomerates, particularly of small particles, can potentially
occur during periods of storage and shipping, and exacerbated in the
presence of compaction forces. During blending it is expected that the
macroscopic bulk lumps break up during convective mixing, but as the
“chunks” break up into smaller elements, it is possible that the areas
of the intrinsic powder that have higher cohesivity will stay intact for
longer. Increased electrostatic effect that is expected for lower bulk den-
sity materials (not measured in this study) may also be a contributing
factor. The concentration spikes observed during blending are likely
due to the inability of the process to quickly and completely break up
cohesive agglomerates within the blend as a consequence of the small
particle size and associated cohesivity of this material. As shown in
this example the presence of such small API particles clearly results in
a slower blending profile, even accounting for the presence of the API
spikes. It is expected that small particles will take longer to blend than
the same mass of large particles because numerically it is difficult to dis-
perse large number of smaller particles compared to larger particles. In
addition when small particles are present as agglomerates it takes addi-
tional shear energy to break them down and as a result these API attri-
butes can lead to incomplete blend with reduced blend homogeneity.
Fig. 11, which shows that blending time to reach uniformity increases
with surface area of the API, clearly demonstrates the influence of pow-
der properties on the manufacturing process. The appropriate
micronization control strategy is one that gives optimal bioavailability
enhancement without detrimental effects to downstream processing
efficiency.
4. Conclusions
The effect of the degree of milling on the surface properties of an API
has been demonstrated. Furthermore the consequence of such milling
induced changes has been evaluated in a unit operation that is com-
monly utilized during pharmaceutical drug manufacture.
Increased milling extent causes changes to the surface as well as
bulk properties of this material, and this has a potential, if not mitigated,
to lead to a direct negative impact on handling and downstream
processing.
In a QbD manufacturing paradigm the effect of bulk properties on all
aspects of the manufacturing process needs to be understood. Here it is
shown that milling parameters produce materials with different bulk
and surface properties, and that these parameters can influence a key
unit process, blending during drug product manufacture. Such work,
carried out on a small scale using appropriate tools, can provide a
guide for setting lower limit of particle size specifications. Specifications
on the upper limit of particle size is obviously important for
biopharmaceutics reasons, while lower limit specifications would en-
sure that powder handling and downstream processing are not made
more challenging than necessary.
232 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233

A key learning point from this study is that when milling is necessary
it is important to conduct it appropriately in order to utilize the benefi-
cial effects for biopharmaceutics requirements while minimizing the
deleterious ones. Milling, one unit operation in a series of drug product
processing steps, beyond what is required can lead to additional chal-
lenges during bulk handling and even downstream processing.
Acknowledgments
The authors would like to thank Dr Denette Murphy (Bristol–Myers
Squibb U.S.A.) for PXRD analyses, and Drs Jatin Patel and Omar Sprockel
(both Bristol–Myers Squibb U.S.A.) for their review and feedback on the
publication.
Appendix A
85
Specific Surface Energy
[10] B.Y. Shekunov, P. York, Crystallisation processes in pharmaceutical technology and
drug delivery design, J. Cryst. Growth 211 (2000) 122–136.
[11] B.R. Rohrs, G.E. Amidon, R.H. Meury, P.J. Secrest, H.M. King, C.J. Skoug, Particle size
limits to meet USP content uniformity criteria for tablets and capsules, J. Pharm.
Sci. 95 (2006) 1049–1059.
[12] International conference on harmonization of technical requirements for registra-
tion of pharmaceuticals for human use, Q8 (R1), Pharmaceutical development revi-
sion 1.
[13] R.W. Bondi Jr., J.K. Drennen III, Quality by design and the importance of PAT in QbD,
Sep. Sci. Technol. 10 (2001) 195–224.
[14] D. Olusanmi, K. Roberts, M. Ghadiri, Y. Ding, The breakage behaviour of Aspirin
under quasi-static and single particle impact loading: effect of crystallographic an-
isotropy, Int. J. Pharm. 411 (2011) 49–63.
[15] J.Y.Y. Heng, A. Bismarck, A. Lee, K. Wilson, D.R. Williams, Anisotropic surface chem-
istry of aspirin crystals, J. Pharm. Sci. 96 (2007) 2134–2144.
[16] F. Thielmann, Introduction into characterisation of porous materials by inverse gas
chromatography, J. Chromatogr. A 1037 (2004) 115–123.
[17] P.P. Ylä-Mäihäniemi, J.Y.Y. Heng, F. Thielmann, D.R. Williams, Inverse gas chromato-
graphic method for measuring the dispersive surface energy distribution for partic-
ulates, Langmuir 24 (2008) 9551–9557.
[18] J.Y.Y. Heng, F. Thielmann, D.R. Williams, The effects of milling on the surface proper-
ties of form I paracetamol crystals, Pharm. Res. 23 (2006) 1918–1927.
[19] P. York, M.D. Ticehurst, J.C. Osborn, R.J. Roberts, R.C. Rowe, Characterisation of the

surface energetics of milled dl-propranolol hydrochloride using inverse gas chroma-

5 % coverage mJ/m2

80 tography and molecular modeling, Int. J. Pharm. 174 (1998) 179–186.

[20] H.E. Newell, G. Buckton, D.A. Butler, F. Thielmann, D.R. Williams, The use of inverse
75
phase gas chromatrography to measure the surface energy of crystalline, amor-
70 phous, and recently milled lactose, Pharm. Res. 18 (2001) 662–666.
Runs 1-4 [21] M. Ohta, G. Buckton, Determination of the changes in surface energetics of
65 cefditoren pivoxil as a consequence of processing induced disorder and equilibra-
Cross-trials tion to different relative humidity levels, Int. J. Pharm. 269 (2004) 81–88.
60
[22] F. Thielmann, D.J. Burnett, J.Y.Y. Heng, Determination of the surface energy distribu-
55 tions of different processed lactose, Drug Dev. Ind. Pharm. 33 (2007) 1240–1253.
[23] R. Ho, D. Wilson, J.Y.Y. Heng, Crystal habits and the variation in surface energy het-
50 erogeneity, Cryst. Growth Des. 9 (2009) 4907–4911.
0 1 2 3 4 5 6 [24] J.Y.Y. Heng, A. Bismarck, D.R. Williams, Anisotropic surface chemistry of crystalline
B.E.T. SSA (m2/g) pharmaceutical solids, AAPS Pharm. Sci. Technol. 7 (2006) E1–E9.
[25] G. Buckton, Assessment of the wettability of pharmaceutical powders, J. Adhes. Sci.
Technol. 7 (1993) 205–219.
Fig. 1A. Variation of specific surface energy at 5% coverage with surface area.
[26] G. Buckton, J.M. Newton, Assessment of the wettability of powders by use of com-
pressed powder discs, Powder Technol. 46 (1986) 201–208.
55
Dispersive Surface Energy

[27] J.Y.Y. Heng, A. Bismarck, A.F. Lee, K. Wilson, D.R. Williams, Anisotropic surface ener-
5 % coverage mJ/m2

getic and wettability of macroscopic form I paracetamol crystals, Langmuir 22

50
45
Runs 1-4
40
Cross-trials
35
30
0 1 2 3 4 5
B.E.T. SSA (m2/g)
Fig. 2A. Variation of dispersive surface energy at 5% coverage with surface area.
References
[1] B. Griffin, Advances in lipid-based formulations: overcoming the challenges of low
bioavailability for poorly water soluble drug compounds, Am. Pharm. Rev. (March
2012) 41–47.
[2] C. Giliyar, D.T. Fikstad, S. Tyavanagimatt, Challenges and opportunities in oral deliv-
ery of poorly water-soluble drugs, Drug Deliv. Technol. 6 (2006) 57–63.
[3] M.B. Maurin, D.J.W. Grant, P.H. Stahl, The physicochemical background: fundamen-
tals of ionic equilibria, in: P.H. Stahl, C.G. Wermuth (Eds.), Handbook of Pharmaceu-
tical Salts: Properties, Selection and Use, Wiley-VCH, New York, 2002, p. 12.
[4] J.M. Merritt, S.K. Viswanath, G.A. Stephenson, Implementing quality by design in
pharmaceutical salt selection: a modeling approach to understanding dispropor-
tionation, Pharm. Res. 30 (2013) 203–217.
[5] G.A. Stephenson, A. Aburub, T.A. Woods, Physical stability of salts of weak bases in
the solid-state, J. Pharm. Sci. 100 (2011) 1607–1617.
[6] R.D. Braatz, Advanced control of crystallization processes, Annu. Rev. Control. 26
(2002) 87–89.
[7] N. Rodriguez-Hornedo, D. Murphy, Significance of controlling crystallization mech-
anisms and kinetics in pharmaceutical systems, J. Pharm. Sci. 88 (1999) 651–660.
[8] S. Desikan, S.R. Anderson, P.A. Meenan, P.H. Toma, Crystallisation challenges in drug
development: scale-up from laboratory to pilot plant and beyond, Curr. Opin. Drug
Discov. Dev. 3 (2000) 723–733.
[9] J.C. Feeley, P. York, B.S. Sumby, H. Dicks, Determination of surface properties and
flow characteristics of salbutamol sulphate, before and after micronization, Int. J.
Pharm. 172 (1998) 89–96.
(2006) 2760–2769.
[28] C. Levoguer, D. Butler, F. Thielmann, D. Williams, Characterising the physico-
chemical properties of pharmaceutical materials, Pharm. Technol. Eur. 12 (2000)
36–42.
[29] J. Domingue, D. Burnett, F. Thielmann, Using inverse gas chromatography (IGC) to
investigate process-related changes in surface and bulk properties of pharmaceuti-
cal materials, Am. Lab. 35 (2003) 32–37.
[30] A. Otte, T. Carvajal, Assessment of milling-induced disorder of two pharmaceutical
compounds, J. Pharm. Sci. 100 (2011) 1793–1804.
[31] J.F. Gamble, M. Leane, D. Olusanmi, M. Tobyn, E. Supuk, J. Khoo, M. Naderi, Surface
energy analysis as a tool to probe the surface energy characteristics of micronized
6 materials — a comparison with inverse gas chromatography, Int. J. Pharm. 422
(2012) 238–244.
[32] P. Begat, D.A.V. Morton, J.N. Staniforth, R. Price, The cohesive–adhesive balances in
dry powder inhaler formulations I: direct quantification by atomic force microscopy,
Pharm. Res. 21 (2004) 1591–1597.
[33] R.R. Vippagunta, R. LoBrutto, C. Pan, J.P. Lakshman, Investigation of metformin HCl
lot-to-lot variation on flowability differences exhibited during drug product pro-
cessing, J. Pharm. Sci. 99 (2010) 5030–5039.
[34] J.F. Gamble, R.N. Davé, S. Kiang, M.M. Leane, M. Tobyn, S.S.Y. Wang, Investigating the
applicability of inverse gas chromatography to binary powdered systems: an appli-
cation of surface heterogeneity profiles to understanding preferential probe-surface
interactions, Int. J. Pharm. 445 (2013) 39–46.
[35] G.M. Dorris, D.G. Gray, Adsorption of n-alkanes at zero surface coverage on cellulose
paper and wood fibres, J. Colloid Interface Sci. 77 (1980) 353–362.
[36] A. Voekel, K. Batko, K. Adamska, B. Strezemiecka, Determination of Hansen solubility
parameters by means of gas–solid inverse gas chromatography, Adsorpt. Sci.
Technol. 26 (2008) 93–102.
[37] J. Schultz, L. Lavielle, C. Martin, The role of the interface in carbon fibre-epoxy com-
posites, J. Adhes. 23 (1987) 45–60.
[38] S. Dong, M. Brendlé, J.B. Donnet, Study of solid surface polarity by inverse gas chro-
matography at infinite dilution, Chromatographia 28 (1989) 469–472.
[39] C.J. van Oss, Interfacial Forces in Aqueous Media, Marcel Dekker, New York, 1994.
[40] R.E. Freeman, J.R. Cooke, L.C.R. Schneider, Measuring shear properties and normal
stresses generated within a rotational shear cell for consolidated and non-
consolidated powders, Powder Technol. 190 (2009) 65–69.
[41] D.K. Owens, R.C. Wendt, Estimation of the surface free energy of polymers, J. Appl.
Polym. Sci. 13 (1969) 1741–1747.
[42] W. Shi, D. Hajedemos, O. Sprockel, An energetic approach of powder mixing by ma-
nipulating cohesive interaction in a magnetic field, Powder Technol. 235 (2013)
400–404.
[43] M. Mantel, J.P. Wightman, Influence of the surface chemistry on the wettability of
stainless steel, Surf. Interface Anal. 21 (1994) 595–605.
 D. Olusanmi et al. / Powder Technology 258 (2014) 222–233 233

[44] M. Bunker, J. Zhang, R. Blanchard, C.J. Roberts, Characterising the surface adhesive
behaviour of tablet tooling components by atomic force microscopy, Drug Dev.
Ind. Pharm. 37 (2011) 875–885.
[45] E. Supuk, A. Zarrebini, J.P. Reddy, H. Hughes, M.M. Leane, M.J. Tobyn, P. Timmins, M.
Ghadiri, Tribo-electrification of active pharmaceutical ingredients and excipients,
Powder Technol. 217 (2012) 427–434.
[46] G. Rowley, Quantifying electrostatic interactions in pharmaceutical solid systems,
Int. J. Pharm. 227 (2001) 47–55.
[47] K.R. LaMarche, F.J. Muzzio, T. Shinbrot, B.J. Glasser, Granular flow and
dielectrophoresis: the effect of electrostatic forces on adhesion and flow of dielectric
granular materials, Powder Technol. 199 (2010) 180–188.
[48] A.J. Tamara, C. Yap, M.A. Mannan, Gravity pipe flow of polymeric bulk solids in
pneumatic conveying system, Chem. Eng. Sci. 61 (2006) 7836–7849.
[49] L.J. Bellamy, A. Nordon, D. Littlejohn, Effects of particle size and cohesive properties
on mixing studied by non-contact NIR, Int. J. Pharm. 361 (2008) 87–91.