Powder Technology 274 (2015) 319–323
`
`Contents lists available at ScienceDirect
`
`Powder Technology
`
`j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / p o w t e c
`
`Macro- and micro-mixing of a cohesive pharmaceutical powder
`during scale up
`Weixian Shi a,⁎, Elizabeth Galella b, Omar Sprockel a
`a Drug Product Science and Technology, Bristol-Myers Squibb Company, 1 Squibb Drive, New Brunswick, NJ 08903, United States
`b Analytical & Bioanalytical Development, Bristol-Myers Squibb Company, 1 Squibb Drive, New Brunswick, NJ 08903, United States
`
`a r t i c l e
`
`i n f o
`
`a b s t r a c t
`
`Article history:
`Received 22 August 2014
`Received in revised form 15 December 2014
`Accepted 21 January 2015
`Available online 26 January 2015
`
`Keywords:
`Macro-mixing
`Micro-mixing
`Shear
`Cohesive
`
`Efficient powder mixing involves a macro- and micro-mixing mechanism, achieved by a combination of various
`types of mixers. Selection of the mixer is based on the understanding of the cohesiveness of the components in
`the mixture. In the current study, a cohesive active pharmaceutical ingredient (API), X, was used as the model
`compound to study the effectiveness of convective mixing in a bin blender and intensive shear mixing with a
`comil (conical mill). Convective mixing in the bin blender only delivered limited macro-mixing for API X and
`the resulting blend was heterogeneous at both micro- and macro-scales. After blending in the bin blender, the
`comilling process added micro-level mixing by introducing locally intensive mechanical shear. The resulting
`blend showed improved homogeneity at the micro-scale, but was still heterogeneous at the macro-scale. An
`additional mixing step in the bin blender after comilling was required to ensure the uniformity of the mixture
`at both micro-and macro-scales. The significance of the second convective mixing to micro-mixing was
`underscored at commercial-scale manufacture as compared to the development scale. Despite the scale
`dependency on the comilling step, the extensive shear exerted during the comilling step facilitated further
`micro-mixing by the convective mixing in the second bin mixing step. The investigation demonstrates that a
`rational selection of mixing steps with various types of mixers is crucial to achieve both macro- and micro-mixing
`of cohesive materials from development to commercial scales.
`
`© 2015 Elsevier B.V. All rights reserved.
`
`1. Introduction
`
`Mixing of powders is a common yet important process that is used to
`achieve uniform mixtures in food, chemical, and pharmaceutical indus-
`tries. It is especially critical in pharmaceuticals as accurate dosing is
`vital to efficacy and safety in patients. The criticality of homogeneity is
`evidenced by the strict requirements for dose uniformity from regulatory
`agencies around the world. As a result, mixing of pharmaceutical pow-
`ders, typically cohesive by nature, is extensively studied in various
`types of mixing equipment to understand mixing mechanism via
`convection, diffusion or shearing [1–4]. While these studies have accu-
`mulated mechanistic understanding of the particular types of mixing
`equipment, few investigations have studied the effect of combining
`different types of mixers to blend cohesive materials. Homogeneity of
`pharmaceutical powder normally requires multiple mixing mechanisms
`to be involved [5], which is difficult to achieve with one type of mixer.
`
`⁎ Corresponding author. Tel.: +1 732 227 6736; fax: +1 732 227 3818.
`E-mail addresses: weixian.shi@bms.com (W. Shi), elizabeth.galella@bms.com
`(E. Galella), omar.sprockel@bms.com (O. Sprockel).
`
`http://dx.doi.org/10.1016/j.powtec.2015.01.049
`0032-5910/© 2015 Elsevier B.V. All rights reserved.
`
`Our study demonstrates that the combination of different types of
`mixers is a practical and effective means to achieve uniform distribution
`of cohesive materials, such as APIs. The key to the uniform distribution is
`to enable both macro- and micro-mixing to reach uniformity at both
`scales. It is achieved by engaging different mixing mechanisms in the
`process, such as convective and shear mixing. Both convective and
`shear mixing can deliver macro- and micro-mixing depending on the
`cohesiveness of the material. Macro-mixing occurs at the bulk level
`via dispersion, and is fast, while micro-mixing occurs at the particle
`level via shearing or diffusion. Although convective mixing in a bin
`blender is an efficient way to achieve macro-scale uniformity, our
`study suggests that macro-scale uniformity in a bin blender is not
`achievable for API X without substantially engaging micro-scale mixing
`via the intensive shear mechanism in a comil first. Although comilling is
`typically used as a size reduction method [6,7], we emphasized more on
`the shear mixing function of comilling in this study. Recent studies have
`shown that comilling is an effective way of distributing minor
`ingredients onto various pharmaceutical powders [8,9]. Only after the
`comilling step delivers some degree of micro-uniformity can mixing in
`a bin blender further convey both micro- and macro-uniformity and
`result in uniform distribution of API X across the powder bed. Such
`
`Cosmo Ex 2023-p. 1
`Mylan v Cosmo
`IPR2017-01035
`
`

`

`320
`
`W. Shi et al. / Powder Technology 274 (2015) 319–323
`
`mechanism was only revealed at commercial-scale manufacture but not
`apparently at the development scale.
`
`3
`
`2. Material and methods
`
`A proprietary formulation containing Microcrystalline Cellulose,
`Lactose Anhydrous, Crospovidone, silicon dioxide, magnesium streareate
`and API X was used in the study.
`API X was micronized to less than 30 μm (100% by a light scattering
`method). The API has a true density of 1.3 g/ml and is needle-like in
`shape. API X was loaded at concentrations for 1.25%, 2.5%, or 5% w/w.
`The batch size in the study ranged from 20 kg to 750 kg. All materials
`were loaded in a bin blender, mixed at 12 rpm for 108 revolutions
`(blend 1) and passed through a comil. The mixture passing through
`the comil (blend 2) was received in a second bin blender and then fur-
`ther mixed at 12 rpm for 120 revolutions (blend 3). Samples were taken
`at the top or bottom of the powder bed after each mixing step for the
`20-kg scale batches in a 68-L bin blender. The schematic process dia-
`gram and sampling locations are shown in Fig. 1. Due to the large
`batch size, additional sample locations were added for a 300-kg batch
`in a 900-L bin blender and a 750-kg batch in a 2000-L bin blender as
`shown in Fig. 2. Samples were tested for uniformity and concentration
`by High-performance liquid chromatography (HPLC). While a Quadro®
`U10 comil with a milling chamber diameter of 127 mm was used for the
`20-kg batch, a Quadro® 196S comil with a milling chamber diameter of
`305 mm was used for the 300-kg and 750-kg batches.
`The following metrics were used to characterize the homogeneity of
`the blends:
`
`1) The maximum difference (MD) in the average potency of API X
`among samples taken from multiple locations of the powder bed
`from the same process step, which is shown in Eq. (1)
`
`
`
`
`MD ¼ Max Xi
`−Min Xi
`
`ð1Þ
`
`where Xi is the average potency at a specific locationi . If samples are
`taken only from the top and bottom of the bin blender (68 L),
`

`

`MD ¼ XT−XB
`
`ð2Þ
`
`where XT and XB stand for the average potency of the sample from
`the top and bottom of the bin blender, respectively. MD indicates
`the macro-mixing behavior and is expected to be close to zero for
`effective macro-mixing.
`
`where k is the number of sampling locations, N is the total number of
`potency measurements across all samples from a particular mixing
`step (N = k ∗ n), and Xm is the observed mean of all these individual
`measurements in the mixing step. RSDm represents the overall mixing
`behavior of a specific mixing step. A powder mixture that is well
`mixed at the micro- and macro-scales yields a low RSDm value while
`
`Top Sample
`
`1st bin
`Blend 1
`
`Bottom Sample
`
`Bottom Sample
`
`Comil
`
`2nd bin
`Blend 2
`
`Top Sample
`
`Top Sample
`
`2nd bin
`Blend 3
`
`Bottom Sample
`
`Fig. 1. Mixing process and sampling locations.
`
`2
`
`1
`
`4
`
`Fig. 2. Sampling locations in the 900-L or 2000-L bin.
`
`2) The relative standard deviation of potency measurements within a
`sample, or RSDS, as is shown in Eq. (3),
`ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
`
`
`Xn
`Xi j−Xi
`j¼1
`n−1
`Xi
`
`vuuuuuut
`
`RSDS ¼
`
`2
`
`ð3Þ
`
`where n is the number of potency measurements within a sample, Xij
`is an individual observation of potency at a specific location i. RSDS is
`an indicator of micro-mixing behavior at a specific location with a
`low RSDS suggesting a uniform distribution of API X in the vicinity of
`that sampling location.
`3) The relative standard deviation for potency measurements from all
`samples within a particular mixing step, or RSDm, which is shown in
`Eq. (4),
`
`ð4Þ
`
`ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
`
`
`k Xn
`Xi j−Xm
`j¼1
`¼1
`N−1
`Xm
`
`2
`
`X i
`
`vuuuut
`
`RSDm ¼
`
`Cosmo Ex 2023-p. 2
`Mylan v Cosmo
`IPR2017-01035
`
`

`

`W. Shi et al. / Powder Technology 274 (2015) 319–323
`
`321
`
`heterogeneity at either micro- or macro-scale results in a high RSDm
`value.
`
`3. Results and discussion
`
`Fig. 3 presents the MDs of three 20-kg batches in a 68-L bin blender at
`each step of mixing with 1.25%, 2.5% and 5% w/w of API X, respectively.
`The MD between the two samples after the first bin mixing step was
`substantial (24.3%–83.5%). However, it decreased dramatically after the
`comilling step (9.5%–17.1%) and was further narrowed after the second
`bin mixing step (3.2%–6.2%). Such drops in MDs clearly suggested that
`regardless of the target concentration of API X in the batch, macro-mixing
`in the first bin mixing step was ineffective, and macro-uniformity still
`relied on subsequent steps, i.e., comilling and the second bin mixing step.
`The drop in MD is accompanied with the improving micro-uniformi-
`ty through the mixing process that is evidenced by the declining RSDS as
`shown in Fig. 4. Blend 1 was heterogeneous at the micro-scale with RSDS
`between 18.9% and 39.3% across the three batches, in line with the
`macro-heterogeneity shown in MDs. In contrast, blends 2 and 3 were
`homogeneous at the micro-scale (RSDS of b4.7%), regardless of sampling
`location and target concentration of API X. This suggests that uniformity
`at the micro-scale was attained via the extensive local shear exerted
`during the comilling step. There was no improvement in micro-scale
`homogeneity between blends 2 and 3, suggesting that maximum
`micro-mixing was achieved at the comilling step and the second blend-
`ing step did not extend micro-mixing further.
`Therefore, regardless of the drug load, blend 1 is heterogeneous across
`the powder mixture, blend 2 is micro-scale homogenous but macro-scale
`heterogeneous, and blend 3 is homogenous across the powder mixture. In
`terms of function of the mixing steps, the first mixing step in the bin
`blender delivers coarse macro-mixing only with no micro-mixing, the
`comilling step delivers micro-mixing with limited macro-mixing, and
`the second mixing step in the bin blender is a macro-mixing step.
`The increase in homogeneity at both macro- and micro-scales across
`the mixing steps also was evidenced by declining RSDm as illustrated in
`Fig. 5. The RSDm started at as high as 43.9% for blend 1 and ended at less
`than 3.5% for blend 3. The trend is similar across the three concentra-
`tions in the study.
`Upon scale-up at 300 kg or 750 kg, the mixing of the API X per-
`formed differently from the 20-kg scale. Fig. 6 shows the maximum dif-
`ference in potency across the mixing steps at 300 kg and 750 kg with
`the 5% w/w drug load and the 20-kg scale batch at 5% w/w is plotted
`in the same graph for comparison. Blend 1 had a smaller MD at
`commercial scales (b20%), suggesting improved yet still ineffective
`macro-mixing upon scale up. This distinction between scales on blend
`1 indicates that macro-mixing in the first blending step is more effective
`at larger scales due to increased shear provided upon scale up [10].
`
`Fig. 4. Within-sample uniformity of blend through the process (20 kg).
`
`The MD dropped slightly from blend 1 to blend 2 for both 300- and
`750-kg batches, unlike the larger decrease observed at the 20-kg
`batches. This distinction between scales suggests that comilling at the
`larger scale had much less impact on macro-mixing due to more effec-
`tive macro-mixing that occurred in the preceding bin mixing step. On
`the other hand, MD for blend 3 dropped significantly to 1.8% (300 kg)
`and 1.4% (750 kg), similar to that from the 20-kg batches. This is an
`indicator that macro-mixing in the second bin blender at the develop-
`ment scale was as sufficient as that at the commercial scale.
`The RSDS of the 300-kg and 700-kg batches from each step of mixing
`are plotted in Figs. 7 and 8, respectively. Although there is a decrease in
`RSDS from blend 1 to blend 2 at each sampling location in the powder
`bed, the degree of micro-scale mixing in the comilling step at large
`scales is not as effective as that observed at the 20-kg scale. This is
`demonstrated by the wide range of RSDs observed for blend 2 after
`comilling, i.e., 2.7%–10.5% for the 300-kg scale and 4.8%–15.1% for the
`750-kg scale. The corresponding range on the three 20-kg scale batches
`is much narrower, 0.8%–4.7%. The noticeable impact from scale up in
`micro-scale mixing delivered by comilling is likely due to the change
`in the size of the comil causing a difference in the residence time and
`working volume of the powder in the chamber.
`Despite the less efficient micro-scale mixing with the comil, the local
`shear exerted through comilling still disrupts the cohesive bonds
`between API X particles, facilitating the micro-scale mixing in the
`second blender. RSDs of blend 3 for the 300-kg and 750-kg batches
`decrease to less than 1.9% and 1.7%, respectively, implying that sufficient
`micro-scale uniformity was achieved. These values are similar to that of
`the 20-kg batches. As there was no discernable decrease in RSDs for the
`20-kg batches from blend 2 to blend 3, such decrease in RSDs upon scale
`up strongly indicates that the second mixing in the bin blender plays
`a crucial role in scale up, i.e., continuing the micro-mixing that was
`initiated by the comilling step.
`
`Fig. 3. Decrease in maximum difference (MD) of mean potency throughout the process
`(20 kg).
`
`Fig. 5. Overall uniformity of blend through the process (20 kg).
`
`Cosmo Ex 2023-p. 3
`Mylan v Cosmo
`IPR2017-01035
`
`

`

`W. Shi et al. / Powder Technology 274 (2015) 319–323
`
`750 kg
`300 kg
`20 kg
`
`Blend 1
`
`Blend 2
`
`Blend 3
`
`Fig. 8. Within-sample uniformity of blend through the process (750 kg).
`
`322
`
`100.0
`
`80.0
`
`60.0
`
`40.0
`
`20.0
`
`0.0
`
`Maximum Difference Or MD (%)
`
`Fig. 6. Decrease in maximum difference (MD) of mean potency (5% API).
`
`It is clear that the second mixing step in the bin blender involves
`both micro-scale mixing and macro-scale mixing. The effect of this
`mixing step on micro-scale mixing is not obvious at a smaller scale,
`because the preceding comilling step is sufficient to achieve the required
`micro-scale mixing for uniformity and, therefore, any further micro-scale
`mixing is negligible. When the comilling step delivers insufficient micro-
`mixing upon scale up, the micro-mixing function of the second bin mixing
`step becomes apparent.
`The detailed mechanism of mixing upon scale up is not revealed
`through RSDm as it lumps effect from both macro- and micro-scale
`mixing mechanisms in one index. The decrease in RSDm could have
`two drastically different starting points from blend 2, i.e., blend 2 is
`not uniform at one scale but is uniform at the other scale, or it is not uni-
`form at either micro- or macro-scales, as suggested by RSDs. Neverthe-
`less, RSDm is a direct indicator of the overall homogeneity as shown in
`Fig. 9. The slight difference between the two commercial scale batches
`is likely due to sampling, while more heterogeneity of blend 1 at the
`development scale reflects less effective shear mixing in the first blending
`step compared to that at the commercial scale.
`Additionally, the effective macro- and micro-mixing achieved at the
`second mixing step is likely associated with the coating effect at the par-
`ticle level exerted from the proceeding coming step [8,9]. SEM images of
`pure API and blends, as shown in Fig. 10, demonstrate that the needle-
`like API covers the surface of the excipients, allowing the excipients
`functioning as API carrier. Since the mixing of API-coated excipients is
`driven by the excipients and is less dependent on the cohesiveness of
`API, the uniformity of API in the second bin blending step is more readily
`achieved than that in the first bin blending step.
`
`In summary, the second mixing step in the bin blender introduces
`additional micro-scale mixing beyond its macro-mixing role due to
`the coating effect from the comilling step. This hidden role in micro-
`scale mixing is crucial in the uniformity of API X at commercial scales.
`Fig. 11 summarizes the mixing mechanisms involved in the mixing
`process for API X.
`
`4. Conclusion
`
`The current investigation revealed that the mechanism of mixing
`cohesive materials in a bin blender changes with bin size. With smaller
`bins, convective mixing occurs only at the macro-scale. Commercial
`scale bins introduce a component of micro-scale mixing due to the
`increased shear provided by the higher drop heights.
`Although the effectiveness of micro-scale mixing in a comil depends
`on the equipment and batch size, a subsequent convective mixing in a
`bin blender removes such dependency by continuing the micro-mixing
`concurrently with macro-mixing. These findings indicate that a
`combination of micro- and macro-scale mixing mechanism is re-
`quired for homogeneous mixing of cohesive materials.
`
`Acknowledgement
`
`The authors would like to acknowledge the following BMS colleagues,
`Deniz Erdemir who provided SEM images of pure API and Lynn
`DiMemmo who provided SEM images of the blend. The authors
`also want to acknowledge the reviewer who provided profound
`insight into the mechanism discussed in the current work.
`
`750 kg
`300 kg
`20 kg
`
`Blend 1
`
`Blend 2
`
`Blend 3
`
`50
`
`40
`
`30
`
`20
`
`10
`
`0
`
`RSDm(%)
`
`Fig. 7. Within-sample uniformity of blend through the process (300 kg).
`
`Fig. 9. Overall uniformity of blend through the process (5% API).
`
`Cosmo Ex 2023-p. 4
`Mylan v Cosmo
`IPR2017-01035
`
`

`

`W. Shi et al. / Powder Technology 274 (2015) 319–323
`
`323
`
`Fig. 10. Coating of excipients by API. Top figures: needlelike pure API; Bottom figures: blend 2 showing coating of excipient by needle like API.
`
`Bin Blender
`
`Macro Mixing
`
`Comil
`
`Micro
`Mixing
`
`Bin Blender
`
`Macro and
`micro Mixing
`
`Effectiveness of mixing is low
`at micro and macro level
`
`Effectiveness of mixing is low at
`macro level but high at micro level
`dependent on batch size. Coating
`of API is the result of micro mixing.
`
`Large MD, RSDS, RSDm
`
`macro and micro levels.
`
`Reduction in RSDS and RSDm,
`but not necessarily MD
`
`Further reduction in RSDS,
`RSDm, and MD
`
`Fig. 11. Schematic mixing mechanism of API X.
`
`References
`
`[1] B. Chaudhuri, A. Mehrotra, F.J. Muzzio, M.S. Tomassone, Cohesive effects in powder
`mixing in a tumbling blender, Powder Technol. 165 (2006) 105–114.
`[2] J.J. McCarthy, Micro-modeling of cohesive mixing processes, Powder Technol. 138
`(2003) 63–67.
`[3] A. Alexander, O. Sudah, P. Arratia, N. Duong, S. Reynolds, F.J. Muzzio, Characterization
`of the performance of bin blenders part 3 of 3: cohesive powders, Pharm. Technol. 28
`(2004) 54–74.
`[4] N. Harnby, An engineering view of pharmaceutical powder mixing, Pharm. Sci.
`Technol. 3 (2000) 303–309.
`[5] M. Poux, P. Fayolle, J. Bertrand, Powder mixing: some practical rules applied to
`agitated systems, Powder Technol. 68 (1991) 213–234.
`[6] L.R. Schenck, R.V. Plank, Impact milling of pharmaceutical agglomerates in the wet
`and dry states, Int. J. Pharm. 348 (2008) 18–26.
`
`[7] M. Llusa, K. Sturm, O. Sudah, H. Stamato, D.J. Goldfarb, H. Ramachandruni, S.
`Hammond, M.R. Smith, F.J. Muzzio, Effect of high shear blending protocols and
`blender parameters on the degree of API agglomeration in solid formulations, Ind.
`Eng. Chem. Res. 48 (2009) 93–101.
`[8] Q. Zhou, L. Shi, W. Marinaro, Q. Lu, C.C. Sun, Improving manufacturability of an
`ibuprofen powder blend by surface coating with silica nanoparticles, Powder
`Technol. 249 (2013) 290–296.
`[9] M.P. Mullarney, L.E. Beach, R.N. Davé, B.A. Langdon, M. Polizzi, D.O. Blackwood,
`Applying dry powder coatings to pharmaceutical powders using a comil for improving
`powder flow and bulk density, Powder Technol. 212 (2011) 397–402.
`[10] F.J. Muzzio, A.W. Alexander, Scale up of powder-blending operations, Pharm.
`Technol. 29 (2005) 34–42.
`
`Cosmo Ex 2023-p. 5
`Mylan v Cosmo
`IPR2017-01035
`
`