Received: 16 May 2019 Revised: 13 September 2019 Accepted: 17 September 2019

DOI: 10.1002/btpr.2923

RESEARCH ARTICLE

Clearance of solvents and small molecule impurities in

antibody drug conjugates via ultrafiltration and diafiltration
operation

Travis J. Gates | Yaqi F. Lyu | Xin Fang | Xiaoli Liao

Process R&D, AbbVie Inc., North Chicago,

Illinois Abstract
Ultrafiltration and diafiltration (UF/DF) processes by tangential flow filtration (TFF)
Correspondence
Xiaoli Liao, Process R&D, AbbVie Inc., North are frequently used for removal of solvents and small molecule impurities and for
Chicago, IL. buffer exchange for biopharmaceutical products. Antibody-drug conjugates (ADCs)
Email: [email protected]
as an important class of biological therapeutics, carry unique solvents and small mole-
Present address cule impurities into the final UF/DF step as compared to standard antibody prepara-
Travis Gates, Department of Biochemistry,
Augustana College, Rock Island, IL. tion. The production process of ADCs involves multiple chemical steps, for example,
reduction and conjugation. The clearance of these solvents and small molecules by
Peer Review
The peer review history for this article is UF/DF, specifically the DF step, has been assessed and described herein. The rates
available at https://publons.com/publon/10. of clearance for all the impurities in this study are close to the ideal clearance with no
1002/btpr.2923.
apparent interaction with either the protein or the TFF membrane and system. The
effect of process variables during DF, such as pH, temperature, membrane loading,
transmembrane pressure, and cross flow rate, has also been evaluated and found to
have minimal impact on the clearance rate. These results demonstrate efficient clear-
ance of solvents and small molecule impurities related to the ADC process by the DF
process and provide a general data package to facilitate risk assessments based on
the sieving factors and program specific needs.

KEYWORDS
ADC, impurity clearance, UF/DF

1 | I N T RO D UC T I O N the protein.2,3 In one study, seven extractables and leachables from

single-use technologies, of different chemical families and molecular
Ultrafiltration and diafiltration (UF/DF) by tangential flow filtration weights, were spiked in and evaluated for clearance from proteins by
(TFF) is a frequently used final purification and buffer exchange step UF/DF process. The results demonstrated effective removal of all
for biopharmaceutical products. TFF allows liquid to flow in the direc- seven extractables and leachables.4
tion parallel to the membrane, generating crossflow current that Antibody drug conjugates (ADCs) have emerged as a new class of

reduces accumulation of protein on the membrane surface and biotherapeutics in areas for oncology, infectious diseases, and immu-

improves permeation of solvents through the ultrafiltration mem- nological indications.5,6 ADCs are comprised of antigen-targeting
monoclonal antibodies (mAbs) and therapeutic drugs that are attached
brane.1 This unit operation is a mainstay in concentration of proteins
covalently via chemical linkages (i.e., linkers). Most ADCs rely on
and buffer exchanges and has shown to be effective at removing sol-
chemical methods to link therapeutic drugs to lysines or reduced
utes when the molar masses of the solutes are significantly lower than
interchain cysteines on the native antibody, for example, the commer-
Travis Gates and Yaqi F. Lyu contributed equally to this study. cial ADCs, Kadcyla, Mylotarg, Besponsa, and Adcetris.6-8 The chemical

Biotechnology Progress. 2019;e2923. wileyonlinelibrary.com/journal/btpr © 2019 American Institute of Chemical Engineers 1 of 8

https://doi.org/10.1002/btpr.2923
2 of 8 GATES ET AL.

processes involved in the ADC production generate a unique collec- of the solvents or small molecules shown in Table 1. As shown in
tion of solvents and small molecule impurities that is not typically Table 2, the spiked solvents and small molecule solutions were: neat
associated with traditional mAb production. However, there is a gen- DMA, neat DMSO, 4.4 mg/ml of TCEP=O in water, and 11 mg/ml of
eral lack of reported data on the UF/DF removal of small molecule NAC in water. EDTA was present at 1.84 mg/ml as a component in
impurities from ADC processes, prompting the herein described PBSE solution. Each load sample contained 20 mg/ml of mAb with dif-
study. ferent final concentrations of the spiked impurities listed in Table 2.
For ADCs using reduced interchain cysteines for conjugation, the For the AB095 mAb, the mAb solution was mixed in a 9:1 ratio
chemical steps include reduction of interchain disulfide bonds, conju- with either neat DMA or 11 mg/ml of NAC in water to give
gation, and quench of thiol-reactive drug-linkers before the final 10.3 mg/ml of mAb with the corresponding final concentration of the
UF/DF operation.9-11 In this article we focus on two commonly uti- spiked impurities listed in Table 2.
lized cosolvents and three small molecules (see Table 1) that are fre-
quently present in the UF/DF unit operation. The clearance of drug-
linker related species by UF/DF is specific to each ADC program and
2.3 | TFF operation: Setup and sampling
needs to be studied separately to meet the specification for each pro- A bench-top UF/DF system shown in Figure 1 was used for the study.
gram, therefore is beyond the scope of this article. Tris A 30 kDa Millipore Pellicon-3 Ultracel (regenerated cellulose) mem-
(2-carboxyethyl)phosphine (TCEP) is typically used for reduction and brane cassette with 88 cm2 surface area and Type C screen (catalog
is converted to the oxidized form (TCEP=O) upon completion of the number P3C030C00) was implemented. Prior to each run, the system
reduction. Ethylenediaminetetraacetic acid (EDTA) is typically added was flushed and sanitized overnight with sodium hydroxide (0.1 N),
for metal chelation to minimize re-oxidation of the newly freed cyste- then flushed with purified water and equilibrated with the DF buffer
ines. Dimethyl sulfoxide (DMSO) and N,N-dimethylacetamide (DMA)
(15 mM Histidine pH 6.0 buffer).
are commonly used to dissolve the typically hydrophobic drug-linker
After the system was emptied, ~50 ml of load sample containing
and serve as cosolvents during the conjugation. Once the conjugation
the solvent or small molecule impurity was added to the UF/DF sys-
is complete, N-acetyl-L-cysteine (NAC) can be used to quench any
tem to yield a loading of 113 g protein per m2 membrane. The system
excess amount of thiol-reactive drug-linkers, often maleimide-
was then operated in constant volume and constant transmembrane
containing drug-linkers.12 In this article, the clearance of these five
pressure (TMP) DF mode, where solution in the reactor was pumped
molecules, with molecular weights ranging from 78.1 to 292.2 Da, is
across the membrane and filtered, while fresh buffer was fed continu-
studied in the presence of one mAb to determine the clearance rate.
ously to keep the constant volume in the reactor. The resulting con-
The operational variables of UF/DF and incorporation of two distinct
stant volume in the system is referred to as diavolume (DV). The
mAbs are also studied to probe the robustness of the clearance. Here
number of DVs (N) is defined as the volume of the permeate removed
mAbs are used as surrogate to the ADCs assuming minimal impact of
divided by DV. In this study, DF operation typically ranged from 8 to
drug-linker conjugation on the clearance rate of the listed solvents
10 DVs. The standard UF/DF conditions were as follows: room tem-
and small molecules, as well as nearly complete retention of mAb or
perature at around 21
C, cross flow rate, or feed flow rate, at
ADC by the membrane.
300 L m−2 hr−1, or 44 ml/min, and TMP ([P1 + P2]/2 – P3) at
18–20 psi. Under this operational condition, the average permeate
2 | MATERIALS AND METHODS flux was about 66 L m−2 hr−1 at the first DV, and increased to about
108 L m−2 hr−1 at 8 DV of DF due to the change in solution composi-
2.1 | Materials tion by buffer exchange.

Antibodies employed in this study were manufactured at AbbVie. Two

mAbs were used, 34C3 monoclonal antibody formulated at 50 mg/ml in T A B L E 1 Solvents and small molecule impurities studied for
15 mM Histidine, pH 6.0 buffer, and anti-TeTx (AB095) monoclonal anti- UF/DF clearance
body formulated at 11.4 mg/ml in 30 mM Histidine, 8% sucrose, pH 6.0 MW
buffer. TCEP=O was synthesized at AbbVie. EDTA disodium salt Compound Purpose in ADC process (Da)
dihydrate, DMSO, DMA, and NAC were purchased from Sigma Aldrich. Tris(2-carboxyethyl) Oxidized form of reducing 266.2
PBSE buffer (potassium phosphate, NaCl, EDTA, pH 7.2) and 15 mM phosphine oxide reagent
Histidine buffer pH 6.0 were purchased from SAFC. Ethylenediaminetetraacetic Chelation reagent for 292.2
acid reduction
Dimethyl sulfoxide Solvent for drug-linker 78.1
2.2 | Load sample preparation
N,N-Dimethylacetamide Solvent for drug-linker 87.1
The load samples were prepared by a gravimetric method based on N-acetyl-L-cysteine Quench reagent for drug- 163.2
the density of each component. For a 50 ml load solution containing linker
the 34C3 mAb, 20 ml of mAb solution was diluted with 25 ml of Abbreviations: ADC, antibody-drug conjugate; MW, molecular weight;
PBSE, followed by ~5 ml of the solution containing specific amounts UF/DF, ultrafiltration and diafiltration.
GATES ET AL. 3 of 8

TABLE 2 Preparation of 50 ml of load sample from spiked compounds at different concentrations

5 ml stock solution of Stock concentration of Final concentration of Final concentration

mAb solution spiked compounds spiked compounds (mg/ml) spiked compounds (mg/ml) of mAb (mg/ml)
20 ml of 34C3 at 50 mg/ml, DMA Neat 94
diluted with 25 ml of PBSE DMSO Neat 110
TCEP=O 4.4 0.43 20
NAC 11 1.1
Water (EDTA in PBSE) 1.84 0.92
45 ml of AB095 at 11.4 mg/ml DMA Neat 94 10.3
NAC 11 1.1

Abbreviations: DMA, N,N-dimethylacetamide; DMSO, dimethyl sulfoxide; EDTA, ethylenediaminetetraacetic acid; NAC, N-acetyl-L-cysteine; TCEP, tris
(2-carboxyethyl)phosphine.

method with mobile phase of 5% ACN in 20 mM Potassium Phos-

phate, pH 6.5 was used. The concentration of DMA in each sample
was calculated by dilution factor and by comparing the peak area to
that of a 0.1 mg/ml DMA working standard.

2.6 | LCMS analytical method for NAC and TCEP=O

Each sample was mixed in a 1:3 ratio with 600 mM HCl in 90/10
F I G U R E 1 Set-up of the bench-top UF/DF system. UF/DF,
ultrafiltration and diafiltration methanol/water to precipitate the protein, followed by centrifugation
for 10 min at 16.1 RCF. The supernatant was collected, diluted with
Samples (1 ml) were taken at different number of DVs of perme-
water, and injected onto an Ascentis Express RP-Amide column
ate, at every one or two DVs from both the permeate port and from
(4.6 × 150 mm, 2.7 μm, Supelco). NAC and TCEP=O were quantified
the reactor (retentate). Before taking retentate samples from the reac-
by single quad LC–MS under selective ion monitor (SIM) mode. The
tor, the valves of DF buffer and permeate were closed, and the
concentration of NAC and TCEP=O in each sample were calculated
retentate solution was recirculated through the membrane for ~5 min
by dilution factor and by comparing the peak area to that of a
to ensure homogeneity of the solution in the system. Sampling also
0.1 mg/ml NAC and TCEP=O working standard. Recovery of NAC and
contributed to a volume decrease of the retentate in the reactor and TCEP=O was between 90 and 105% depending on the type of protein
was corrected for the following DF operation. and concentration of the analytes.

2.4 | HPLC analytical method for DMSO 2.7 | HPLC analytical method for EDTA
An amount of 200 μl of each sample solution was accurately transferred Each sample was 1:1 mixed with acetonitrile to precipitate the pro-
into a 5 ml volumetric flask and diluted to volume with acetonitrile. The tein, and was placed on the centrifuge for 10 min at 16.1 RCF. The
mixture was agitated and the resulting protein precipitation was allowed supernatant was collected, and was either diluted with water or
to settle for ~5 min, followed by filtration through a 0.2 μm PTFE filter injected neat onto a PRIMESEP D column (4.6 × 150 mm, 5 μm,
to remove the precipitate. Each filtrate was either diluted with 15 mM SIELC). Mobile phases contained 1.26 mM CuSO4, and EDTA formed
Histidine buffer or injected neat onto an Ascentis Express HILIC column a UV active complex with copper ion when mixed with mobile phase
(3.0 × 150 mm, 2.7 μm, Supelco, Catalog No: 53972-U). The concentra- through the HPLC system. The concentration of EDTA in each sample
tion of DMSO in each sample was calculated by comparing the peak area was calculated by dilution factor and by comparing the peak area to
to that of a 0.1 mg/ml DMSO working standard and accounting for the that of a 0.012 mg/ml EDTA working standard.
dilution factor of the sample preparation. Recovery of DMSO was per-
formed for the analytical method and was above 95% at both working
2.8 | TFF parameter exploration
standard level and quantitation limit level.
Several TFF parameters were explored for DMA to determine their
impact on the clearance rate. Two parameters were studied separately
2.5 | HPLC analytical method for DMA
while keeping the other parameters the same, such as the pH of the
Each sample was either directly diluted with 15 mM Histidine buffer DF buffer (15 mM Histidine), which varied from pH 6.0 to 5.0 and
or injected neat onto an ODS-2 HYPERSIL column (4.6 × 150 mm, pH 7.0, and the membrane loading, which was increased to 454 g pro-
5 μm, Thermo Scientific, Catalog No: 31605-154630). An isocratic tein per m2 membrane. Three other parameters were studied in four
4 of 8 GATES ET AL.

experiments with different combinations of the following: the temper- the DF, and Ci is the initial concentration of the species of interest at
ature was varied from 21 to 28
C and 12
C; cross flow rate was var- the start of DF, S is the sieving factor, and N is the number of DVs.
ied from 300 to 186 L m−2 hr−1 and 420 L m−2 hr−1; TMP deviated
from 18 to 30 psi and 12 psi. Some of the conditions might be in the C f = Ci e − ðS × NÞ ð1Þ
pressure dependent regime for this TMP range explored.
To evaluate the clearance rate of the solvents and small molecule
impurities listed in Table 1, each was spiked at different levels in
3 | RESULTS AND DISCUSSION
34C3 mAb solution and was subjected to a constant volume DF as
described above. The samples collected at different DVs were ana-
3.1 | Determination of clearance rate of solvents and
lyzed for the concentration of each compound in both retentate solu-
small molecule impurities by UF/DF
tion and permeate solution. To determine the sieving factor for each
During UF/DF process, sieving factor (S), defined as the ratio of con- species, the natural logarithm of the concentration (mg/ml) of each
centration of the species in the permeate to the concentration of the species was plotted against the number of DVs, and the negative
13
species in the retentate, is often used to characterize the clearance slope from the linear regression represents the sieving factor. Here,
rate. The sieving factor can be species dependent. A species that is the calculated sieving factor represents the apparent sieving factor
completely retained will have a sieving factor of 0, while a species that that is based on the permeate concentration and mixing-cup concen-
is ideally cleared or freely permeable will have a sieving factor of tration of the retentate.14 Figure 2 shows the plot, the linear regres-
1. The clearance of each solvent and small molecule species for a con- sion, and the corresponding sieving factor from both the retentate
stant volume DF can be described by Equation (1),2 where Cf is the solution and the permeate solution for the solvents and small mole-
concentration of the species of interest at certain number of DVs into cule impurities studied here. The clearance trends and the sieving

(a) DMA (b) DMSO

4 Retentate S=0.97 4 Retentate S=0.98

Ln (Conc. mg/mL)

Ln (Conc. mg/mL)

Permeate S=0.96 Permeate S=0.94

2
2
0
0
-2
-2
-4

-6 -4
0 2 4 6 8 10 0 2 4 6 8
Number of Diavolumes Number of Diavolumes

(c) NAC (d) EDTA

0 0
Retentate S=0.91 Retentate S=0.95
Permeate S=0.94
Ln (Conc. mg/mL)
Ln (Conc. mg/mL)

Permeate S=0.97
-2
-2

-4
-4

-6
-6
0 2 4 6 8 0 2 4 6
Number of Diavolumes Number of Diavolumes

(e) TCEP=O
0
Retentate S=0.85
Ln (Conc. mg/mL)

Permeate S=0.99
-2

F I G U R E 2 The UF/DF clearance

trend and sieving factor for solvents and
-4
small molecule impurities in 34C3 mAb
solution; each line represents the linear
-6 regression result and the R2 values are all
0 2 4 6 equal to or greater than .99. UF/DF,
Number of Diavolumes ultrafiltration and diafiltration
GATES ET AL. 5 of 8

factors matched well for the retentate solution and the permeate therefore the load samples were prepared directly in a 9:1 ratio of the
solution of each species, and all the sieving factors approximated ideal mAb solution to the solution of spiked impurity. While maintaining
clearance. Except for the study of DMA where no protein precipita- similar starting concentration of the spiked impurities, the composi-
tion method was used, there were slight differences observed tions of the load samples were very different for these two mAbs,
between the concentrations and sieving factors obtained from the with AB095 samples containing 10.3 mg/ml mAb in 27 mM Histidine
retentate solution and permeate solution for the other four molecules, buffer, 7.2% sucrose and pH around 6, while the 34C3 mAb con-
possibly due to the protein precipitation during sample treatment that taining 20 mg/ml mAb in 6 mM Histidine, 50% PBSE and pH
impacted differently for the retentate solution that contained mAb around 7.
and permeate solution that contained no mAb. Despite the differences in mAb properties and load sample com-
Both solvents, DMA and DMSO, were spiked at 10% vol/vol into positions, DMA and NAC showed similar clearance trends for these
the mAb solution, representing a typical level in a conjugation reaction two mAb solutions, using 15 mM Histidine pH 6.0 as the DF buffer
mixture. They were both cleared efficiently with sieving factors close (Figure 3). In the case of DMA, the clearance trend between these
to 1. DMA had closely matching clearance trends of the retentate two mAbs solutions aligned, and the calculated sieving factors were
solution and permeate solution, while DMSO showed slight differ- similar and close to ideal clearance (S = 1). The sieving factors of NAC
ences between these two, possibly due to the sample treatment dur- in both mAb solutions were calculated as >0.9 and NAC reached
ing the analysis (Figure 2a,b). Starting at around 105 ppm in the <0.03% wt/wt after 6 DV of DF. Despite the difference of sieving fac-
solution or 5 × 10 ppm relative to the mAb, both solvents were
6
tors of 0.97 versus 0.91 for the two mAbs, both resulted in NAC con-
below 2,500 ppm and 50 ppm in the solution after 4 DV and 8 DV of centrations below the acceptable levels during a typical process
DF, respectively, or below 2,500 ppm relative to the mAb after involving more than 6 DVs of DF. Therefore, both clearance trends of
8 DV of DF. DMA and NAC were close to ideal clearance with different mAbs and
NAC and EDTA were spiked at ~1 mg/ml in the mAb solution. sample compositions. We propose that these results would hold true
Both impurities were cleared efficiently, and the sieving factors were as well for other impurities, other mAbs, or other buffer components,
determined to be 0.91 and 0.95 in the retentate solution, respectively provided there is no interaction between the impurities and the mAb
(Figure 2c,d). The concentrations measured in the retentate solution or buffer components.
were slightly lower than that in the permeate solution, which could be
due to the protein precipitation in the retentate solution and slightly
3.3 | Clearance of DMA under various UF/DF
lower recovery of the analytes when preparing the analytical sample.
conditions
In the retentate solution, both impurities reached <0.032% wt/wt rel-
ative to the 20 mg/ml mAb after 5.2 DV of DF. Multiple UF/DF parameters were studied for their impact on DMA
TCEP=O was spiked at 0.43 mg/ml in the mAb solution, which clearance, including the pH of DF buffer, the membrane loading of the
corresponded to 1.6 mM, or 11.8 equivalents relative to the 34C3 mAb, temperature, TMP, and cross flow rate.
mAb at 20 mg/ml. Samples were collected from both retentate solu- The DF buffer was added to the retentate to maintain constant
tion and permeate solution up to 8 DVs. The 6 DV and 8 DV samples volume during DF, and 15 mM Histidine, pH 6.0 buffer was used. The
showed concentrations of ≤4 μg/ml, which were lower than the buffer pH was adjusted to 5.0 by addition of HCl, or to 7.0 by addition
detection limitation of the analytical method for TCEP=O and were of NaOH. As shown in Figure 4a, when the DMA clearance in 34C3
therefore excluded for the determination of fit (i.e., linear regression). mAb solution was tested with the DF buffer at pH of 5.0 or 7.0, no
As shown in Figure 2e, the linear regression based on the first 4 DVs impact was observed on the clearance trend and sieving factor from
yielded sieving factors of 0.85 and 0.99 for retentate solution and per- the retentate solution. Both pH conditions showed the clearance of
meate solution, respectively. In the retentate solution, TCEP=O DMA close to ideal clearance, similar to that encountered with the
reached 0.05% wt/wt relative to the 20 mg/ml mAb after nominal pH 6.0 experiment.
4 DVs of DF. Membrane loading is an important UF/DF parameter. In this
Based on these results, both solvents and all three small molecule study, the membrane loading was increased four folds for the 34C3
impurities tested here exhibited clearance rates close to ideal clear- mAb, from 113 to 454 g/m2, by increasing the volume of the load
ance, and their levels were well controlled for a typical process using sample from 50 to 200 ml, while keeping the concentration of all the
7–10 DVs of DF. solution components the same. As shown in Figure 4b, at the range
tested, the DMA clearance rate was not impacted by increased mem-
brane loading, and the sieving factors were close to 1 for both low
3.2 | Clearance of solvents and small molecule
and high loadings. This result indicated a wide operational range for
impurities with different antibodies
the membrane loading, and possible higher loadings could be tolerated
One solvent (DMA) and one small molecule impurity (NAC) were also without impacting the clearance of small molecule impurities.
tested with a second mAb, AB095. AB095 mAb has higher isoelectric Finally, temperature, TMP, and cross flow rate were studied in
point (pI 9.24) than that of 34C3 mAb (pI 8.7). Also, AB095 was for- combination in four sets of experiments. During UF/DF, retained sol-
mulated at 11.4 mg/ml in 30 mM Histidine, 8% sucrose, pH 6.0 buffer, utes, mostly protein, build up on the membrane surface and form a gel
 6 of 8 GATES ET AL.

(a) DMA (b) NAC F I G U R E 3 The UF/DF clearance

0 trend and sieving factor of the retentate
4 34C3, S=0.97 34C3, S=0.91
solution for DMA and NAC in 34C3 and
Ln (Conc. mg/mL)

Ln (Conc. mg/mL)
AB095, S=0.98 AB095, S=0.97 AB095 mAb solutions; each line
2 -2
represents the linear regression result and
0 the R2 values are all equal to or greater
-4 than .99. DMA, dimethylacetamide; NAC,
-2
N-acetyl-L-cysteine; UF/DF, ultrafiltration
-4 and diafiltration
-6
-6
0 2 4 6 8 10 0 2 4 6 8
Number of Diavolumes Number of Diavolumes

(a) (b) F I G U R E 4 The UF/DF clearance

trend and sieving factor of retentate
4 pH 5, S=0.97 4 113 g/m2, S=0.97 solution for DMA in 34C3 mAb solution
pH 7, S=0.98
Ln (Conc. mg/mL)

Ln (Conc. mg/mL)

454 g/m2, S=1.02 with different pHs (a) or mAb loadings (b);
2
2 each line represents the linear regression
0 result and the R2 values are all greater
0 than .99. DMA, dimethylacetamide;
-2
UF/DF, ultrafiltration and diafiltration
-2
-4

-4 -6
0 2 4 6 8 0 2 4 6 8 10
Number of Diavolumes Number of Diavolumes

TABLE 3 Effect of different UF/DF operational conditions on the clearance rate of DMA

Temperature TMP Feed flux Permeate flux at first DV Conversion ratio at Sieving factor Sieving factor
(
C) (psi) (L m−2 hr−1) (L m−2 hr−1) first DV (%) (retentate) (permeate)
Nominal 21 18 300 68 23 0.97 0.96
Cond. A 12 30 186 46 25 0.95 0.93
Cond. B 12 12 420 59 14 0.95 0.94
Cond. C 28 12 186 56 30 0.98 0.96
Cond. D 28 28 420 97 23 0.99 0.97

Abbreviations: DV, diavolume; TMP, transmembrane pressure; UF/DF, ultrafiltration and diafiltration.

layer. The thickness of this layer could depend on diffusivity of the buffer composition by buffer exchange. Condition D showed higher
solute, permeate flux, and the hydrodynamic shear tangential to permeate flux rate than the nominal condition, while all the other
the membrane.15 Temperature has an influence on the viscosity of three conditions showed lower permeate flux rate than the nominal
the solution and diffusivity of the solute. The cross flow rate condition. In terms of the conversion ratio in Table 3, defined as
directly impacts the shear force on the membrane surface. The pro- the ratio of permeate flux to the feed flux, condition A and D
tein concentration at the membrane surface increases with TMP showed similar conversion ratio as the nominal condition, while
until it reaches the gelation concentration. As a result, all of these condition B and C showed lower and higher conversion ratio than
factors, temperature, TMP, and cross flow, could have an effect on the nominal condition, respectively. The conversion ratio here
gel layer formation and membrane performance. Four conditions reflected the impact of both TMP and temperature. The processing
(A–D) listed in Table 3 with either higher or lower temperature, time and the impact on mAb and ADC properties are other factors
TMP, and/or cross flow rate than the nominal conditions were to consider for the UF/DF unit operation, and are beyond the
studied for DMA clearance in 34C3 mAb solution. The parameter scope of this study. As shown in Figure 5, despite very different
ranges were chosen as high or low end toward the nominal condi- processing conditions used for these four experiments, the DMA
tion, in order to show the tolerance of operational variables during clearance trends aligned well. The sieving factors calculated from
UF/DF process. The permeate flux at the first DV for each condi- both the retentate solution and permeate solution were all very
tion is listed in Table 3 to indicate the differences. The permeate close to those obtained from the nominal condition, which were
flux typically increases in subsequent DVs due to the change in close to ideal clearance (Table 3). Therefore, the operational range
GATES ET AL. 7 of 8

would further depend on the actual dose, duration, and indication of

4 Cond. A
the ADC. The findings from this study can serve as a foundation to
Ln (Conc. mg/mL)
Cond. B
assess the risk that an impurity poses to the quality of the ADC on a
2 Cond. C
case-by-case basis. The levels would likely be further reduced if a
Cond. D
chromatography purification step is involved in the ADC process,
0
which would reduce the amount of solvents and small molecule impu-
-2 rities prior to the UF/DF operation. Additionally, this study evaluated
the impact of process variables during DF, such as pH, temperature,
-4 membrane loading, TMP and cross flow rate, and demonstrated mini-
0 2 4 6 8
mal impact of these variables on the clearance rate. These results
Number of Diavolumes
demonstrate efficient and robust clearance of solvents and small mol-
F I G U R E 5 The UF/DF clearance trend of the retentate solution ecule impurities that are unique to ADC process by the DF operation,
for DMA in 34C3 mAb solution under different operational and provide a general data package to facilitate risk assessments of
conditions; each line represents the linear regression result and the R2 controlling their levels.
values are all greater than .99. DMA, dimethylacetamide; UF/DF,
ultrafiltration and diafiltration

ACKNOWLEDG MENTS
for temperature at 12–28
C, TMP at 12–30 psi, and cross flow rate
We thank Anthony Haight, Dennie Welch, Calvin Becker, Steve Rich-
at 186–420 L m−2 hr−1 had no impact on the clearance rate
ter, and Robert Leanna from AbbVie for helpful discussions and manu-
of DMA.
script reviewing.
These results demonstrate that the operational variables during
UF/DF, including the pH of DF buffer, the membrane loading, temper-
ature, TMP, and cross flow rate, were all well tolerated and had mini- DISCLOS URE OF I NTE RE ST S
mal impact on DMA clearance. Since all the other solutes of interest
All authors are AbbVie employees or summer worker, and may own
evaluated in the study, that is, DMSO, EDTA, TCEP=O, and NAC, had
AbbVie stock. AbbVie sponsored and funded the study; contributed
similar behavior to DMA at the nominal condition, it is plausible that
to the design; participated in the collection, analysis, and interpreta-
the operational variables have minimal impact on clearance of these
tion of data, and in writing, reviewing, and approval of the final
molecules as well.
publication.

4 | CO NC LUSIO N OR CID

Xiaoli Liao https://orcid.org/0000-0002-9692-9855

This study showed conditions when solvents and small molecule
impurities from a typical ADC process (Table 1) were spiked at rele-
vant concentrations into mAb solutions to provide a simulated post-
RE FE RE NCE S
conjugation reaction mixture. From this study, all small molecule
impurities were cleared efficiently by UF/DF, specifically the DF step. 1. Brose DJ, Dosmar M, Jornitz MW. Membrane filtration. In: nail S.L.,
Akers M.J. (eds) development and manufacture of protein pharmaceu-
The rates of clearance for the species were characterized by sieving
ticals. Pharm Biotechnol. 2002;14:213-279.
factor, which were all close to the ideal clearance (S = 0.9–1). Based 2. Grandison AS, Lewis MJ. Ultrafiltraion. Separation Processes in the
on the ICH guideline, the permissible daily exposure of DMA and Food and Biotechnology Industries - Principles and Applications; Cam-
DMSO are 10.9 and 50 mg/day, or 1,090 ppm and 5,000 ppm relative bridge, England: Woodhead Publishing Limited, 1996:97-122.
3. Kurnik RT, Yu AW, Blank GS, et al. Buffer exchange using size exclu-
to the pharmaceutical products based on 10 g/day dosage, respec-
sion chromatography, countercurrent dialysis, and tangential flow fil-
tively. In our study, without taking into account of further ultrafiltra- tration: models, development, and industrial application. Biotechnol
tion step, DMA and DMSO were cleared more than 2,000 times at Bioeng. 1995;45(2):149-157.
8 DV of DF and reached below the conservative 1,090 ppm after 4. Magarian N, Lee K, Nagpal K, Skidmore K, Mahajan E. Clearance of
9 DV, and 5,000 ppm after 8 DV, respectively. As for the other three extractables and leachables from single-use technologies via ultra-
filtration/diafiltration operations. Biotechnol Prog. 2016;32(3):
small molecules, the ICH guidance is to limit to 0.15% wt/wt relative
718-724.
to the ADCs if dosed ≤2 g/day, or 0.05% wt/wt relative to the ADCs 5. Chari RV, Miller ML, Widdison WC. Antibody-drug conjugates: an
if dosed >2 g/day. In our study, the impurities were cleared below emerging concept in cancer therapy. Angew Chem Int Ed Engl. 2014;
0.05% wt/wt after 4 DV of DF for TCEP=O and after 5 DV of DF for 53(15):3796-3827.
6. Hoffmann RM, Coumbe BGT, Josephs DH, et al. Antibody structure
NAC and EDTA. Their levels should all be well controlled below the
and engineering considerations for the design and function of anti-
acceptable level in a UF/DF process that typically employs 7–10 DVs body drug conjugates (ADCs). Oncoimmunology. 2018;7(3):
of DF. For a specific program, the acceptable levels of these impurities e1395127/1-e1395127/11.
8 of 8 GATES ET AL.

7. Hamann PR, Hinman LM, Hollander I, et al. Gemtuzumab ozogamicin, a 13. Millipore Corporation. Protein concentration and diafiltration by
potent and selective anti-CD33 antibody-calicheamicin conjugate for tangential flow filtration. Millipore Technical Brief. Lit. No. TB032;
treatment of acute myeloid leukemia. Bioconjug Chem. 2002;13(1):47-58. 2003.
8. Doronina SO, Toki BE, Torgov MY, et al. Development of potent 14. Arunkumar A, Etzel MR. Negatively charged tangential flow ultrafil-
monoclonal antibody auristatin conjugates for cancer therapy. Nat tration membranes for whey protein concentration. J Membr Sci.
Biotechnol. 2003;21:778-784. 2015;475:340-348.
9. Sun MM, Beam KS, Cerveny CG, et al. Reduction-alkylation strategies 15. Jornitz MW, Meltzer TH, eds. Filtration and Purification in the Biophar-
for the modification of specific monoclonal antibody disulfides. Bio- maceutical Industry. 2nd ed. Boca Raton, FL: CRC Press; 2007.
conjug Chem. 2005;16:1282-1290.
10. Avilés P, Domínguez JM, Guillén MJ, et al. MI130004, a novel antibody–
drug conjugate combining Trastuzumab with a molecule of marine origin,
shows outstanding in vivo activity against HER2-expressing tumors. Mol
Cancer Ther. 2018;17:786-794.
11. Li Y, Gu C, Gruenhagen J, Zhang K, Yehl P, Chetwyn NP. Medley
How to cite this article: Gates TJ, Lyu YF, Fang X, Liao X.
CD. A size exclusion-reversed phase two dimensional-liquid chroma- Clearance of solvents and small molecule impurities in
tography methodology for stability and small molecule related species antibody drug conjugates via ultrafiltration and diafiltration
in antibody drug conjugates. J Chromatogr A. 2015;1393:81-88. operation. Biotechnol Progress. 2019;e2923. https://doi.org/
12. Kozak KR, Tsai SP, Fourie-O'Donohue A, et al. Total antibody quanti-
10.1002/btpr.2923
fication for MMAE-conjugated antibody-drug conjugates: impact of
assay format and reagents. Bioconjug Chem. 2013;24:772-779.