Safety risk management for low molecular weight process‐related impurities in monoclonal antibody therapeutics: Categorization, risk assessment, testing strategy, and process development with leveraging clearance potential

Abstract Process‐related impurities (PRIs) derived from manufacturing process should be minimized in final drug product. ICH Q3A provides a regulatory road map for PRIs but excludes biologic drugs like monoclonal antibodies (mAbs) that contain biological PRIs (e.g. host cell proteins and DNA) and low molecular weight (LMW) PRIs (e.g., fermentation media components and downstream chemical reagents). Risks from the former PRIs are typically addressed by routine tests to meet regulatory expectations, while a similar routine‐testing strategy is unrealistic and unnecessary for LMW PRIs, and thus a risk‐assessment‐guided testing strategy is often utilized. In this report, we discuss a safety risk management strategy including categorization, risk assessment, testing strategy, and its integrations with other CMC development activities, as well as downstream clearance potentials. The clearance data from 28 mAbs successfully addressed safety concerns but did not fully reveal the process clearance potentials. Therefore, we carried out studies with 13 commonly seen LMW PRIs in a typical downstream process for mAbs. Generally, Protein A chromatography and cation exchange chromatography operating in bind‐and‐elute mode showed excellent clearances with greater than 1,000‐ and 100‐fold clearance, respectively. The diafiltration step had better clearance (greater than 100‐fold) for the positively and neutrally charged LMW PRIs than for the negatively charged or hydrophobic PRIs. We propose that a typical mAb downstream process provides an overall clearance of 5,000‐fold. Additionally, the determined sieving coefficients will facilitate diafiltration process development. This report helps establish effective safety risk management and downstream process design with robust clearance for LMW PRIs.

it clearly states that biologic drugs are excluded. For biologic drugs, safety risk concerns from PRIs are currently addressed primarily on a case-by-case basis and carried out in different ways developed by pharmaceutical companies. 6 For biologic drugs like mAbs, PRIs generally arise from the cell substrates (e.g., host cell proteins and host cell DNA), the cell culture process (e.g., media components and antifoam), and the purification process (e.g., Protein A leachate from affinity column and detergents used for viral inactivation). 1 Safety risks of the biologically derived macromolecules or biological PRIs (such as host cell protein, DNA and protein A leachate) are managed through routine testing to assure to be below acceptable ranges. [6][7][8][9] The risk assessments, process clearance, and assays for biological PRIs have been reviewed in multiple recent publications. [10][11][12][13][14][15] Putatively acceptable residual levels that are based on human consumption safety history or observations from clinical trials are often used to guide process development, such as 100 parts-per-million (ppm) for residual host cell proteins (HCPs) 10 and 10 ng per dose for DNA. 14 Most of the upstream PRIs (e.g., vitamins and anti-foam) and downstream PRIs (e.g. buffers and reagents) have low molecular weight (LMW) compared to the biological PRIs such as HCPs and DNA. These PRIs are usually considered too small to constitute epitopes that can be recognized by the mammalian immune system, 16 thus the immunogenicity risk is fairly low and can be neglected. Some LMW PRIs (e.g. metal ions) potentially impact protein stability as discussed in a recent review paper 13 and the impact can be evaluated by stability studies, therefore, these risks are not discussed in this report.
We are focused on the safety risk arising from potential toxicity of the LMW PRIs. ICH Q3C (R6), 17 Q3D (R1), 18 Q6B, 1 Q9, 19 and M7 (R2) 20 guidelines provide relevant guidance and recommendations, however, safety risk assessment for PRIs in biologics drugs remains complicated. 9,21,22 Testing every LMW might be the most assuring approach to guarantee no safety risk to patients, but routine tests of all LMW PRIs for every manufacturing lot are unrealistic and unnecessary. Therefore, a science-based safety risk assessment is highly encouraged to meet regulatory expectations and pharmaceutical companies often implement a safety risk assessment-guided-testing strategy for LMW PRIs. 6,22,23 In this report, we discuss a LMW PRI safety risk management process that consists of multiple stages that can be integrated with CMC development activities. The categorization, risk assessment approaches, testing strategy, downstream clearance, decision tree, and process development aiming for robust PRI removals are also discussed.

| Risk assessment approaches, impurity safety factor and clearance calculation
A risk assessment can be carried out using PDE (permissible daily exposure), which is the maximum acceptable intake per day of an impurity in pharmaceutical products. 17 A PDE is usually derived preferably from NOEL (no-observed-effect level) with the following Equation (1): where F1 accounts for extrapolation between species, F2 is a factor of 10 to account for variability between individuals, F3 is a variable factor to account for toxicity studies of short-term exposure, F4 is a factor that may be applied in cases of severe toxicity, and F5 is a variable factor applied if LOEL (low observed effect level) is used. 17 For a LMW PRI with no available NOEL or LOEL, that is, PDE cannot be determined through Equation (1), a safety risk assessment can be carried out with an impurity safety factor (ISF) calculation. 6 ISF represents the distance between a toxicity dose and a PRI dose in a product dose. ISF is calculated with the following Equation (2) When the testing result for the PRI was "not detectable," the assay limit of detection (LOD) was used in the equation.  0.1 M sodium hydroxide for membrane storage. The impurity to be tested was spiked before the start of diafiltration. Samples were taken after each DV and tested by the corresponding qualified assay. Clearance of the tested PRIs was analyzed with the following Equation (5) 24 :

| Chromatography instrument and operations
where C is the final concentration of the PRI, C 0 is the initial PRI concentration, N is the number of DV, and S is the sieving coefficient.
S was determined by fitting the equation to the experimental data.
Unless mentioned otherwise the fittings had R factors greater than 95%.

| Analytical assays and sample testing
The assays for the 13   The safety risk level is determined according to the safety and toxicity data available in scientific literature and public databases, as well as information from regulatory guidance. 4 In brief, PRIs carry low safety risk are considered as "known-to-be-safe" and can be eliminated from the safety risk management process. PRIs with reported medium toxicity are considered to pose medium risk, while PRIs with reported genotoxicity and carcinogenicity are considered to pose high risk. PRIs with medium and high risks are carefully managed in the following three parts. The Part (1) categorization mainly focuses on toxicity of PRI and the risk associated with usage amount is evaluated in the following Part (3). Part (2) consists of process development. As a rule of thumb, high-risk PRIs should be avoided; mediumrisk PRI usage should balance risk and process benefit after process clearance knowledge is obtained; while low-risk PRI usage may have more flexibility to maximize process benefits. Additionally, acceptance criterion can be set up for raw materials to simplify risk management and reduce the testing burden. Maintaining process clearance data for LMW PRIs builds the knowledge base about the process clearance potential for different PRIs, which has the potential to reduce future process development activities. Part (3) is a safety risk assessment for the remaining high-and medium-risk PRIs to further define their safety risk levels. Generally, when a PRI has a significant safety margin or its residual level is well below the safety dose, the PRI can be considered to pose a low safety risk. PRIs with limited safety margin (a level close to the safety dose) or no safety margin (the level is at or above the safety dose), additional actions must be taken (such as testing or process change) to minimize their safety risks. Part (4) consists of assay development and testing for LMW PRIs that are identified in Part (3) to demonstrate process clearance. A suitable assay with sufficient sensitivity needs to be developed and confirmed compatible with the samples to be tested.
Appropriate testing points need to be selected and a PRI testing plan is established for GMP manufacturing.
Overall, implementation of safety risk management processes help to systematically eliminate safety risk and meet regulatory expectations, as well as streamline CMC development.
F I G U R E 1 Schematic of safety risk management process for LMW PRIs F I G U R E 2 Decision tree for LMW PRIs risk assessment. PDE, permitted daily exposure; TTC, threshold of toxicological concern 3.2 | Categorization, safety risk assessment approaches, and decision tree for LWM PRIs Categorization of LMW PRI is an initial risk identification step. Generally, LMW PRIs can be categorized into three groups based on toxicological risks: A, B, and C ( Figure 2).
Category A contains LMW PRIs that inherently pose no safety risk and are termed "known-to-be-safe" PRIs within the safety risk assessment. Many LMW PRIs derived from upstream processes are nutrients (such as amino acids, vitamins, salts, lipids, carbohydrates and trace elements) required for cell growth. Many of these PRIs can be found in humans as naturally existing chemicals, that is, human metabolites.
Metabolites and the concentration range in humans can be found in the Human Metabolome Database. 26 23 Schenerman et al 23 proposed an approach termed "impurity safety factor (ISF)" to measure the distance between the PRI level in a dose of product to the established toxicity dose. The PRI is considered to pose no safety risk only when the ISF is greater than the defined threshold value. Subsequently, the CMC Biotech Working Group, consisting of industry experts, adopted this ISF approach in a white paper entitled "A Mab: A Case Study in Bioprocess Development" 6 ; and the PhRMA working group included the ISF approach in its advice on applying "quality by design for biotechnology products." 7 To measure safety risk of Category B1 PRIs, ISF values can be calculated using Equation (2). The threshold ISF value can be carefully determined based on the dose-response relationship. 23 For Category C PRIs, risk assessment is performed by comparing PRI dose in a product dose to a TTC value (typically 1.5 μg/day). 20 Step 2 safety risk assessment is divided into two sub-steps, as shown in Figure 2.
Step 2a risk assessment uses worst-case assumptions. The main assumption is that the PRIs are copurified with the product to final drug substance, that is, process clearance is not con- improve the process to get sufficient removal) or the analytical method (e.g., poor sensitivity) need to be improved, and the ISF should be recalculated until it is acceptable.
Additionally, for PRI that has no available safety/toxicity data or PRI without chemical identity, risk assessment can be carried out assuming that the PRI has the highest safety risk and follow the assessment workflow for a Category C PRI. Step 2a assessment was carried out for the nine PRIs.

| Safety risk assessment of LMW PRIs in a mAb
Step 2a results suggested that six out of the nine PRIs posed no safety risk even without accounting for process clearance. The remaining three PRIs were considered to pose safety risks without accounting for process clearance. Accordingly, in-process testing for three PRIs was added to the testing plan for GMP manufacturing, and testing was carried out accordingly at the viral filtration step to demonstrate the process clearance. The three PRIs (an antifoam, an anti-shear protectant, and a chemical reagent for cell line selection) were not detected in the samples by the corresponding assays.
Step 2b assessment was carried out using the corresponding assay detection limits and the results demonstrated that the three PRIs posed no safety risks. Therefore, the 105 PRIs in this example posed no safety risk and their safety risks were successfully managed.

| Clearance data for 6 LMW PRIs from 28 mAbs
Removal of LMW PRIs by downstream manufacturing processes is one aspect that determines their risk level. Therefore, downstream process clearance is critical for the PRI safety risk mitigation. Knowing the process clearance potential should help risk mitigation. Figure 5 shows the clearance data from large-scale GMP manufacture of 28 dif-  Table 1). These PRIs have molecular weights ranging from 70 to 9,000 g/mol and have different physical properties such as charge and hydrophobicity.

| Clearance of LMW PRIs in Protein A chromatography
The results from the spiking/clearance study on Protein A chromatography are summarized in Table 2 and Figure 7a. As shown in Table 2, Interestingly, low levels of dextran sulfate and Triton X-100 were detected in the Wash fractions but they were not detected in the Eluate fraction. The results indicate that these two PRIs were weekly retained on the Protein A column during loading, likely due to weak interactions with the mAb proteins or the resins. These weak interactions were effectively disrupted by the wash condition because the two PRIs were not detected in the Eluate. Therefore, a wash condition can further improve PRI removal capability of Protein A chromatography. Due to potential weak interactions between LMW PRIs and the mAb, mAb properties (such as charge and hydrophobicity) may affect LMW PRIs removal. As shown in Figure 7a, clearance of the same PRI was similar between the two different mAbs, suggesting the contribution from mAbs to PRI removal may be negligible.
As shown in Figure 7a, more than 1,000-fold clearance was achieved for all tested LMW PRIs. Clearance for EDTA, polysaccharide, MSX, PEG8000, and Triton X-100 were greater than 10,000-fold. Along with the historical data summarized in Figure 5 and the recent publication, 4

| Clearance of LMW PRIs in cation exchange chromatography
The results from spiking/clearance studies from cation exchange chromatography are summarized in Table 3 and Figure 7b. Except for dextran sulfate and Triton X-100, all of the spiked PRIs in the feed were removed in the column flow-through, and the remaining level in the elution fraction was very low with most as "not detectable." The clearance for BME, polysaccharide, monothiol glycerol, Pluronic F68, and simethicone was more than 1000-fold. Unlike Protein A chromatography, removal of dextran sulfate by cation exchange chromatography was not as effective compared to the other tested PRIs.
Considering that dextran sulfate is negatively charged under the pH conditions, 31 its interactions with the positively charged mAbs may reduce the clearance. The removal of Triton X-100 on cation exchange chromatography was also less than that on Protein A chromatography. Similar clearance of dextran sulfate and Triton X-100 on cation exchange chromatography were also observed on mAb B (Figure 7(b)), suggesting that mAb-specific interactions are unlikely the major reason. The retention mechanism is likely weak interactions between Triton X-100 and the mAb; however, further investigation is needed to confirm this hypothesis.  Figure 7c, copper ion and MSX were effectively removed by diafiltration. A typical 6 DV diafiltration resulted in greater than 300-fold clearance for these two PRIs. Equation (5) was fitted to the data shown in Figure 7c and the sieving coefficients for copper(II) and MSX were estimated to be 1.09 and 1.02, respectively, suggesting nearly ideal sieving. Significant removal of EDTA, tropolone, and caprolactam was also achieved by diafiltration, although the clearance for the three PRIs was not as effective as copper ion and MSX. Accordingly, the obtained sieving coefficients for these three PRIs were in the range of 0.58-0.83. Very limited clearance was obtained for Pluronic F68, even with a spiked concentration (450 μg/ml) that was significantly lower than the critical micelle concentration (1,900 μg/ml). 32 Poor clearance is expected when the Pluronic F68 concentration is higher than critical micelle concentration because the size of the micelles is greater than the TFF membrane MWCO. The sieving coefficient for Pluronic F68 was estimated out to be 0.11.

| Clearance of LMW PRIs during diafiltration
Copper ion had a sieving factor slightly greater than one likely due to electrostatic repulsions between the positively charged copper ion and the positively charged mAb. 33,34 Similarly, EDTA removal was not as effective as copper ion or the neutrally charged MSX  In terms of reasonably reducing testing burden, removal of all inprocess testing for PRIs and assuming good clearance certainly poses significant risks. Based on the results presented in this work, assuming no process clearance is highly conservative, and assuming some degree of process clearance for typical mAb downstream process has scientific basis and is reasonable. Our studies showed that Protein A chromatography and cation exchange chromatography (operated in bind-elute mode) had greater than 1,000-and 100-fold clearance, respectively. The typical diafiltration process is also capable of removing LMW PRIs, generally more than 100-fold but the clearance potential can be affected by PRI chemical properties such as charge and hydrophobicity as demonstrated in this study and several recent reports. 24,34 The downstream clearance potential must be considered in the safety risk assessment to avoid any unnecessary testing. In the absence of clearance data, the initial risk assessment (Figures 2 and 3) based on the worst-case assumption that there is no clearance during downstream processing is likely to lead to some testing. However, the clearance data presented here suggests that it is quite reasonable to assume some conservative level of clearance, which can help reduce the testing burden. For the overall process, a minimum clearance of 5,000-fold can be assumed for mAb purification processes, with 100-fold clearance from the Protein A chromatography step, 10-fold clearance from the cation exchange chromatography (in bind-elute mode), and fivefold clearance from the diafiltration process. With this assumption, an additional assessment taking into account a minimum process clearance can be added to the decision tree in Figure 2. After accumulating sufficient clearance data, testing for some LMW PRIs may be avoided for mAbs using the platform process. The gained clearance potential of each unit operation also facilitates the process development for a new mAb. It is noteworthy that clearance of dextran sulfate and Triton X-100 by bind-elute cation exchange chromatography was significantly lower than the clearance by Protein A chromatography. The poor clearance of dextran sulfate may be explained by the potential electrostatic interactions between dextran sulfate and the resins or bound mAbs. The mechanism for the retention of Triton X-100 by the cation exchange chromatography would need further studies. Furthermore, we found that the properties (such as charge and hydrophobicity) of protein and/or PRI could impact the clearance during tangential flow filtration through potential weak interact rations between PRI and proteins. These undesired interaction led to lower sieving coefficients for several commonly seen PRIs.
The sieving coefficients obtained in our study for the commonly seen LMW PRIs can be used to guide diafiltration development to achieve desired clearance. Taken together, this report establishes an effective safety risk management and rational design of robust downstream process for LMW PRIs. writing-original draft; writing-review and editing.