Elizabeth Joseph is research scientist II, BioPharma Solutions at Baxter Healthcare Corporation.

Elizabeth Joseph

Lyophilization is a common step in drug product manufacturing to improve stability of the formulation. Optimization of the lyophilization cycle for each formulation/fill combination includes assessment of the residual moisture level of the lyophilized cake. Residual moisture is a critical parameter for stability of the API in lyophilized drug products, particularly with large-molecule formulations such as monoclonal antibody (mAb) and other protein formulations. Often a minimum moisture level is required for protein structural integrity; however, a maximum moisture limit is implemented to reduce API degradation and ensure elegant cake appearance over time. It is imperative that a suitable moisture method is developed for each product to ensure accurate monitoring of the moisture level.

	Several methods are commonly used in the pharmaceutical industry for residual moisture determination of lyophilized drug products. Methods range from simple drying of samples (loss on drying [LOD]) to titrations (Karl Fischer [KF]) to complex spectroscopic techniques (near infrared spectroscopy [NIR]). Factors such as sample interference, method accuracy, method range, and ability for non-destructive measurement must be considered when choosing an appropriate method for each product. The intent of this article is to provide an overview of common residual moisture methods for lyophilized drug products and discuss the key points to consider during method selection and method development.

	Drug product formulations for many types of molecules are not stable in aqueous solution and often require refrigerated or frozen conditions or the addition of excipients to prevent degradation. Protein and mAb formulations are particularly susceptible to aggregation, deamidation, oxidation, and other degradation pathways not only in solution, but as freeze-dried solids. Residual moisture acts as a plasticizer in predominantly amorphous systems and lowers the glass transition temperature, Tg, of the formulation (1). Even relatively small changes in residual moisture can lead to significant changes in both physical stability and solid-state chemical stability. These types of degradation are more likely to occur with higher levels of water in the formulations; therefore, lyophilization is an appropriate processing step to remove moisture from the formulation.

	In the lyophilization process the solution is frozen, then a high percentage of water is removed through sublimation of the ice under low pressure during the primary drying process. This is followed by the secondary drying process, during which the product temperature can be increased to remove additional water to meet moisture level specifications. Optimization of the lyophilization cycle is required to achieve acceptable cake appearance and reconstitution time without comprising the drug activity and stability. Some residual moisture remains in the drug product after lyophilization and must be measured and monitored as part of lyophilization secondary drying process development, product characterization, and stability studies.

	While low residual moisture levels may seem ideal to prevent degradation, an intermediate level of residual moisture may be needed to ensure acceptable protein activity and reconstitution of the sample (1). Too low residual moisture may cause aggregation, loss of protein activity, reconstitution issues, or even impact further processing steps such as viral inactivation for safety purposes (1,2). The ability to determine the residual moisture of drug products with acceptable accuracy and precision is crucial for product characterization and stability to ensure the moisture levels remain within the desired range of values. Water may be present in lyophilized samples as bound (absorbed), surface (adsorbed), or trapped (1,3,4). Use of different methodologies for residual moisture analysis may yield variable results depending on which type of water is being measured. Typically, residual moisture for lyophilized drug products reflects the amount of surface water present (1,4).

	Standard-setting organizations such as the US Pharmacopeial Convention (USP) include chapters for well-established moisture methods that are routinely used for residual moisture determination in lyophilized drug products (

). While these methods are well-established techniques, they still must be assessed for specificity, accuracy, linearity, precision, and range for products labeled as meeting good manufacturing practice (GMP) requirements.  

	Loss on drying (LOD) is the simplest and oldest technique for residual moisture analysis and may be performed without any specialized laboratory equipment. The sample is removed from the vial and weighed into a dried container. The sample is dried using heat, vacuum, or desiccant to remove moisture, then it is cooled and reweighed. The amount of moisture present is calculated by dividing the weight lost during drying by the original weight (5). While this method is simple, it lacks the accuracy and specificity required for most applications. The loss on drying test yields the amount of weight lost only and is not specific for water.

	Larger sample amounts are needed than with other methods, which can require multiple vials for low-fill products. Other volatiles in the product such as ethanol may also be driven off, yielding inaccurate results. Sample degradation may occur with higher temperatures, which may contribute to falsely high results particularly with proteins. Incomplete drying of samples may yield falsely low results.

	The technique is very sensitive to test conditions, such as how sample containers are dried and stored, how quickly and uniformly the sample is prepared, how the sample is cooled after drying, and how quickly measurements are obtained after drying to prevent ambient humidity contamination. Test conditions can be especially critical for hygroscopic samples or high humidity laboratories. These factors all contribute to poor accuracy and reproducibility when compared to other methods.

	Thermogravimetric analysis (TGA) is a similar technique to LOD, but with the advantage of being able to use a smaller sample size and having improved control over weighing and temperature. The sample is removed from the vial and weighed into an analysis pan inside a low humidity glove box. The analysis pan is sealed for transfer to a commercially available instrument, which is comprised of a programmable oven and microbalance (6). The weight of the sample is continuously monitored while the temperature is increased from ambient to a temperature that will drive off water without degrading the sample.

	Similar to LOD, method development is required to determine the appropriate temperature and time required for the sample to reach a constant weight and moisture cannot be distinguished from other volatiles that may evolve at lower temperatures. TGA may be combined with other techniques such as mass spectrometry to differentiate which transitions are due to residual moisture and which are related to other volatiles (1). Drawbacks of TGA include the need to open the sample vial and lack of specificity for water.

	Karl Fischer (KF) titration was first developed in 1935 (7) and is based on the generation of electrons during the reaction of iodine and water, as follows (8).

	Alcohol (typically methanol) and sulfur dioxide mix in a basic solution (typically imidazole) (

	Iodine is generated electrochemically by the instrument through anodic oxidation of iodide in the solution (

	The iodine reacts with water in the sample. Once all water has reacted with iodine, the generation of iodine stops and the titration ends. The amount of water present is determined by measuring the current and duration of the reaction (

	Measurement of residual moisture in lyophilized products is usually performed using coulometric KF titration, per 

) <921>, Method Ic (9). KF analysis may be performed using manual titration but is most commonly performed using commercially available instrumentation. Commercial reagents and standards are also available, which allow for easy implementation in quality control (QC) laboratories, operation by entry-level scientists, and consistency between laboratories. Coulometric titration is preferred over volumetric titration because lower moisture readings are more accurate and reproducible using this technique.

	In coulometric KF titrators, iodine is generated electrically from iodide in the KF reagent when the drift in moisture measured is above a defined threshold. Once the drift drops below a setpoint due to all the water in the system reacting with the free iodine, the instrument stops producing iodine and the titration reaches the endpoint. For volumetric KF titration, the iodine is delivered from a buret as needed, and the solution is standardized for calculation of the residual water in the sample (7). Coulometric KF instruments are available with or without a diaphragm to separate the anode and cathode.

	Instruments without a diaphragm require only one reagent and reduction of iodine back to iodide at the cathode is prevented by the design of the instrument. Instruments with a diaphragm require separate anolyte and catholyte reagents for operation and are preferred for very low levels of moisture typically found in lyophilized drug products, as reactions reach the endpoint more quickly, and the instrument is more accurate and precise at lower levels.

	Lyophilized samples may be analyzed by direct addition to the titrator or by external extraction using an anhydrous solvent. Typically, extraction using anhydrous methanol is the preferred method because the drug product sample container has minimal exposure to the laboratory environment and methanol is compatible with the KF reaction. Direct addition of the sample to the titrator requires opening the product vial, transferring material to weigh on a balance, and then opening the KF titrator to add the sample. These steps all allow for environmental moisture contamination of the measurement.

	External extraction requires analysis of a blank measurement to subtract the diluent moisture content from sample results, but substantially reduces exposure of the sample to the environment. Anhydrous methanol is transferred to a stoppered container to reduce gradual increase of moisture in the diluent during the test duration. Lyophilized samples are generally stoppered with partial vacuum in the headspace, which can impact the accuracy of the weighed vial; therefore, any vacuum in the sample container is released using a dry syringe needle prior to weighing. Syringes and needles are used to reconstitute the lyophilized sample with anhydrous methanol, and the sample is mixed by vortex mixing or sonication to dissolve or suspend the cake solids.

	Both the weight of the lyophilized cake and the diluent added must be determined. The reconstituted solution is analyzed by KF, and then the moisture content of the cake is calculated by subtracting the diluent moisture level. Duplicate or triplicate analysis of samples is recommended to ensure reproducible results. Common issues during KF analysis include sample interference and sample dissolution difficulties. Many drug product formulations include excipients or active ingredients that interfere with the KF reaction, such as reducing agents like ascorbic acid which react with the iodine in solution or boron compounds like boric acid which react with the methanol to yield water (7, 10). Both types of interference yield artificially high moisture results.

	Performance of spiking studies to determine accuracy and linearity are essential in verifying that no side reactions are occurring that may artificially inflate results. While complete dissolution of the cake in the extraction solvent is not required for reliable, accurate results, incomplete dissolution can result in lower moisture values, clogged syringes, and poor precision. The diluent may be modified to include other reagents, such as formamide for sugars, sugar alcohols, and proteins, which may improve dissolution. Interference due to the presence of ketones or aldehydes in the sample may be overcome using special KF reagents that contain different alcohols or increasing the pH (7).

	Sample interference may often be overcome using KF instruments with an oven attachment. With an oven, the sample is heated, and moisture is driven into the titrator via dry air or nitrogen so that the sample never directly contacts the KF reagent. Because only the moisture is titrated, side reactions are often eliminated. Use of a KF Oven instrument does require the development of appropriate settings for oven temperature, run time, and endpoint criteria, but enables moisture analysis using a well-established and accepted technique.

	Use of KF Oven typically requires opening the sample vial to transfer weighed material to an analysis vial, and so use of a low humidity glove box is recommended, and determination of a method blank is important. Overall, coulometric KF has many benefits and is usually a first-choice method for residual moisture analysis. Tedious manual analysis, use of hazardous reagents, and destruction of the sample are the main drawbacks of this method.

	Spectroscopic methods such as near infrared spectroscopy (NIR) have more recently been adapted for residual moisture determination of lyophilized products. The main reason for the popularity of these types of methods is that they enable lyophilized samples to be analyzed non-destructively. Lyophilized samples typically are produced in glass tubing vials which have a flat bottom surface. The spectrum of the cake is read through the bottom of the vial and correlated to the known moisture result using chemometric software. Moisture determination by NIR uses two regions of water absorption: the OH stretch/HOH bend combination band at approximately 5150–5200 cm

and the first overtone of OH stretch at approximately 6800–6940 cm

	The NIR absorption is not a direct measurement of water content. Instead, a calibration curve must be generated using a reference method such as KF, and the KF moisture values are correlated to the NIR spectra. Lyophilized vials spanning the desired region of residual moisture levels are first scanned using the NIR to obtain the spectrum. Then the same vials are analyzed using a reference method such as coulometric KF to obtain the residual moisture. The residual moisture values are then entered into the NIR chemometric software to correlate the values for each vial and create a calibration curve.

	Several options for correlation are available, including stepwise multiple linear regression (SMLR), partial least square (PLS), and principle component regression (PCR), with PLS being the most commonly used method. The second derivative of the spectrum is used for quantitative analysis. Once the calibration has been created, the correlation must be evaluated for acceptable fit. The coefficient of determination (R2), root mean square error of calibration (RMSEC), and other software specific values such as Performance Index are useful to assess a good calibration curve (11). Settings for smoothing, minor adjustments for the spectral regions, and removal of outlier values may aid in creating a good calibration. At least 20 independent values are required to provide sufficient data for the correlation.

	Once a suitable calibration is obtained, a cross-validation is typically performed to further assess performance. One standard is removed from the model at a time, re-calibration is performed, and predicted values are generated. Cross-validation will yield the R2 of cross validation and root mean square error of cross validation (RMSEV), which are useful in comparing calibrations to find the optimal model. Additionally, more lyophilized samples may be scanned by NIR, analyzed by KF, and then predicted results are compared to the KF result. This external cross-validation allows calculation of the root mean square error of prediction (RMSEP) (11).

	Drawbacks for this type of method include shifting or shrinkage of the cake within the vials resulting in error in obtaining the spectra, necessity of calibration curve creation for each formulation, needing to validate the chemometric software, and potentially missing moisture in the vial headspace. The laser for NIR systems typically reads about 3–4 mm into the cake, through the glass vial. If the cake does not contact the vial surface, the reading will not be as accurate. Furthermore, lyophilized samples with very low fill volumes may not be suitable if the cake is too thin. Having a cake of “infinite thickness” or more than 4 mm eliminates the need to consider depth of the reading into the sample (12).

	As described previously, development of NIR methods requires correlation to a direct method, and the reference method must first be validated prior to supporting NIR calibration creation. Within a standard chemometric software there are many mathematical tools to analyze and fit the correlation data, and achieving acceptable correlation may be difficult. Large sample sizes are required, and samples must be available that span the range of anticipated moisture results. Preparing lyophilized samples across a wide residual moisture range may be difficult, because cake meltback may occur at higher moisture levels and spiking samples with water requires care not to dissolve the cake and to ensure moisture is fully equilibrated throughout the cake prior to NIR measurement. Because other components in a formulation, particularly sugars, may yield changes in the NIR spectrum, a new calibration model must be developed for each formulation (11).

	Chemometric software packages are now available that meet 21 

) Part 11 requirements, but completing such a validation takes time and resources. Despite these drawbacks and the need for more upfront work, NIR is a useful technique, especially when large numbers of samples must be analyzed, or sample quantities are limited and non-destructive testing is desired.

	New techniques for moisture measurement continue to be developed. One such instrument is the Vapor Pro Moisture Analyzer manufactured by Ametek (Rx and XL models), which has been evaluated in the lab as an alternative to KF or LOD, especially for samples that are not compatible with the KF reaction or include other volatiles. The lyophilized cake is broken up using a sharp probe by piercing the stopper and is then placed into the heated chamber of the instrument. As the vial is introduced into the instrument, the stopper is pierced by a hollow needle so that dry nitrogen can flow through the sample as it is heated. Moisture is driven off and measured by a calibrated relative humidity sensor. Sensor readings are combined with nitrogen flow and temperature data to yield the sample moisture result.

	Advantages of the system are that no hazardous reagents are required, most side reactions found in KF titration are eliminated, measurement are specific for moisture, lyophilized sample vials do not need to be opened prior to analysis, and the system may be used by operators with minimal training. Drawbacks of this technique are the need for method development to optimize temperature and analysis time, destruction of the sample, and potentially longer run times than KF. Currently, this technique is not listed in regulatory guidelines or compendia. Validation of a method using the Vapor Pro would be conducted in a manner similar to KF.

	Other techniques such as frequency modulated spectroscopy (Lighthouse analyzers) are also under development, in which moisture is measured by laser absorption in the gas headspace of a vial through the glass container.  The laser is tuned to match the internal absorption frequency of water molecules so that the amount of laser light absorbed is correlated to the amount of water in the headspace.  While this technique has advantages of being non-destructive, rapid, specific for water, and easy for operators to perform, sensitivity of the technique for practical use in residual moisture determination has yet to be demonstrated.

	A 25 mg/mL sucrose/25 mg/mL glycine solution was prepared and lyophilized in 20-mL vials. Historical data supported using coulometric KF as a reference method because no interference or solubility issues were anticipated, and a NIR method was desired because several hundred vials would require residual moisture analysis to support residual moisture mapping of production scale freeze dryers. Vials were removed from the lyophilizer at periodic intervals during the secondary drying process using a sample “thief”, to yield samples with varying moisture levels. The vials were allowed to stand for several days so that residual moisture could equilibrate throughout the cake.

	The NIR spectrum of each sample was recorded in duplicate through the bottom of the vial using the integrating sphere sample model for the NIR (Thermo Antaris II MDS with RESULT software suite for collection and TQ Analyst for calibration model). A total of 51 vials were analyzed by NIR, and 32 vials were analyzed in triplicate by KF using extraction by anhydrous methanol. The KF values were correlated with the appropriate spectra using Thermo TQ Analyst software to create a calibration curve.

	A model was created using partial least square (PLS) technique, constant pathlength, second derivative, Norris smoothing, and spectral regions of 5203.00–5511.56 cm

. The KF moisture value and the average predicated moisture value from the calibration were compared to check performance of the correlation model. See 

for example NIR spectra and second derivative spectra and 

for comparison of KF and NIR predicted moisture results. All predicted results were within ± 0.2 % H2O (absolute difference) of the KF result, except for Vial 3, which was atypically high. The curve created would be most suitable for moisture levels between 1% and 4% based on the data points used to create the curve. Vial 3 was more than 6% moisture and did not fit the curve as well (absolute difference was 0.7% H2O).

	Case study: small-molecule lyophilized product

	A new residual moisture method was required for a small-molecule product containing ascorbic acid (vitamin C) and mannitol. Previous efforts had utilized coulometric KF and direct addition of the solid to the titrator with unacceptable results. Ascorbic acid is known to interfere with iodine in KF reaction (

	This interference may yield falsely high results, because the ascorbic acid is measured as water by the titrator and may be contributing to high variability of moisture results using the current method. Therefore, a new method for moisture analysis was developed to minimize exposure of the lyophilized product to ambient humidity and improve the accuracy and precision of sample data. 

	Various techniques for moisture analysis were evaluated with samples to evaluate compatibility with each method type. Samples were prepared for evaluation to contain varying levels of ascorbic acid while keeping other drug product ingredient levels constant, to confirm any impact of ascorbic acid interference on the results. Samples were equilibrated for a few days with stoppers removed in a desiccator filled with saturated potassium hydroxide (KOH) having a relative humidity of approximately 8.2% to ensure that all vials were equivalent in moisture prior to testing. Vials were stoppered after equilibration under conditions of low humidity (approximately 20%) in a nitrogen-purged glove box.

	Samples were analyzed by KF using anhydrous methanol extraction, KF Oven, and Vapor Pro. LOD was not evaluated as a potential method because it is not specific for moisture analysis. NIR was not pursued at this time due to low projected sample testing requirements. Vials containing the target amount of ascorbic acid were spiked with water to evaluate method accuracy. Only KF Oven and Vapor Pro were used to evaluated spiked samples due to the known interference of ascorbic acid with KF results.

	For KF titration, each vial was weighed (w

) and then reconstituted with 5 mL of anhydrous methanol.  The diluent weight was recorded (wd), and the vial was vortex mixed for 30 seconds to mix. Each sample was analyzed in triplicate. The vials were rinsed and dried, and the empty vial weight was obtained (w

). The moisture of each cake was calculated as shown in 

	For KF Oven, each sample vial was weighed (w

), and then reconstituted with 5 mL of anhydrous methanol (w

). Vials were vortex mixed for 30 seconds. Extraction was used with the oven sampler to avoid the interfering ascorbic reaction while also not having to open the product vial containing dried material. Approximately 1 g of this solution was transferred to a KF Oven vial and sealed. Each sample was analyzed in triplicate. Samples were analyzed using an oven temperature of 110 °C, nitrogen flow of 150 mL/min (2.0 psi), mix time of 1800 s, max time of 600 s, and end criteria of ≤ 5 µg/min. The mix time describes the time when the sample vial is lifted into the oven so that moisture begins to be released, but the nitrogen flow is bypassed so the titration does not start yet. The titration begins, nitrogen flows through the sample vial in the oven, and the titration starts. The titration ends when the maximum time or drift rate is reached. Sample vials were rinsed, dried, and weighed (w

). The moisture of each cake was calculated as above, using a vial filled with anhydrous methanol for a blank.

	For Vapor Pro, each sample vial was weighed (w

g). The lyophilized cake was broken up using a sample breaker (needle-like tool inserted through the stopper). Each vial was analyzed using a temperature setting of 110 °C and a nitrogen flow of 75 ± 2 mL/min. Stop criteria for the analysis included a minimum time of 10 minutes, followed by a rate criteria of ≤ 0.05 µg/s. After analysis, the vial was rinsed, dried, and weighed (w

g). The percent moisture of each cake was calculated by dividing the amount of water measured by the weight of the sample cake, in other words, (µg water/(w

)) x (1 g/106 µg) x 100. The entire vial is analyzed at once for Vapor Pro.

	Samples analyzed by KF using anhydrous methanol extraction were difficult to dissolve in methanol and showed a proportional increase in apparent residual moisture with respect to increase in ascorbic acid (

) demonstrating that KF is not suitable for the product. 

	Samples analyzed by KF oven with anhydrous methanol extraction were quite variable and even negative, likely due to very low moisture levels in the product which may have been below the sensitivity of the instrument (

). No trend in moisture relative to ascorbic acid content was observed. A diaphragm-less cathode must be used for KF oven instruments, which impacts instrument sensitivity.  Samples analyzed by Vapor Pro showed improved precision and no trend in moisture related to ascorbic acid content (

	The spiking study for the KF oven using anhydrous methanol showed that results were linear (R2 = 1.00), but not as accurate as expected (

). All recoveries were near 75%, supporting that some aspect of the spiking technique or measurement technique was impacting the result. Spiking of the methanol diluent instead of direct addition of water to the sample has been performed with success for other KF accuracy studies and may yield improved accuracy results here as well. Precision for triplicate measurements was acceptable at 2–11% (

	The spiking study for Vapor Pro showed that Linearity was acceptable (R2 = 1.00), and accuracy was also acceptable with recoveries of 92–104% (

). Precision could not be evaluated because only one vial was analyzed at each level, but previous Vapor Pro studies demonstrated acceptable precision.  

	Though the Vapor Pro method demonstrated better precision, accuracy, and linearity than the KF Oven method, additional method optimization and evaluation were performed for the KF Oven method in an effort to yield a suitable method that complied with 

	An extraction solvent consisting of anhydrous methanol:formamide (2:1) with KF oven analysis was found to yield more consistent results for samples, indicating that the samples may not have dissolved well in previous KF oven testing, leading to falsely low residual moisture values. The anhydrous methanol:formamide (2:1) mixture was used to repeat a spiking study for the KF oven.  Sample vials were equilibrated to constant humidity, and approximate moisture levels were evaluated by NIR before spiking to ensure that spectra for each vial matched.  

	The repeat spiking study for KF oven with anhydrous methanol:formamide (2:1) diluent was assessed for recovery/accuracy, precision, and linearity. Moisture for an unspiked sample was used to determine the spike recoveries for each level. Spike recoveries ranged from 95–127%. Precision ranged from 6–17%.  Linearity was evaluated by plotting the experimental spike recovery vs. the theoretical spike recovery.  Linearity was demonstrated, with an R2 value of 0.99 (

For this case study, suitable Vapor Pro and KF oven methods were developed for residual moisture analysis of the drug product containing mannitol and ascorbic acid (

). Method validation would be required prior to using either method for testing for GMP purposes. Ultimately, the KF oven method was chosen using the anhydrous methanol:formamide (2:1) diluent because this technique is described in 

	Residual moisture is a critical attribute of lyophilized drug product formulations, and suitable methods must be developed to accurately measure residual moisture for drug product characterization and stability. Methods should be specific for water, and coulometric KF, NIR, and emerging techniques such as Vapor Pro are favored over non-specific methods such as LOD and TGA.

	Method development must include evaluation of interfering reactions and sample solubility for KF methods and potential for heat degradation for oven methods. Sample analysis from an unopened lyophilized vial is preferable, and creation of a NIR calibration curve should be considered for larger projects where the additional work would be a worthwhile investment in the long term. All methods should be validated prior to use for GMP testing, including spike recovery studies.

	1. J. Towns, “Moisture Content in Proteins: Its Effects and Measurement,” 

, 705 (1) 115–127 (1995).
	2. M. Savage, “Determination of Adequate Moisture Content for Efficient Dry-Heat Viral Inactivation in Lyophilized Factor VIII by Loss on Drying and by Near Infrared Spectroscopy,” 

, 26 (2) 119–124 (1998).
	3. USP <1241>, “Water-Solid Interactions in Pharmaceutical Systems” (US Pharmacopeial Convention, Rockville, MD, 2019).
	4. J. May, 

, 10 (3) 249–259 (1982).
	5. USP <731>, “Loss on Drying” (US Pharmacopeial Convention, Rockville, MD, 2019).
	6. USP <891>, “Thermal Analysis” (US Pharmacopeial Convention, Rockville, MD, 2019).
	7. S. MacLeod, 

 63 (10) 557–566 (1991).
	8. Mettler Toledo, “Fundamentals of the Coulometric Karl Fischer Titration with Selected Applications,” 

Mettler Toledo Titrators DL32/DL39 Applications Brochure

 32 (2002)
	9. USP <921>, “Water Determination” (US Pharmacopeial Convention, Rockville, MD, 2019).
	10. Metrohm, “Water Determination by Karl Fischer Titration,” 

PDA Journal of Pharmaceutical Science and Technology

 56 (4) 196–205 (2002).
	12. USP <1119>, “Near-Infrared Spectroscopy” (US Pharmacopeial Convention, Rockville, MD, 2019).

	Elizabeth Joseph is research scientist II, BioPharma Solutions at Baxter Healthcare Corporation.


	Vol. 43, No. 11
	November 2019
	Pages: 30–39, 56

Articles

There are several common methods for detecting residual moisture in drug products that have been lyophilized; here are some key points to consider when selecting a method or developing one.

Detecting Residual Moisture in Lyophilized Product

The author provides an overview of common residual moisture methods for lyophilized drug products and key points to consider during method selection and method development.

Residual Moisture Determination in Lyophilized Drug Products

Aluminum compounds have been used as adjuvants in human vaccines for more than 75 years (1), and they are the most widely used adjuvants in both human and veterinary vaccines. Billions of doses of aluminum salts containing vaccines have been administered safely to a diverse population of patients globally (2). Despite their widespread usage, there is little regulatory guidance or published literature regarding release and stability testing of these products. Though there have been great gains in knowledge of therapeutic protein chemical and physical stability, there are few published examples of the applications of bioanalytical stability methods applied to protein once adsorbed on to the aluminum adjuvant. This article provides an introduction to aluminum adsorbed vaccines, reviews studies of antigen stability, and proposes a panel of test methods for the analysis of aluminum vaccine release and stability analysis.

	Early vaccines were prepared as alum precipitates, which could be heterogeneous. Preformed gels of aluminum hydroxide and, later, aluminum phosphate replaced alum precipitation and have resulted in standardized formulations of aluminum-adsorbed vaccines. Aluminum-based adjuvants continue to be used in new vaccine preparations as recombinantly expressed protein antigens are identified, which can serve as safer replacements for inactivated or attenuated pathogens.  Additionally, vaccines containing combination adjuvants linking aluminum with synthetic and natural ligands have been approved (2). The addition of adjuvants is often needed to induce a robust immune response and effective immunization to protein antigens, which may not possess strong immunogenicity. The physiological mode of action of aluminum adjuvants, however, is not completely understood (3). Potential mechanisms include that the aluminum forms a depot at the injection site from which the antigen is released slowly; that the particulate form of the adsorbed vaccine results in uptake by macrophages, neutrophils, and dendritic cells; and the immune system is stimulated directly. Additionally, it has been proposed that adsorption of the protein on the aluminum gel could result in structural destabilization increasing the protein’s immunogenicity (4).

	There are two common aluminum salt preparations used in human vaccines: aluminum hydroxide (Alhydrogel) and aluminum phosphate (Adju-Phos). The pH at which the net surface charge of the aluminum adjuvant is zero is called the point of zero charge (PZC), which is similar to the isoelectric point (pI) of a protein. The adjuvant surface is positively charged when the solution pH is below the PZC and negatively charged when the solution pH is greater than the PZC (5). Aluminum hydroxide has a PZC of approximately 11 and aluminum phosphate has a PZC of 4–5.5 (4), though consideration should be given to the buffer used in vaccine formulation as substitution of ions at the surface may lower the PZC and impact the adsorption of the antigen (5). For many proteins, antigen adsorption is best in pH intervals between pI of the protein antigen and PZC of the aluminum adjuvant because it is in this pH range that the protein antigen and the adjuvant will have opposite electric charges (1). For this reason, lysozyme, which has a pI of 11.35, has been used as a model antigen in studies of aluminum phosphate. Bovine serum albumin (pI of 4.7) and ovalbumin (pI of 4.5) bind more efficiently to aluminum hydroxide (4). Generally, Alhydrogel has a higher adsorption capacity than Adju-Phos, and the adsorption capacity of Alhydrogel decreases as its particle size increases and as the molecular weight of the proteins increase (6).

	The adsorption mechanism of proteins to aluminum adjuvants is not completely understood (6) and is likely a combination of physical phenomena including “electrostatic attraction, hydrogen bonding, apolar interactions, ligand exchange, and van der Waals’ forces” (4).  Studies of model proteins found that electrostatic interactions were the major mechanism for adsorption of lysozyme, human growth hormone, diphtheria toxoid, a monoclonal antibody, and PEGylated growth hormone, but that for one protein, ovalbumin, adsorption involved ligand exchange (6). The predominant mechanism for adsorption of HBsAg by aluminum hydroxide was determined to be ligand exchange between the phospholipids in HBsAg and the surface hydroxyls in aluminum hydroxide adjuvant (7). HBsAG was found to bind more tightly to an aluminum hydroxyphosphate sulfate adjuvant when the concentration of phosphate buffer was increased (8). The correlation of strength of adsorption to vaccine effectiveness is related to the adsorption mechanism. Phosphorylated proteins adsorbed via ligand exchange have an inverse relationship between high adsorption strength and poor antigen presentation and immune response (9). Studies of a trivalent Pneumococcal  protein vaccine demonstrated differences in immunogenicity resulting from selection of adjuvant impacted immunogenicity and found that lower microenvironment pH at the antigen surface and decreased strength of adsorption improved antigen stability (10).

	A significant challenge in vaccine formulation development is the understanding of the antigen stability following adsorption. While aluminum-precipitated and aluminum-adsorbed vaccines have been in use for 75 years, it has only been in the past 15 years that studies investigating the impact of adsorption on protein structure and stability have been reported, presumably as new techniques developed to characterize recombinant proteins were applied to vaccine research. In a review of the subject, Clapp and colleagues describe how the environment of the protein at the adjuvant surface differs from that in solution, potentially driving structural change (11). Specifically, studies of the rate of acid-catalyzed hydrolysis of glucose-1-phosphate (G1P) adsorbed to aluminum hydroxide concluded that the pH of the microenvironment at the surface of the adjuvant was approximately two pH units higher than that of the bulk solution (12). Differences in polarity and charge density on the adjuvant surface may also result in structural changes in the bound protein (11).

	Spectroscopic methods have been applied to the study of protein structural perturbations resulting from adsorption. Techniques have been well described in the literature and have included fluorescence spectroscopy (13), transmission Fourier-transform infrared (FTIR) spectroscopy (14), FTIR spectroscopy using an attenuated total reflectance (ATR) cell (13, 14, 15), and circular dichroism (CD) (15). Differential scanning calorimetry (DSC) has also been applied (13, 16). In studies of model vaccines, structural changes by FTIR–ATR, fluorescence spectroscopy techniques (13, 15), and DSC (13) were observed by following adsorption.  Studies using transmission FTIR (14) concluded that no structural changes resulted from adsorption in studies of six model proteins. Overall, a review of studies indicates that variation in the buffer concentration, buffer pH, and properties of the antigen itself may influence the structure of the adsorbed protein antigen. In studies of a hepatitis B vaccine, the interaction between the antigen and aluminum hydroxide adjuvant was modified by optimizing the phosphate ion concentration and could possibly ensure that the native concentration of the antigen was maintained (17). While no structural information was presented, the authors concluded that a formulation containing 40 mM phosphate had improved thermal stability by 

 antigen reactivity. Additionally, FTIR–ATR studies determined that the structural changes were dependent on the amount of protein adsorbed, with more native protein observed in samples with maximum adsorption (17). The authors theorize that at lower adsorption concentrations, a monomolecular layer is formed in which the confirmation of the adsorbed protein is affected by the properties of the adjuvant surface and are more denatured. At higher adsorption levels, the protein is more weakly linked and may retain the native structure. While the focus of aluminum-absorbed vaccine characterization studies has protein antigen structure, questions remain about the applicability of studies to individual vaccine formulations and the impact that antigen structural perturbations may have on vaccine effectiveness and stability throughout the shelf life of the product.

	Regulatory guidance for vaccine stability studies

	There is little general guidance on release and stability test methods for aluminum adsorbed vaccines. The FDA document, 

Guidance for Industry, Content and Format of Chemistry, Manufacturing and Controls Information and Establishment Description Information for a Vaccine or Related Product 

(18), states that identity, purity, and potency should be included in drug product release specifications but does not provide specific test methods or guidance. 

) Chapter <1235> Vaccines for Human Use–General Considerations (19) does not provide specific recommendations for release and stability tests beyond general requirements of testing for potency, general safety, sterility, purity, identity, and constituent materials. There was only one aluminum-adsorbed vaccine USP monograph available as an example (Anthrax Vaccine Adsorbed), but this was deleted as of August 2018 (20). The monograph includes tests performed on the sterile filtrate which does not include aluminum. Tests performed on the sterile filtrate include Identity by Western Blot, total protein by using a Bradford assay per 

<1057> (21), and 83kDa Protein using sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS–PAGE). Finished product tests are 

 relative potency, concentration determination of aluminum by atomic absorption spectroscopy, safety, sterility, pH, sodium chloride by ion-selective probe, benzethonium chloride by titration, and a limit test for formaldehyde using UV-Vis spectroscopy.

	While there is not specific guidance on the inclusion of test methods that are potency-indicating, there are parallels that can be drawn from International Council for Harmonization (ICH) Q5E 

Comparability of Biotechnological/Biological Products Subject to Changes

 in their manufacturing process (22). The guidance states that “the manufacturer should consider the limitations of biological assays, such as high variability, that might prevent detection of differences that occur as a result of a manufacturing process change.” However, it also states, “In cases where the biological assay also serves as a complement to physicochemical analysis (e.g., as a surrogate assay for higher order structure), the use of a relevant biological assay with appropriate precision and accuracy might provide a suitable approach to confirm that change in specific higher order structure has not occurred following manufacturing process changes.” While the intent of this analysis is to assess stability rather than manufacturing change, this guidance can be extrapolated to stability test methods. If potency methods lack sufficient accuracy and precision to detect small physical and chemical changes occurring during storage, additional methods may be required to provide early indication of loss of potency and expiry.

	Stanley Hem advocated for the application of preformulation studies, common in the development of biopharmaceuticals, to aluminum adjuvant containing vaccines (23). In addition to characterizing the adjuvant and the free antigen, Hem recommends evaluation of the antigen-adjuvant formulation to understand strength of adsorption as well as the structural and chemical stability of the protein antigens.

	Ideally, test methods to provide stability data for adsorbed vaccine products would not require extensive sample preparation or desorption of the antigen from the adjuvant. There are a variety of different mechanisms suitable for desorption but all share disadvantages. The desorption procedure itself may result in structural changes or additional modifications or impurities in the protein antigen. For analysis in most test methods, desorbed samples will require a buffer exchange step, which will result in solutions with an unknown concentration, which must be determined using a separate assay prior to analysis by stability indicating methods. Lastly, desorption recovery may vary and recovery can decrease in aged samples (15, 24).

	The measurement of unbound antigen by its definition is performed on finished drug product (FDP) without desorption. The intention of the test method is to measure the amount of free protein that was not adsorbed or has desorbed from the aluminum adjuvant. Samples are prepared by centrifugation, and the supernatant is removed for analysis. Typically, chromatography is utilized to quantify the amount of free protein in the sample against a standard curve of the same protein. Chromatographic methodologies vary and may include size-exclusion chromatography (SEC) (25) or reverse-phase high performance liquid chromatography (RP–HPLC). Typical methods can be based on those used for analysis of the antigen bulk drug substance (BDS), which are then optimized for sensitivity and validated using a low-level standard curve.

	Some immunological test methods are feasible in the presence of Alhydrogel or aluminum phosphate and may be used for identification or, in some cases, quantitation without desorption. While Western Blots are not a practical alternative because SDS–PAGE cannot be performed on FDP, slot blots and dot blots eliminate the separation step. In a slot blot or dot blot method, the product is applied directly to a membrane and may be pulled through the membrane using a microfiltration apparatus and vacuum. Additional rinses may be performed as needed to ensure no surface residue remains on the membrane. Following application of samples, standards, and controls to the membrane, the membrane is removed from the apparatus and immunoblotted using standard procedures. Though no published procedures were identified describing the use of dot or slot blots for the analysis of aluminum adsorbed vaccines, the technique was used to quantitate the Hemagglutinin and Neuraminidases in influenza vaccine (26). An advantage of dot and slot blots is that multiple samples can be analyzed simultaneously, and membranes can be cut into strips for immunoblotting using multiple antibodies allowing for the identification of multiple antigens simultaneously.  Zhu demonstrated the feasibility of a direct Alhydrogel formulation immunoassay (DAFIA) for the determination of antigen identity and content (27). In the method, the FDP is added to wells of a black multititer plate. Following centrifugation and washing, the wells are blocked, then probed with primary antibody followed by a secondary antibody tagged with fluorescein. Signal is read with a fluorometer at 485nm/535 nm.

	Concentration measurement of the aluminum adjuvant can be determined directly, using inductively coupled plasma atomic emission spectroscopy (ICP–AES). Traditional methods for protein quantitation such as UV-Vis spectroscopy or RP–HPLC are not suitable for bound antigens. Method 6 of 

 <1057> utilizing o-phthalaldehyde (OPA) (21) has been applied to Alhydrogel vaccines. The OPA reacts with N-terminal amine groups or the amine groups on the side-chains of lysine and results in a fluorescent signal at 340nm/455 nm. Protein concentration of the FDP is compared to a standard curve. The OPA method was applied to Malaria vaccine candidates containing Alhydrogel, and it was determined that for accurate results Alhydrogel must be included in standards at an equivalent concentration as samples (28).  Accuracy was 87–100%, and linearity, though protein dependent, ranged from 25–400 µg/mL. In internal studies, the method had a range of 20–250 µg/mL using a non-linear curve fit and resulted in accuracy of 98–110% and repeatability of ≤2% for samples (see a representative standard curve in 

	HPLC is not suitable for the analysis of intact aluminum-adsorbed FDP because the vaccine molecules would be filtered by the column frit or, if they do pass into the column, would result in column clogging and increased back-pressure. RP–HPLC has been applied to concentration measurement, and to a limited extent, chemical stability of vaccines after desorption (10). While a desorption step is required to perform chromatographic analysis of the intact antigen, peptide mapping provides a powerful tool to verify the identity of FDP as well as evaluate the chemical stability of the antigen following adsorption and during stability. In peptide mapping, the protein is digested with an enzyme such as Lys C or Trypsin to produce a set of peptides. The resulting mixture of peptides can be separated chromatographically and compared visually to a control or the BDS or it can be analyzed by mass spectrometry and compared to the expected masses of peptides based on the primary sequence. The peptide mixture can also be analyzed without separation using matrix-assisted laser desorption mass-spectrometry with time-of-flight detection (MALDI–TOF). Lys C digestion of the desorbed antigens followed by MALD–TOF analysis of a trivalent botulinum vaccine containing Alhydrogel detected oxidation and deamidation reactions in all three protein antigens and reaction rates were accelerated in the presence of the adjuvant (29). Initially a 250 mM succinate buffer, pH 3.5 was utilized to desorb samples, but following incubation, desorption became more difficult, and use of a 4 M Urea pH 7.5 buffer was required. The desorption itself complicates the analysis because, first, it may not be completely reproducible, especially for aged samples, and, second, it may result in additional degradation. In unpublished studies, trypsin was used to successfully digest the antigen bound to Alhydrogel without desorption. Standard digestion procedures of incubation in Tris/HCL buffer, pH 8.0 for four hours at 37 °C were followed by RP–HPLC using an ACN/TFA mobile phase and C18 column. Visually, chromatograms (

) from digested FDP closely matched those of digested BDS, indicating the non-desorbed peptide mapping is a feasible method for the evaluation of antigen stability in FDP. With the increased availability of high mass accuracy/high resolution mass spectrometers, peptide mapping has evolved into multi-attribute methods (MAM), which can replace multiple electrophoretic and chromatographic stability indicating methods for therapeutic proteins (30). The application of MAM to aluminum-adsorbed vaccines has the opportunity to further understanding of protein-antigen stability and improve analysis to support quality by design (QbD) development as well as quality control (QC) release.

	MAM has the potential to replace the electrophoretic and chromatographic stability indicating methods that all require desorption prior to analysis. Desorption methods range from the use of phosphate buffer pH 7.4 (25), phosphate buffer and Zwittergent (9), 250 mM succinate buffer pH 3.5 (29) to harsher conditions such as 20 mM sodium dodecyl sulfate (SDS) or 20 mM cetylpyridium chloride in citrate-phosphate buffer (31), and up to 4M Urea (29). Multiple investigators have noted that desorption is more challenging in aged samples (14, 24). Therefore, development of desorption procedures for stability indicating methods should be performed with aged samples when possible. The selection of desorption method will vary depending on the adjuvant used as well as the intended analysis method and compatibility with the desorption solution. For example, SDS would be a good choice for desorption buffer when preparing samples for methods utilizing SDS in the procedure, such as SDS–PAGE or CE-PAGE. Additionally, SDS does not significantly interfere with RP–HPLC procedures. Urea is often used in sample preparation for imaging capillary electrophoresis (ICE) procedures and is a preferred buffer for preparation of samples for charge separation methods such as ICE and ion-exchange chromatography (IEX). A standard protocol for desorption using SDS involved the following steps:

		Mix buffer, SDS, and vaccine FDP and incubate at 70 °C for 10 minutes.

		Centrifuge, remove supernatant, and determine protein concentration using RP–HPLC. Ensure that SDS concentration in the reference standard matches that in the samples.

		Concentrate using spin-filters and desalt with water wash. Utilize volume to determine approximate concentration prior to analysis.

		A standard protocol for urea desorption follows a similar procedure:

		Mix buffer, urea, and vaccine FDP and shake for 18 hours at refrigerated conditions.

		Centrifuge, remove supernatant, and determine protein concentration using RP–HPLC. Ensure that urea concentration in the reference standard matches that in the samples.

		Dialyze samples to reduce urea concentration.

		Repeat RP–HPLC analysis to determine concentration, concentrate using spin filters and desalt with water wash.

	Following desorption, SDS–PAGE can be performed using standard procedures and commercial gels and reagents. The analysis of a model vaccine pictured in 

 utilized a pre-cast gel (Thermo Invitrogen NuPAGE 4–12% Bis-Tris) with recommended sample and running buffers (NuPAGE LDS Sample Buffer and NuPAGE MOPS SDS Running Buffer). The gel shows results from analysis of SDS-desorbed model vaccine samples that had been stored at 37 °C for 0 to 5 days. Samples are reduced prior to loading the gel and stained using a Coomassie blue stain (NuPAGE Colloidal Blue Staining Kit). Gels were analyzed by densitometry (Bio-Rad GS-800 Densitometer with Quantity One software) for presence of additional bands due to degradation as well as % purity. Purity of the main band ranged from 96% for the Day 0 sample to 90–91% at Days 4 and 5, therefore demonstrating that the desorbed SDS–PAGE method is stability-indicating.

	ICE analysis of a model BSA-Alhydrogel vaccine was performed using an imaged capillary electrophoresis system with detection at 280nm (ICE280 and ICE3 systems, Protein Simple). Samples were prepared using urea desorption and dialysis as described previously and then concentrated using spin filters. Samples, reference material, and a control sample were diluted in a diluent consisting of a mixture of pharmalyte 4–6.5, pharmalyte 3–10, 1 g/mL urea, 0.7% methylcellulose, and pI markers 5.12 and 6.61. Analysis was performed using acidic anolyte (0.1% methylcellulose, 0.08 M H3PO4), basic catholyte (0.1% methylcellulose, 0.10 M NaOH), and separation for 10 minutes at 3000V. The resulting electropherograms were analyzed (ChromPerfect 7 software) for isoform purity. Samples of model vaccine were stored at 37 °C for 0 through 5 days prior to analysis. See 

for an electropherogram overlay showing the isoform profiles for each sample. The main peaks are observed at pI 6.1 and 6.2 in non-degraded samples, but a significant shift toward more acidic isoforms is observed with longer storage at 37 °C. The desorbed ICE method is stability-indicating for detection of changes in isoform profile.

	The nature of aluminum adsorbed vaccines presents challenges to the application of traditional bioanalytical methods commonly used for stability studies of therapeutic proteins. While studies have demonstrated structural and chemical modification of the protein antigen resulting from adsorption, and that adsorption may impact stability, little regulatory guidance is provided beyond the recommendation to include identity, purity, and potency in drug product release testing. Potency methods may not have the necessary accuracy and precision to predict vaccine effectiveness and stability throughout the shelf life of the product. Instead, a panel of tests adapted from traditional therapeutic protein analysis is proposed in 

 Though historically, antigen characterization was performed following desorption, most of the methods proposed can be performed on the FDP directly. As MS-based peptide mapping methods, such as MAM, continue to grow in popularity, the need for desorption followed by electrophoretic and chromatographic methods for the determination of fragmentation or chemical modification may be eliminated.

, 82 (5), pp. 497-505 (2004).
	2. D. Hagan, 

, 47, pp. 93-102 (2017).
	3. B. Lambrecht, 

, 96 (3), pp. 547-557 (2007).
	5.  S. Seeber, 

, 9 (3), pp. 201-203 (1991).
	6. M. Huang, 

, 466 (1-2), pp. 139-146 (2014).
	7. S. Iyer, 

, 22 (11-12), pp. 1475-1479 (2004).
	8. P. Egan, 

, 27 (24), pp. 3175-3180 (2009).
	9. B. Hansen, 

, 27 (6), pp. 888-892 (2009).
	10. B. Ljutic, 

, 30 (19), pp. 2981-2988 (2012).
	11. T. Clapp, 

, 100 (2), pp. 388-401 (2011).
	12. A. Wittayanukulluk, 

, 22 (9-10), pp. 1172-1176 (2004).
	13. L. Jones, 

, 280 (14), pp. 13406-13414 (2005).
	14. A. Dong, 

, 351 (2), pp. 282-289 (2006).
	15. Y. Zheng, 

, 21(4), pp. 211-226 (2007).
	16. C. Vessely, 

, 98 (9), pp. 2970-2993 (2009).
	17. J. Jezek, 

Guidance for Industry, Content and Format of Chemistry, Manufacturing and Controls Information and Establishment Description Information for a Vaccine or Related Product

(CBER, 1999).
	19. USP, Chapter <1235> Vaccines for Human Use- General Considerations, 

, In: Official Prior to 2013. s.l.:s.n.
	20. USP, Anthrax Adsorbed Monograph. 

, In: First Supplement Official as of August 1, 2018. s.l.:s.n.
	21. USP, USP Chapter <1057>Biotechnology Derived Articles- Total Protein Assay, 

, In: Official Prior to 2013. s.l.:s.n.
	22. ICH, Q5E 

Comparability of Biotechnological/Biologic Products Subject to Changes in their Manufacturing Process

, 28 (31), pp. 4868-4870 (2010).
	24. E. Zhu, 

, 30 (2), pp. 189-194 (2012).
	25. S. Jendrek, 

, 21 (21-22), pp. 3011-3018 (2003).
	26. C. Gravel, 

Quantitative Analyses of all Influenza Type A Viral Hemagglutinins and Neuraminidases using Universal Antibodies in Simple Slot Blot Assays

, 344 (1), pp. 73-78 (2009).
	28. D. Zhu, 

, 27 (43), pp. 6054-6059 (2009).
	29. T. Estey, 

, 98(9), pp. 2994-3012 (2009).
	30. R. Rogers, 

, 7 (5), pp. 881-890 (2015).
	31. J. Rinella, 


	Vol. 43, No. 9
	September 2019
	Pages: 42–52

	When referring to this article, please cite it as W. Saffell-Clemmer and E. Joseph, "Stability Indicating Methods for Aluminum Adsorbed Vaccine Products," 

The authors provide an introduction to aluminum adsorbed vaccines, review studies of antigen stability, and propose test methods for the analysis of aluminum vaccine release and stability analysis.