Freeze-dry microscopy of protein/sugar mixtures: Drying behavior, interpretation of collapse temperatures and a comparison to corresponding glass transition Data



The purpose of this study is to investigate the change in collapse appearance and temperature of protein/sugar mixtures as a function of nucleation temperature (Tn), sublimation velocity (Vsub) and the sugar/protein mole ratio when performing freeze-dry microscopy experiments. BSA and HSA were used as sample proteins and mixed with either sucrose or trehalose. Differential scanning calorimetry was used to determine the corresponding glass transition temperatures equation image. To allow a more representative comparison between these analytical methods, a collapse midpoint temperature (Tc-50) was introduced. While there was no distinct correlation between Tn and the onset of collapse (Toc) for either mixture, Vsub was found to correlate with the measured collapse temperature which is important for comparability of experiments. Furthermore, Vsub could be used to qualitatively investigate the product resistance to water vapor flow. A dramatic change in the appearance of collapse was found for high sugar/protein mole ratios (≥362:1) which needs to be considered to avoid a misinterpretation of Toc data. At low protein concentrations midpoint equation image data showed good agreement with Toc values but were found significantly lower at higher protein concentrations. Application of the Gordon–Taylor equation failed to predict the critical temperature for any of the protein/sugar mixtures studied. © 2008 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 98:3072–3087, 2009


Freeze-drying is still the method of choice to achieve improved stability of biopharmaceuticals when the product is not sufficiently stable in a liquid formulation. A well designed freeze-drying cycle forms a glassy solid which can minimize degradation reactions of complex structures such as proteins or peptides and may provide acceptable shelf life for worldwide shipping and storage.1, 2 It was estimated that about 200 new antibody products are currently in development, many of them in a lyophilized form.3, 4 However, formulation and cycle development becomes challenging if protein therapeutics require a large amount (>50 mg/mL) of active in the formulation to achieve the desired therapeutic effect.5 While highly concentrated protein formulations may even produce an acceptable cake without any excipients added, most proteins require at least one formulation component (i.e., a stabilizer) to assure stabilization during the process and/or acceptable long term stability during storage. Here, one of several stabilization mechanisms discussed in the literature is the “water replacement theory” which implies that a stabilizer (e.g., a disaccharide) serves as a substitute for the water molecules which form the protein's hydration shell.6–8 A review of the freeze-drying literature shows that sucrose and trehalose are the most commonly used stabilizers for protein pharmaceuticals,9–11 but they also need to be added in a sufficient quantity to achieve the desired stabilization effect. A molar stabilizer/protein ratio of 300/1 up to 500/1 was recently proposed.12, 13 Taking the above mentioned aspects in consideration (i.e., high protein concentrations and the need for additional excipients which remain in the amorphous state), an optimized freeze-drying cycle design might be difficult to develop and time consuming to run and requires substantial understanding of the drying behavior of the protein/disaccharide formulation.

Freeze-drying is generally considered a time consuming and therefore expensive process which can, in many cases, be attributed to unnecessarily long primary drying times when the cycle conditions are designed far from optimum.14–16 Recent emphasis within the FDA on manufacturing sciences and process analytical technology (PAT) encourages the pharmaceutical industry to further optimize and improve current freeze-drying processes and to design new cycles which are robust and economical from the very beginning.17 The only way to truly optimize a freeze-drying cycle is to control the product temperature at the sublimation interface close to (but not above) the “critical temperature” of the formulation; this critical temperature refers to the “collapse temperature” or “glass transition temperature” for amorphous or the “eutectic temperature” for crystalline systems.18 The collapse or glass transition temperature is much lower than the eutectic temperature for the corresponding crystalline structures. Even more important, if the product temperature during primary drying exceeds Tc (or equation image) shrinkage or collapse within the product matrix may occur which will then compromise product quality attributes; vials that show severe product shrinkage or collapse have a lack of product elegance, and were found to have elevated residual moisture levels as well as extended reconstitution times. Extended reconstitution time is normally caused by small amounts of undissolved solid remaining in the product solution19, 20 or instability (i.e., precipitation) of the active component. Indeed, Passot et al.21 recently reported that after 6 months of storage, and regardless of the protective medium used, the losses of antigenic activity of toxins A and B of Clostridium difficile increased from 0% to 25% when primary drying was performed at a product temperature higher than equation image. A similar observation on instability was reported by Chang et al.22 and Allison et al.23 for porcine pancreatic elastase and actin, respectively. However, no significant impact on protein stability was found for IgG and lactate dehydrogenase, both of which maintained their full activity even when the cake was completely collapsed.24 However, even when stability of the protein is not an issue, there is no doubt that a shrunken or collapsed cake lacks the elegance expected for a freeze dried product, and acceptance of such a product by the customer may be a serious issue.

In order to design an optimum freeze-drying process, a representative value for the maximum product temperature during primary drying is required. Two analytical methods have been commonly used for many years in freeze-dry formulation and cycle development to evaluate this critical temperature: differential scanning calorimetry (DSC)25 and freeze-dry microscopy (FDM).19, 25–27 While DSC has been used over several decades in pharmaceutical industry to characterize thermal transitions such as glass transitions (Tg), eutectic melting points (Teut) and the glass transition temperature of the maximally freeze-concentrated product equation image, FDM is arguably a better measure of collapse in a product that has been widely used in more recent years; here, a user visually measures via microscopy the collapse temperature (Tc) of a given product during primary drying. It is important to emphasize that the two technologies do not use the same experimental conditions to describe the physical property parameter, that is, the maximum allowable product temperature for primary drying. With DSC, it is an apparent glass transition temperature, equation image that is measured, which is commonly described as the glass transition of a maximally freeze-concentrated solution; this transition appears as an endothermic shift in heat capacity which arises from a decrease of the viscosity of the glassy structure in a small temperature range which in turn allows the system to access additional degrees of freedom. We do acknowledge an alternative interpretation of the equation image thermal event that has been discussed recently where equation image is said to arise from either a combination of ice melting and Tg28, 29 or ice melting alone.30 Important for the present study is, however, that during an typical DSC measurement, the amorphous matrix is in direct contact with ice, and the measurement itself is performed at atmospheric pressure with no drying during the measurement. In contrast, during FDM the sample is first frozen under atmospheric pressure, followed by evacuation of the FDM stage to a (controlled) pressure setpoint which is relevant for freeze-drying. Controlled heating of the sample is applied to allow the sublimation of ice from the product matrix. The microscopic collapse of the dried structure is determined as a function of visually recognizable structural changes in the dried product matrix over the observation time.19 At first glance, both methods seem to describe similar effects, that is, the decrease in viscosity of a glassy matrix (note that a decrease in viscosity per se does not necessarily result in thermal events). However, a Tc measurement is based on a dynamic process (i.e., during sublimation drying) which obviously is more representative of product behavior in a “real” freeze-drying process. However, the freeze-drying literature predominantly provides data for equation image. In a few cases, both Tc and equation image values for excipients or a final formulation are reported.9, 31 An important observation for freeze-drying cycle design is that collapse temperatures have been reported 2–5°C above the corresponding equation image for various systems;19 in short, the two methods do not necessarily measure the same property, and the difference is not negligible. In addition, information about the temperature “tolerance” of a given product to product temperature excursions exceeding the predefined critical temperature is helpful for robustness testing in process design, but such data are typically not available in the literature. Therefore, a systematic study and comparison of equation image and Tc data is of great interest with regard to process optimization as such comparisons provide an opportunity to further shorten the cycle time. As reported recently, an increase of only 1°C in product temperature during primary drying was found to shorten process time up to 13%.32

The present study illustrates, for the first time, a comprehensive investigation on appropriate FDM measurement methodology for evaluation of collapse temperatures. This work provides a new collapse temperature classification and reporting procedure, delineates collapse temperature dependency on experimental conditions, and describes for the first time a different collapse appearance found in protein/disaccharide systems. Human serum albumin (HSA), bovine serum albumin (BSA) and mixtures of HSA or BSA with either sucrose or trehalose at different mole ratios were used as a model to study the impact of nucleation temperature and rate of sublimation on the measured values of collapse temperature data for such formulations. This work further presents an improved classification procedure for observed collapse behavior. A direct comparison of Tc values with equation image data obtained from DSC measurements and a theoretical estimation by the Gordon–Taylor equation33 guide the discussion whose objective is the identification of the “maximum allowable primary drying temperature” such that a truly optimized freeze-drying process may be designed for modern protein therapeutics.

Materials AND MetHODS


BSA, HSA, sucrose and trehalose, potassium chloride (KCl), sodium chloride (NaCl) and magnesium chloride (MgCl2) were purchased from Sigma Chemical Company (St. Louis, MO). All excipients used were of highest available analytical grade. Water was double distilled from an all-glass apparatus and filtered through a 0.22 µL membrane before used (Millipore®, Billerica, MA). Liquid nitrogen was obtained from Linde (Nuremberg, Germany).

Experimental Procedure

Solutions of pure BSA, HSA, sucrose and trehalose, as well as binary mixtures of BSA and HSA with both disaccharides, were prepared with a constant total solid content of 50 mg/g. The ratio of the protein/disaccharide concentration was varied to obtain a different stabilizer to protein mole ratio (further reported in the text as sugar/protein mole ratio, according to Ref.12). Ratios were calculated by using the following molecular weights: sucrose (Mr: 342.3), trehalose (Mr: 342.3), BSA (Mr: 66,430), and HSA (Mr: 66,437), which were obtained from the certificate of analysis (Sigma Chemical Company). A summary of the individual sample compositions and the calculated mole ratios of the sugar/protein mixtures is provided in Table 1.

Table 1. Overview of Sample Solutions Investigated in This Study
Sample NameBinary Mixture (Protein/Sugar)c [Protein] (mg/g)c [Sugar] (mg/g)Mole Ratio, Sugar/Protein
  • Note that the total solid content for all sample solutions was fixed at 50 mg/g.

  • a

    The abbreviation “t” refers to the disaccharide trehalose, the abbreviation “s” to the disaccharide sucrose.

BSA or HSAPure BSA or HSA50.00.0
BSA/x (95/5)BSA/x = t or sa47.52.510: 1
HSA/x (95/5)HSA/x = t or sa47.52.510: 1
BSA/x (90/10)BSA/x = t or sa45.05.022: 1
HSA/x (90/10)HSA/x = t or sa45.05.022: 1
BSA/x (80/20)BSA/x = t or sa40.010.049: 1
HSA/x (80/20)HSA/x = t or sa40.010.049: 1
BSA/x (65/35)BSA/x = t or sa32.517.5104: 1
HSA/x (65/35)BSA/x = t or sa32.517.5104: 1
BSA/x (50/50)BSA/x = t or sa25.025.0194: 1
HSA/x (50/50)HSA/x = t or sa25.025.0194: 1
BSA/x (35/65)BSA/x = t or sa17.532.5362: 1
BSA/x (20/80)BSA/x = t or sa10.040.0776: 1
HSA/x (20/80)HSA/x = t or sa10.040.0776: 1
BSA/x (10/90)BSA/x = t or sa5.045.01747: 1
Trehalose (t)Pure trehalose0.050.0
Sucrose (s)Pure sucrose0.050.0

Freeze-Dry Microscopy (FDM): Technical Setup

The freeze-dry microscope system consisted of a Zeiss Axio Imager.A1 microscope (Carl Zeiss MicroImaging, Göttingen, Germany) with a lambda plate plus spectrum analyzer and a FDCS 196 freeze-drying stage (Linkam Scientific Instruments, Surrey, UK) with a liquid nitrogen cooling system and a programmable temperature controller. The magnification used was 200×. Pictures were captured in 1 s intervals by a digital camera system with a resolution of 1.3 mega pixels and analyzed using the LinkSys 32 software (Linkam Scientific Instruments). About 2 µL of the sample solution were placed on a thin glass cover slide (LAT Labor- und Analysentechnik GmbH, Garbsen, Germany) during each experiment, which was then placed on the silver block oven in the center of the freeze-drying stage. Note that about 5 µL of silicon oil was used to improve thermal contact between the silver block and the 20 mm glass cover slide. A second, smaller (10 mm) glass cover slip was then placed on top of the sample droplet. Custom-made, precision cut spacers (height 0.025 mm) were used to maintain the thickness of the sample layer constant.

Prior to the experiments, the temperature sensor (accuracy: 0.1 K) located beneath the surface of the silver block was calibrated by using well known eutectic melting temperatures of the three different salts KCl (Teut: −10.7°C), NaCl (Teut: −21.1°C), and MgCl2 (Teut: −33.6°C).34 Ten percent (w/w) stock solutions were prepared, and the eutectic melting point was measured in triplicate for each substance. The deviation to the reported values was within ±0.2°C. The collapse was then determined by using the following measurement routine: after an equilibration phase at 5°C for 1 min, the sample was frozen with a ramp rate of 10°C/min to −40°C. The sample was kept at this temperature for 10 min to ensure thermal equilibration of the product. After 8 min of equilibration, the stage was evacuated and the pressure was held as low as possible (<10 mTorr), except for the sublimation velocity experiments where the pressure was controlled as described in the text. Then, the sample was reheated by using a 1°C/min ramp rate through the point of collapse. During each experiment, pressure was determined using a calibrated Pirani gauge. All FDM measurements were performed in duplicate.

Freeze-Dry Microscopy (FDM): Collapse Temperature Classification

Throughout this work the collapse behavior was classified according to the observed (and calculated) three different phases of structural loss. The temperature at which first observable structural alterations occurred (e.g., holes, gaps or fissures in the dried structure adjacent to the sublimation interface) was denoted as the temperature of the “onset” of collapse (Toc). Note that most authors refer to this temperature as the “collapse temperature” when Tc values are published in the literature.35 Next, a Tc-50 value was newly introduced in this work. This parameter is used as a simple way to estimate a temperature where roughly 50% of structural loss in the product is present. For this purpose, Tc-50 was defined as the “midpoint” collapse temperature and calculated as an average value of Toc and the temperature of full collapse (Tfc) of the structure. At Tc-50, many holes or fissures form in the dried structure due to elevated viscous flow of the material. The overall structure, however, still appears to be coherent and dense. Since it is an established standard according to ASTM/IEC that equation image is reported as midpoint and not as an onset, the calculation of Tc-50 may facilitate a direct comparison with equation image data.25, 36 Note that the physical basis for collapse observed in a microscope is quite complex and cannot be easily correlated to glass transition temperatures. Again, the goal of introducing a Tc-50 is to have a measure of a “midpoint” temperature for a given FDM experiment, rather than to establish a direct physical comparison between equation image and Tc-50.

Finally, “full” collapse (Tfc) was defined as the temperature at which the product forms no coherent product layer adjacent to the sublimation interface right after the sublimation process. Here, most of the structure in the dried region degenerates and huge holes can be seen. In our experience, even chewing gum-like structures develop, that is, thin, highly viscous and stretchy filaments which connect the sublimation interface and the already dried product matrix. As a supportive argument to identify full collapse may serve an elevated elasticity of the dried product matrix (i.e., a vibrant behavior) close to the sublimation interface, in particular when high protein concentrations are present. A change in elasticity for proteins has already been described in the literature37 but a detailed discussion is certainly beyond the scope of this article. The reader is referred to the results and discussion section for a more detailed discussion of the Toc, Tc-50, and Tfc classification procedure.

Differential Scanning Calorimetry (DSC)

The glass transition temperature of the maximally freeze concentrated solute equation image of each individual sample was determined using a Mettler Toledo DSC822e (Mettler Scientific Instruments, Göttingen, Germany) and analyzed using the Mettler STARe Software (version 9.0). Twenty to 30 µL of each sample were sealed in 40 µL Al pans, cooled to −60°C at 10°C/min and then reheated at 10°C/min through the point of equation image. During all experiments, nitrogen was used to purge the DSC cell. equation image values were reported as “midpoint” temperatures throughout this work unless otherwise stated.


Nucleation Temperature and Corresponding Collapse Temperature of the Sample

Nucleation temperatures (Tn) for both the pure protein solutions and the binary mixtures with sucrose and trehalose were observed in a temperature range from approximately −9 to −20°C (cf. Fig. 1) which is in good agreement with the degree of supercooling found during laboratory scale freeze-drying experiments.1, 2 Performing replicate experiments under identical conditions (i.e., same solution but new preparation of experiment) clearly showed that Tn cannot be controlled during FDM experiments: differences in Tn between duplicate measurements of up to 7°C were found.

Figure 1.

Impact of freezing temperatures of BSA/sucrose (BSA/s) and BSA/trehalose (BSA/t) mixtures on collapse temperatures. Collapse temperatures are reported as onset temperatures (Toc). Symbols represent: crossed squares = Tn (experiment #1), crossed circles = Tn (experiment #2), filled squares = Toc (experiment #1), filled circles = Toc (experiment #2).

Differences in Tn are expected since it is well known that the formation of ice crystals during cooling is a random event, even in a well controlled process.38 The degree of supercooling, however, governs the rate of nucleation and therefore the number of ice crystals formed. In turn, the number of ice crystals significantly affects the product morphology during a freeze-drying run in vials and is known to be a scale-up issue in freeze-drying.38 A product with higher Tn tends to form larger ice crystals during the freezing step. As a result, the dried product matrix imposes less resistance to water vapor flow during primary drying and may dry at higher mass flow rates relative to a product with a low Tn.38 Based on previous reports in the literature, there is a reasonable basis for speculations that a difference in mass flow rate might impact the collapse temperature of a given structure during an FDM measurement. For example, Pikal and Shah19 predicted differences in collapse temperatures on the order of 1°C for systems subjected to major variations in freezing rate, based on calculations for formulated moxalactam di-sodium. This prediction was based on observations that differences in freezing rate produce differences in pore size on the order of a factor of two.19, 39 Therefore, a more in-depth evaluation of a potential correlation between Tn and Toc for the present protein/sugar system would be of great value.

Figure 1 also provides the corresponding Toc values found for each sample investigated, indicating no obvious correlation between nucleation temperatures and the value for Toc. For example, BSA/sucrose (20/80) showed a 6°C difference between two identical measurements, but the onset of the collapse temperature was observed at −29.2 and −29.5°C, respectively. In contrast, the BSA/trehalose (50/50) mixture showed an almost identical Tn for replicate experiments but the Toc determined differed by more than 2°C between runs. Similar observations were made with pure HSA and HSA/sugar mixtures in this study (data not shown). Conventional wisdom suggests that Tn of the product is predominantly determined by particulate contamination or, if the solution is sterile filtered, even the roughness of the container surface is important. In separate FDM experiments, solutions were sterile filtered before performing FDM experiments, and glass cover slides were thoroughly cleaned before using them. However, no change in the nucleation temperatures relative to unfiltered samples could be observed (data not shown). Further, the results in Figure 1 demonstrate that there is no direct impact of Tn on Tc which is consistent with recent studies which investigated the impact of the freezing rate (1 or 10°C/min) on collapse behavior of pure sucrose.35 Even if sample solutions in FDM and a subsequent freeze-drying cycle show supercooling in the same temperature range, the product morphology in a vial with typical fill depths >0.3 cm might not be comparable to an FDM experiment using 2 µL of solution and an effective sample thickness of about 2.5 × 10−3 cm. However, for the investigation of the collapse temperature of protein formulations and complex mixtures by FDM, a user must consider the possibility that high freezing rates may generate a less homogeneous frozen product matrix which may allow significant spatial variation in sublimation rates (i.e., velocity of the sublimation front). In addition, it should be noted that it is not advisable to use high freezing rates during FDM with formulations containing excipients which are expected to crystallize during the freezing step in a freeze-drying process (e.g., mannitol). Indeed, the rate of sublimation was found to have an impact on the collapse temperatures of a sample, which was attributed to a change in the observation time for viscous flow (cf. following paragraph) and not related to differences in product morphology.

Correlation Between Velocity of Sublimation and Onset of Collapse Temperature

Figure 2 illustrates the velocity of sublimation (Vsub, µm/s), determined for a BSA/sucrose 95/5 (upper plot) and a BSA/sucrose 50/50 (lower plot) mixture as well as the corresponding values for Tn, Toc, and Tfc. Vsub was evaluated by timing the movement of the ice–vapor interface as described earlier in the literature.19 Different pressure settings (0.075–1.5 Torr) were used during this set of experiments to alter the pressure gradient between sublimation interface (P0) and drying chamber (Pc) of the stage. It might be important to note that typical chamber pressures (Pc) used during freeze-drying are in the range of 0.05–0.2 Torr, but technical restrictions in older FDM equipment sometimes do not allow much lower pressures than 0.5 Torr. The reported data for Vsub were determined using a calibration scale, which is commercially available microscopy equipment, to determine the size of a given object. Vsub is reported as an averaged value of the sublimation velocity visually determined for the sublimation interface at temperatures just below the onset of structural changes in the product matrix. As a first approximation, this parameter may be taken as a qualitative indicator of the resistance of the product matrix to vapor flow during FDM experiments, that is, as a relative comparison of Vsub data for different formulations during formulation development. A user might also be tempted to compare these results with the time dependence of the receding ice layer thickness (Lice, cm) in a vial during a freeze-drying run in the laboratory.40, 41 However, such a comparison requires a thorough evaluation of the gas composition in an FDM stage during experiments.

Figure 2.

Interrelationship between collapse temperatures and velocity of sublimation (or drying rate19) during a freeze-dry microscopy experiment. Toc and Tfc of BSA/sucrose (95/5) and BSA/sucrose (50/50) mixtures were analyzed using different pressure settings during the microscopy experiment (0.075–1.5 Torr) and the speed of the sublimation front was determined graphically by a calibrated scale. The upper plot represents the BSA/sucrose (95/5) results using the following symbols: crossed squares = Tn, filled squares = Toc, open circles = Tfc, and stars = speed of sublimation (µm/s). The lower plot represents the BSA/sucrose (50/50) results. Here symbols are used as follows: crossed diamonds = Tn, filled diamonds = Toc, open diamonds = Tfc, and stars = speed of sublimation (µm/s).

While it is well known that the vapor in a freeze-dryer is mainly water vapor during primary drying,1 the vapor composition in the FDM stage used in this study (equipped with a gas control valve) may be a mixture of residual water and nitrogen, and therefore, the sublimation rate at a fixed temperature and given pore size might be expected to be slightly greater than in a vial undergoing freeze-drying in a normal process. However, the authors suggest that the overall impact of such a difference in gas composition on Vsub is small. Thus, a comparison of Vsub observed during an FDM experiment and Vsub in vials is instructive even if not exact. The results in Figure 2 indicate that Vsub is much higher for a BSA/sucrose (95/5) mixture (about 2.0 µm/s, Tp: −6°C) relative to a BSA/sucrose (50/50) mixture (between 1.5 µm/s at 0.075 Torr/Tp: −17.5°C and about 0.5 µm/s at 1.5 Torr/Tp: −22°C) for the pressure and temperature conditions used. While a higher Vsub at a given Pc was observed for BSA/sucrose (95/5), as expected due to a much higher Tp at the point of the Vsub measurement,18 the differences in Vsub (Fig. 2) between these two mixtures are surprisingly small. For example, assuming (1) the Pc in the FDM chamber is controlled at 75 mTorr (a pressure setting relevant for freeze-drying) and (2) Tp at the point of the velocity measurement can be used to calculate the vapor pressure of ice,18 the corresponding pressure differential (i.e., the driving force for sublimation, P0Pc) for BSA/sucrose (95/5) is about four times higher than for the BSA/sucrose (50/50). Since the sublimation rate, Vsub, is directly proportional to the pressure differential P0Pc and inversely proportional to product resistance,1, 18 one may expect roughly a four times higher Vsub at the higher temperature if product resistance is assumed to be identical for both mixtures. The effective difference in Vsub at this pressure is, however, only 0.5 µm/s, giving a ratio of only 1.33 (BSA/sucrose 95/5 = 2.0 µm/s, BSA/sucrose 50/50 = 1.5 µm/s, cf. Fig. 2). These results suggest the high protein content system has a higher resistance to water vapor flow than does the 50:50 mixture. We note the observation of similar behavior in recently published resistance data for a freeze-drying run in vials using 100 mg/mL pure sucrose and a mixture of 25 mg/mL BSA and 75 mg/mL sucrose at chamber pressures <0.1 Torr.41 Furthermore, while an increase in Pc to 0.375 Torr results in about a 50% decrease in Vsub for the BSA/sucrose (50/50) mixture, Vsub for the BSA/sucrose (95/5) formulation shows little sensitivity to pressure. Note that the decrease in Vsub for BSA/sucrose (50/50) is likely not a result of increasing product resistance, but rather is probably a result of Pc approaching P039 which is not the case for BSA/sucrose (95/5) due to a higher product temperature. In general, Toc for BSA/sucrose (95/5) mixtures appear to have less dependency on Vsub (cf. Fig. 2). It is important to note that Tn values were found to be similar in both mixtures. Unfortunately, no analytical technology (e.g., SEM, SSA, etc.) could be applied to reliably evaluate the pore size distribution of the thin (0.025 mm) cake produced by FDM. In addition, careful observation and analysis of Vsub during individual experiments clearly showed that some specific areas of the observed sample dry slightly faster than others. As already reported in the literature, observation time is a critical factor in determining reliable Toc data19 since the amorphous structure requires time to undergo viscous flow. A higher Vsub decreases observation time and therefore leads to a Toc shift to higher product temperatures. This effect becomes more important at higher disaccharide concentrations and lower pressures. In the pressure range of 0.075–0.375 Torr, Vsub and Toc data appear to decrease as pressure increases (temperature of observation was in the same temperature range) and remain at a constant level until 1.5 Torr. This impact on the Toc result was not observed for the (95/5) system. Therefore, for comparison and evaluation of Toc data in protein/sugar mixtures, it is advisable to also determine Vsub. This recommendation is in qualitative agreement with measurements performed previously on formulated moxalactam di-sodium19 where the authors indicated this trend in collapse temperature was only marginal (for the system studied). In the current study, however, the increase in pressure from 0.075 to 0.375 mTorr resulted in a 50% decrease in Vsub and a 4°C difference in Toc for the BSA/sucrose (50/50) mixture, which is significant.

Sugar/Protein Mole Ratio and Its Impact on Collapse

To maintain the highly porous, sponge-like product morphology during primary drying, the product temperature at the sublimation interface must be controlled below the collapse temperature of the product.18, 19 In particular for protein/sugar mixtures, Toc measurements and further interpretation of this transition are often difficult which is, in part, due to the inconsistency of Toc data in the literature. A standardization of appropriate measurement methodology and standardization of the interpretation of structural alterations in the product matrix are required for optimal application of this technique. Even more important, some recent reports have revealed that a given product may be dried well above the FDM measured collapse temperature without negative impact on product quality. For example, the evaluation of Tmicro (corresponding to Toc in this study) and Tmacro (corresponding to Tfc) and subsequent freeze-drying of a series of cytokine:sucrose systems of increasing protein concentration in a constant 25 mg/mL sucrose matrix showed that at product temperatures both below and well above Tmicro during primary drying, the resulting products showed comparable degree of shrinkage, moisture content and stability.42 In addition, the authors mentioned that interpretation of Tmacro is difficult, in particular at high cytokine content. A similar observation was made for monoclonal antibodies, which are frequently stabilized by lyoprotectants (e.g., sucrose or trehalose2) at a preferred mole ratio of 360:1. Indeed, in some cases, FDM observations do not always predict the loss of microscopic structure that is interpreted as collapse in a vial, for a variety of other reasons. Therefore, to allow a further optimization of a freeze-drying cycle (i.e., cut cycle time and increase product turnover), a more refined interpretation of the collapse behavior might be necessary and helpful. The following experiments and corresponding collapse data analysis provide an example of a more detailed interpretation procedure for protein/disaccharide mixtures, a procedure which should also be considered for formulations others than investigated in the present study.

Figures 3 and 4 illustrate Toc, Tc-50, and Tfc data for replicate experiments at various concentrations of BSA/sucrose and BSA/trehalose as well as HSA/sucrose and HSA/trehalose. The results obtained indicate that the mole ratio of sugar to protein becomes critical for collapse temperatures measured only when higher sugar concentrations are present in the mixture, that is, at sugar/protein mole ratios between 49:1 (protein/sugar weight %: 80/20) and 104:1 (protein/sugar weight %: 65/35). No significant difference in Toc was found among those binary mixtures where the concentration of the protein was extremely high. For example, an increase in either sucrose or trehalose up to 10 mg/g in a 50 mg/g BSA/sugar solution (mole ratio of 49:1) showed little effect on Toc (roughly a 2°C decrease in Toc), but was lowered significantly when the protein to sugar ratio was elevated to a weight % of 65/35 (decrease in Toc: ∼10°C). In addition, no difference between Toc and Tfc of HSA and BSA sugar mixtures can be seen, which might be expected given the similarity in these proteins. However, note that in mixtures with either protein, both sucrose and trehalose show Toc values in the same temperature range (±2°C) at a given mole ratio, confirming their equivalence in decreasing the formulation Toc in the mixture. The Toc of a pure 50 mg/g BSA solution is found at −5°C (−5.2 and −4.9°C for two replicate experiments) with an observable ice melt at −3°C. Note that due to the small temperature difference between Toc and ice melting temperature (Tm), Tfc could not be determined. However, the difference in structural change between the onset of collapse and the onset of melting could be visually distinguished clearly in this narrow temperature interval (see below). Surprisingly, the difference between Toc and Tfc was found smaller (about 3°C) and more consistent if one component was dominant in the mixture relative to intermediate mixing ratios. Here, the temperature difference was found to be as much as 6°C.

Figure 3.

Collapse temperature determination for various mixtures of BSA/sucrose (BSA/s) and BSA/trehalose (BSA/t), total solid content fixed at 50 mg/g. As a comparison, the collapse temperatures of pure BSA, sucrose and trehalose (all 50 mg/g) are reported as well. Experiments were performed in duplicate from the same stock solution (run #1, run #2). Symbols represent: filled squares = Toc (experiment #1), half filled squares = Tc-50 (experiment #1), open squares = Tfc (experiment #1), filled circles = Toc (experiment #2), half filled circles = Tc-50 (experiment #2), open circles = Tfc (experiment #2).

Figure 4.

Collapse temperature determination for various mixtures of HSA/sucrose (HSA/s) and HSA/trehalose (HSA/t) at a fixed total solid content of 50 mg/g. Experiments were performed in duplicate from the same stock solution (run #1, run #2). Symbols represent: filled squares = Toc (experiment #1), half filled squares = Tc-50 (experiment #1), open squares = Tfc (experiment #1), filled circles = Toc (experiment #2), half filled circles = Tc-50 (experiment #2), open circles = Tfc (experiment #2).

Furthermore, for both BSA and HSA mixtures, our study revealed clear differences in the visual appearance of collapse as sucrose or trehalose concentration increased (cf. Fig. 5). A robust measurement of Toc is dependent upon a correct interpretation of the morphological changes, which might be difficult for an inexperienced user. For example, the pure solutions of BSA and HSA and mixtures with little sugar content dry predominantly with fissures and cavities, even when keeping the temperature far below the collapse temperature (cf. Fig. 5A and B). When reaching the point of collapse, the transition may be described as “chewing gum-like” behavior, that is, a formation of thin, highly viscous and stretchy filaments. It is important to underline that the disruptions below Toc appear as rigid and fibrous like fissures in a structure similar to a lattice. In turn, in the presence of an elevated sugar concentration, Toc can be observed at much lower temperatures as a formation of “bright and distinct holes” which develop adjacent to the sublimation interface (cf. Fig. 5C and D). This discrepancy in the collapse appearance in the various mixtures may arise from the difference in the molecular weight of the component molecules. The product morphology formed from small sugar molecules undergoes viscous flow at a given temperature in a focused region close to the sublimation front. Here, the sponge-like product structure degrades locally (i.e., degradation of only a few pores which have a size of a few microns) which can be observed based on the increased transmission of light from the bottom light source through the product matrix. With increasing protein content, the collapse behavior changes and becomes more representative for the “chewing gum” behavior of the protein.

Figure 5.

(A) 50 mg/g pure BSA, onset of collapse was observed at a temperature of −4.9°C, magnification 200×. (B) 50 mg/g BSA/sucrose (90/10), onset of collapse: −6.2°C, magnification 200×. (C) 50 mg/g BSA/sucrose (50/50), onset of collapse: −24.0°C, magnification 200×. (D) 50 mg/g BSA/sucrose (10/90), onset of collapse: −29.5°C, magnification 200×.

Summarizing the observations made for high protein concentrations (i.e., sugar/protein mole ratio >49:1) in terms of measured collapse temperatures as well as collapse appearance suggests that a minimum disaccharide concentration is necessary (1) to obtain a distinct plasticizing effect of the protein/sugar mixture and (2) to observe a change in the collapse appearance which is then more representative for the disaccharide. Considering the difference in molecular size between the protein and the sugar, a plausible hypothesis is that at low sugar concentrations the smaller disaccharides molecules act as a “gap filler” between the protein molecules and may closely interact with the protein structure.43, 44 If the concentration of the disaccharide is still too low to fill the gap between the protein molecules, disruptions develop during an FDM measurement (Fig. 5A). In turn, this hypothesis would suggest that at a given mole ratio of sugar to protein the molecular interaction between a protein and a given number of disaccharides is depleted. To follow this line of arguments, a method is required to estimate the number of disaccharide molecules which can maximally interact with the protein structure. Imamura et al.45 recently proposed a quantitative evaluation of the hydration state of the model protein BSA in a freeze-dried sugar matrix and investigated the amount of sugar required to embed this model protein. The authors suggested that at high residual moisture content within the cake structure (which would be the case for the dried structure in the FDM experiment), the hydration water in the BSA molecule was substituted by the sugar (both trehalose and sucrose) and corresponded to about 20–25% of the hydration water for BSA alone.45 They attributed this finding to the about 20-fold volume of the disaccharide molecule relative to a water molecule, that is, steric hindrance. Note that BSA as well as HSA tends to form hydrogen bonds to sugar molecules rather than to water molecules.45–47 Furthermore, the monolayer water substitution (M0) for a protein structure can be determined according to the Pauling-Green Theory.47, 48 Here, moieties in the protein sequence can be classified into weak and strong binders for water adsorption. The knowledge of such residues can be used to estimate M0 in the absence of sorption data. For BSA and HSA, a total of 187 strong polar groups can be calculated from the molecular structure, based on the actual Swiss Prot data base sequence.49 As a rough estimate, the boundary for the mole ratio sugar/protein at which the sugar is expected to become more dominant for the Toc behavior would be ∼50:1, assuming (1) about 25% of effective interaction between BSA/HSA and the sugar and (2) a total of 187 strong binding sites as a basis for interaction. Note that this ratio is in excellent agreement with the results illustrated in Figures 3 and 4.

Toc Versus Midpoint Glass Transition and Gordon–Taylor Estimation

Three different approaches can be used to evaluate the maximum temperature at which a product can be dried during primary drying. The first approach is more of empirical in character and describes the relationship between the glass transition and the composition known as Gordon–Taylor/Kelley–Bueche equation.50 If densities of the components of the mixture are equal, the Gordon–Taylor equation may be simplified to the Fox equation.50 However, for most low molecular weight glass formers (e.g., sugars) and proteins, the Fox equation will not be satisfactory. The utility of the Gordon–Taylor equation is that by using literature equation image of the pure components, the equation image of the mixture may be predicted without a direct measurement on the mixture (cf.2, 25). The weakness of the Gordon–Taylor equation used in this manner is, however, that it is necessary to assume that the water content in the three component freeze concentrate is simply a weighted average of the water contents in the individual two component systems. Strictly speaking, water content in the three component freeze-concentrated solution (and water Tg) should be included in the equation to obtain equation image values for the three component mixture. However, for the mixtures studied, water contents in the freeze-concentrated solutions are unknown. Besides empirical calculations, the most commonly used methodology for determining the critical temperature of a given formulation is (modulated) DSC and the (direct) measurement of equation image.25 It is frequently reported that collapse temperatures by FDM were found at higher temperatures (about 2–5°C) relative to the corresponding equation image data.18, 19, 35 This temperature difference was attributed to (1) different heating rates during FDM and DSC experiments19, 51, 52 as well as (2) the difference in the measurement principle: collapse is the result of a viscous flow in the dried region adjacent to the ice over a time when water content is decreasing due to secondary drying, while equation image refers to a glass transition in the amorphous phase of fixed water content in contact with ice.19 Unfortunately, there is no literature reference available at the moment which reports Toc, equation image and data obtained from the Gordon–Taylor estimation for a given formulation composition.

Figures 6 and 7 illustrate the Toc, equation image and the Gordon–Taylor/Kelley–Bueche calculated estimates for BSA/sucrose and BSA/trehalose mixtures, respectively. The Gordon–Taylor calculation was performed by using (1) K = 0.25 which was calculated from the densities of the two components50 and (2) equation image data for BSA (−11°C),25, 53 sucrose (−32°C)35 and trehalose (−29°C)25 as reported in the literature. For DSC and corresponding glass transition data (heating rate: 10°C/min), the Gordon–Taylor calculation was found to greatly overpredict equation image for both BSA/sucrose and BSA/trehalose over the entire composition range. For FDM and corresponding Toc data, an overprediction for the critical temperature of the sugar/protein mixture was found up to a mole ratio of about 104:1 (protein/sugar weight %: 65/35), followed by an underestimation at higher protein content (cf. Figs. 6 and 7, open squares). This may be expected, since Toc data for the pure protein were found at much higher temperatures during our study than reported in the literature for equation image (equation image: −11°C vs. Toc: −5.1°C). Note that glass transitions for protein/sugar mixtures could only be identified up to a protein concentration of 32.5 mg/g in the 50 mg/g mixture (protein/sugar weight %: 65/35), using a 10°C/min heating rate. Even with high heating rates during the DSC experiment (>30°C/min), the step change in heat flow signaling the change in heat capacity at the glass transition was not sufficiently large to detect the protein equation image.2

Figure 6.

Impact of decreasing BSA content (in %, from left to right) in a BSA/sucrose mixture on Toc (average value, n = 2), equation image calculated by the Gordon–Taylor equation and equation image determined by DSC. Symbols represent: solid line = equation image predicted by Gordon–Taylor equation, filled squares = equation image (DSC), open squares = Toc (FDM).

Figure 7.

Impact of decreasing BSA content (in %, from left to right) in a BSA/trehalose mixture on Toc (average value, n = 2), equation image calculated by the Gordon–Taylor equation and equation image determined by DSC. Symbols represent: solid line = equation image predicted by Gordon–Taylor equation, filled squares = equation image (DSC), open squares = Toc (FDM).

equation image values by DSC were in excellent agreement with Toc data at low protein concentrations (i.e., lower than a protein/sugar weight % of 35/65, cf. Tab. 1). For BSA/sucrose mixtures differences are within ±1°C (cf. Fig. 6). However, an increasing bias in equation image and Toc values could be observed at higher protein content in the mixture. Toc data could be determined clearly for high concentrated protein mixtures (protein concentration: >32.5 mg/g) and even for the pure proteins BSA and HSA. We note that the measurement of a protein (or polymer) equation image is often difficult (or even impossible) by DSC due to an extremely weak step change in heat capacity at the equation image. Thus, one may tempted to predict glass transitions for proteins by extrapolating to zero excipient concentration using equation image values measured for binary mixtures of protein and another glass-forming excipient such as sucrose over a range of excipient concentrations.2, 53 The results obtained in Figures 6 and 7 suggest, however, that this procedure would lead to an underestimate of the transition temperature of a pure protein.

The Width of the Observed Collapse and Corresponding Glass Transition: A Comparison

The procedure most frequently used in industry during the development of a protein or peptide formulation for freeze-drying is to evaluate equation image by DSC as the critical benchmark temperature for the freeze-drying process. Indeed, a DSC has a wide range of applications and allows (besides a equation image measurement in the frozen state) also the characterization of the freeze-dried cake after the process. In addition, this technology can be used with auto samplers and therefore allows a higher throughput of samples than FDM. Considering the extra costs of a freeze-dry microscope, its limited applicability to formulation design and the time needed for a user to be able to perform FDM experiments, one may be tempted to base the process design exclusively on equation image information. Therefore, it is critical to know if a equation image for a given system is comparable to the corresponding Tc by FDM. Note that even if data are available from DSC and FDM for a given formulation, traditionally single temperature values are compared (i.e., onset of collapse by FDM and midpoint glass transition temperature by DSC). While this practice might greatly simplify a comparison of DSC or FDM data for routine formulation work, the information obtained might be insufficient for process optimization. For process design and optimization, one needs to know the maximum temperature tolerated by the product which is, for many cases, not necessarily the onset of collapse or midpoint glass transition temperature. For example, it was reported in the literature that cytokine/sucrose mixtures tolerated much higher temperatures than Toc during freeze-drying without additional shrinkage of the final cake.42 While it is not the goal of this article to show transferability of the temperature data obtained by DSC and FDM to freeze-drying in vials, the original method of data analysis for both DSC and FDM might be of value: that is, reporting a glass transition/collapse temperature as a temperature region rather than a fixed temperature.50, 51 The actual observation collapse event during an FDM experiment may reveal valuable information about the robustness of a structure to temperature changes. As mentioned earlier, most of the literature refers to the “onset of collapse” (i.e., the observation of the first visible changes in the product matrix) when reporting FDM data. This information is then compared to equation image data which is reported as “midpoint” temperatures. Note that “midpoint” represents a large viscosity change from the viscosity at the onset. There is typically a 3–6 orders of magnitude change in viscosity between the onset and end of the glass transition, at least at systems of constant composition.29 Here, ice melting provides a significant contribution to the observed DSC event. To allow a more reliable interpretation of the collapse behavior and facilitate data comparison between glass transition and collapse temperatures, some authors introduced a more complex procedure to describe the collapse phenomenon by FDM.35, 42 However, a detailed comparison between Toc data and “onset” equation image values for the same product and investigation of representative heating rates cannot be found in the literature.

equation image values (“onset,” “midpoint,” and “end”) and corresponding Tc data (Toc, Tc-50, and Tfc) for various BSA/sucrose and BSA/trehalose concentrations are illustrated in Figure 8. The width of the glass transition (from “onset” to “end”) for both BSA/sugar mixtures is found to be essentially constant over the entire concentration range (about 2.5°C), whereas the width of the collapse (from Toc to Tfc) steadily increases with increasing fraction of protein (0.5°C: low protein concentration, 3.0°C: high protein concentration). At low protein concentrations (mole ratio sugar/protein: 776:1, protein/sugar weight %: 20/80), equation image midpoint values were in excellent agreement with Toc data for BSA/sucrose, but were found to be lower for BSA/trehalose mixtures in most cases. In turn, the “onset” of equation image was found significantly lower than the corresponding Toc for all protein/sugar mixtures. The protein to sugar ratio had a large impact on the comparability of glass transition and collapse temperature data. For example, a protein concentration of 17.5 mg/g or higher in a 50 mg/g mixture leads to a much lower “end” temperature of equation image than the corresponding Toc. This temperature difference increases dramatically with further elevation of protein concentration in the mixture (cf. Fig. 8). These observations are important when deciding on an appropriate upper temperature limit for a given product during primary drying. As mentioned before, drying above the collapse temperature (i.e., Toc) was found to result in cake appearance similar to drying well below Toc.42 Using a Tc-50 value for comparison might (at least in part) account for this apparent problem in transferability; that is, for the majority of protein/sugar mixtures the Tc-50 is (on average) 2°C higher than the Toc at a low, 1°C/min ramp rate. In addition, use of Tc-50 would allow one to quantify the effect of elevated temperature on product shrinkage or collapse. Thus, we tentatively conclude that Tc-50 should be used in place of Toc in characterizing collapse which is also consistent with the common practice of reporting “midpoint” temperatures for glass transitions. We suggest that, in general, process optimization should not be performed using the more conservative Toc data as the “representative collapse temperature” because in most cases the product withstands higher process temperatures. This issue will certainly be clarified in the future when Toc, Tc-50, and Tfc data are used for process design in a freeze-dryer with subsequent analysis of the overall shrinkage of the product.

Figure 8.

Comparison between Toc, Tc-50, and Tfc obtained from FDM and equation image (onset of equation image, midpoint and end of transition) by DSC: BSA/sucrose and BSA/trehalose mixtures. Symbols represent: open squares = Toc (1°C/min), open diamonds = Tc-50 (1°C/min), open stars = Tfc (1°C/min), filled squares = equation image “onset” (10°C/min), open circles = equation image “midpoint” (10°C/min), open squares = equation image “end” (10°C/min). Note that Tc-50 is calculated from Toc and Tfc.


In this study, collapse temperatures and their visual appearance during an FDM experiment were evaluated for various protein/sugar mixtures. BSA and HSA were used as sample proteins with varying concentrations of either sucrose or trehalose. The present results indicate it is advisable to report the temperature interval for glass transitions measured by DSC as well as for collapse events measured by FDM, in particular if this information is used for freeze-drying cycle design and optimization. Low protein concentrations showed good agreement between onset of collapse (FDM) and midpoint equation image (DSC), whereas a dramatic difference between FDM and DSC information was found for sugar/protein mole ratios of 362:1 (protein/sugar weight %: 35/65) or a lower presence of disaccharide in the mixture. Glass transitions could only be determined up to a protein/sugar mixture of 35/65 (weight %) using a 10°C/min heating rate. By contrast, even a 1°C/min heating rate during the FDM experiments allowed the determination of the onset of collapse for the pure proteins. The experiments further revealed that the collapse appearance dramatically changes with increasing protein concentrations. This is a significant observation which may greatly impact FDM methodology; it is easier for an inexperienced FDM user to evaluate the onset of collapse for sugar rich formulations. In turn, the collapse appearance for protein rich formulations changes to a “chewing gum” appearance associated with fissures and cavities. These fissures might falsely be interpreted as the onset of collapse. The collapse event during an FDM measurement should be observed and studied itself, and the onset as well as the full collapse temperatures need to be reported to gain more insight into product behavior at critical temperature conditions. The introduction of a 50% collapse (Tc-50) calculated from the onset and the full collapse was found to be more representative of a “critical” formulation temperature for protein/sugar mixtures than is the onset of collapse. Additionally, the present investigations revealed that the velocity of the sublimation front may also impact the results obtained for collapse temperatures and must be taken into consideration when FDM data are compared with each other (i.e., when replicate experiments are performed and greater differences are found for the same sample solution). Based on the results obtained, we suggest that future work needs to focus on transferability of collapse temperature data obtained by FDM to freeze-drying cycles, which would be of great value in supporting cycle optimization and robustness testing in freeze-drying.


The authors would like to thank Dr. Michael J. Pikal for his active participation in discussions for bringing understanding to the results obtained during this study.