α-CT was chosen as the model protein since it has been employed prior to this investigation by us in solid-phase stability studies and also as model protein to test drug- delivery systems [1,30,31]. α-CT is prone to undergo aggregation and thus is an excellent sensor for aggregation-related instability events. Our experimental approach was to expose solid α-CT formulations to accelerated storage conditions (i.e., high temperature) under controlled humidity conditions as it is common in the field [7,10,32]. We then monitored various stability related parameters as a function of time, i.e., loss in monomer concentration (by SEC-HPLC), residual specific activity of soluble α-CT, formation of buffer-insoluble aggregates, tertiary structure intactness of soluble α-CT (by CD spectroscopy), and structural dynamics (by FTIR H/D exchange).

To investigate the aggregation of α-CT at different RH, formation of insoluble aggregates was measured by solution-depletion experiments after incubation at 50°C and at 11%, 51%, 75%, 81%, and 96% RH for up to 3 weeks. Water-sorption isotherms demonstrated that all samples were completely equilibrated after 12 h (Additional file 1). All α-CT samples exhibited significant solubility loss during incubation due to aggregation (Figure 1). In contrast, α-CT stored at -20°C remained completely soluble in buffer. A control stored at 0% RH and 50°C did not form more than 2.1% aggregates. Thus, the detrimental process occurred during the high temperature/humidity incubation of solid α-CT, and not during lyophilization. Long-term storage of the protein at increasing levels of RH increased the formation of aggregates with this effect also increasing with storage time (Figure 1). Its aqueous solubility plummeted in particular after various days of incubation at 96% RH.

More than 90% of the aggregates were dissolved in 6 M urea which indicates that the α-CT molecules in the aggregates were held together by non-covalent interactions.

The Kruskall-Wallis test was employed to statistically investigate whether aggregate formation was significantly different at the various RH employed. The analysis revealed that there are marked differences in the amount of aggregates formed at the various RH employed (p < 0.05). To compare the effect of the RH on aggregate formation quantitatively, we performed a post hoc test (Dunn's test). This test revealed that aggregation was markedly different at RH of 11% and 51% RH when compared to 75%, 81%, and 96% RH and at 75% RH when compared to 96% RH. Furthermore, statistical analysis revealed that there was a significant difference in the formation of aggregates at various incubation times at each RH. Aggregation was markedly different after 24 h of incubation when compared to the samples stored for 1 and 2 weeks at RH of 51%, 75%, 81%, and 96%.

To study the kinetics of α-CT monomer loss, we investigated the water soluble fraction after dissolving the solid samples in buffer by SEC-HPLC. The amount of monomeric α-CT was measured based on the area of the corresponding HPLC peak (see Supporting Information for details) and the fraction of residual soluble monomer was plotted as a function of time (Figure 2). Two degradation pathways were identified: formation of soluble aggregates and fragmentation of the enzyme into subunits [22]. The kinetics of the monomer loss can be modeled using the following equation:

(1) X = A 1 exp ( − k 1 t ) + A 2

where X is the fraction of residual soluble monomer and A1 represents the population of protein undergoing aggregation or fragmentation with the rate k1, and A2 is the population that remains unaffected.

The data obtained reveal that monomer loss was not as pronounced at 11% and 51% RH as at higher RH (Table 1, Figure 2). Between 51% and 95% RH monomer loss increased at increasing RH level. The differences observed when compared to formation of buffer-insoluble aggregates at the various RH were more pronounced. This can be understood because more than one off-pathway leads to monomer loss and in particular at high RH values fragmentation becomes a significant degradation pathway.

In general, the instability data parallel the water-sorption at the various RH (see Additional file 2 for details). Water sorption becomes significant at 51% RH and then steadily increases.

To determine protein structural changes induced during incubation, near-UV-CD spectroscopy was employed to examine the tertiary structure of the soluble fraction of the protein. After incubation the protein was redissolved in 10 mM phosphate buffer at pH 7.3 and 25°C and the solution was filtered to obtain the soluble fraction. The CD spectrum of α-CT displays two maxima at 288 and 297 nm (Figure 3) which completely disappear upon thermal unfolding. The data compiled demonstrate that structural changes were significant for almost all RH after a short period of storage (i.e., 24 h). This effect increased markedly at increasing RH, being most significant for the samples stored at 96% RH. After 1 week of incubation only the samples stored at 11% and 51% RH retained more than 60% of native structure. The data parallel those obtained with SEC- HPLC on monomer loss.

Non-invasive FTIR microscopy was conducted to investigate the secondary structure of the samples during incubation using diamond windows [33] rather than pressing the powders into KBr pellets which is usually done to investigate the structure of lyophilized proteins [34,35]. FTIR analysis was performed in the amide III spectral region because water and water vapor have an insignificant contribution in this spectral area [34]. The amide III spectral region has the advantage that structural transitions can be monitored easily without spectral deconvolution because the absorptions of α-helix (ca. 1330-1290 cm^-1), other, and β-sheet (ca. 1250-1215 cm^-1) secondary structure are well separated [34]. First, we obtained FTIR spectra of the lyophilized powder using the usual KBr-pellet method [34] and by FTIR microscopy [33]. The spectra obtained were very similar and demonstrate that the method of FTIR spectral acquisition and sample preparation did not influence the result substantially in agreement with literature data [33].

Qualitative analysis of the spectra revealed that some spectral changes occurred in the amide III region after incubation of all powders under the RH conditions described. The shape of the spectra obtained even at 11% RH is somewhat different from those obtained without incubation. A slight increase in FTIR absorption is visible in the region corresponding to α-helix structure (ca. 1330-1290 cm^-1), while β-sheet structure contributions to the spectra (ca. 1250-1215 cm^-1) decreased. There is also more non- repetitive secondary structure present as indicated by the relative increase of the component around 1260 cm^-1. We can conclude from this that some structural changes occur to αCT powder upon exposure to even low levels of moisture. Incubation of the samples for 96 h at different RH resulted in marked spectral changes characterized by an increase in the components around 1230 cm^-1 indicating an increase in the β-sheet content while the α-helix content suffered a drastic decrease even for the samples incubated at the lowest RH (11%). Quantitative analysis by calculation of the spectral correlation coefficient (SCC) also demonstrated that moisture sorption induced structural changes in αCT even at low RH (Table 2). A value of 1 demonstrates spectral and thus structural identity while lower values indicate spectral and thus structural changes. Again, as in all other tests, the magnitude of structural changes increased at increasing RH.

From the results presented thus far it is evident that structural changes occur to the solid protein formulations even when exposed to low RH levels and that structural changes become more pronounced at increasing RH levels. This leads to the hypothesis that moisture sorption increases structural dynamics in the lyophilized formulation.

Amide H/D exchange measurements have been used for over four decades to investigate protein dynamics. Despite of the compact folded structure of most proteins, there are solvent exposed amide bonds that exchange rapidly, but the amide bonds that are inaccessible to the solvent or are participating in stable hydrogen bonds must undergo temporary structural distortion to allow for hydrogen exchange and thus have significantly slower exchange rates. This decrease in the exchange rates of buried amide bonds in a folded protein makes H/D exchange an excellent and sensitive probe for monitoring protein conformational dynamics processes [23,25,36].

In order to study the global structural dynamics, H/D exchange kinetics were measured by FTIR spectroscopy by following changes in the amplitude of the amide II band. The amount of amide hydrogen exchange (HX) was plotted in the form of HX decay plots where the fraction of unchanged amide hydrogen atoms (X) decreases over time due to the exchange process (Figure 4). Visual inspection of the Figure 4 demonstrates that at increasing RH level a decrease in the fraction of slowly exchanging amide bonds is observed. This demonstrates reduced structural rigidity of the enzyme at increasing RH values. A quantitative analysis of the exchange kinetics was attempted but led to insufficient fitting results because the fast exchange component could not be described appropriately.

Activity of α-CT after incubation at different RH was determined using n- succinyl-Ala-Ala-Pro-Phe-p-nitroanilide as the substrate [23]. The activity of α-CT was reduced depending on the RH and the time the protein was exposed to it. Exposure to increased RH caused a more pronounced activity decrease (Table 3). The activity decrease parallels the loss of native-like structure observed by CD spectroscopy.

Combining all data obtained thus far reveals the following likely scenario compatible with all observations: moisture sorption increases the structural dynamics of the protein which in turn enables structural changes leading to protein fragmentation, inactivation, and aggregation as the end result. All detrimental events increase at increasing RH during incubation and reach their maximum at the highest RH tested, namely 96% RH. Based on this we developed the following working-hypothesis: if it were possible to prevent water-protein interactions by shielding the protein surface and thus prevent an increase in structural dynamics, it should be possible to minimize or prevent the subsequent detrimental events. One method previously found to accomplish this goal is through the chemical glycosylation of the protein surface [21-24]. Solá et al. found through experimental and computational studies that chemically bound glycans can locally shield the protein surface from the interaction with the solvent (water) electrostatics thus reducing structural dynamics and increasing protein thermodynamic and colloidal stability [21,24]. These effects were mainly related to a reduction of solvent dielectric shielding effects within the proteins internal core [24].

Based on the developed working hypothesis, we next performed chemical glycosylation using chemically-activated glycans as previously described by Solá et al. [23]. Glycans of differing length (500 Da lactose and 10 kDa dextran) were employed since it was previously found that for different instabilities both the amount and the length of glycosylation had impacts on stability. For example, while thermodynamics effects were found to depend only on the amount of glycosylation, colloidal stabilization effects were also found to depend on the size of the glycans.

Glycans were covalently attached to α-CT via the reactivity afforded by the surface lysine residues (ε-amine groups) towards a succinimidyl functionalized linker. Synthesis conditions were adjusted to achieve an average number of glycan molecules bound to the protein between 4 and 7 since maximum thermodynamic and colloidal stabilization effects in solution were previously reported for these constructs by Solá et al. [21-24]. Since the protein has 14 lysine residues, these modifications correspond to 30% and 50% of the total glycan content that can theoretically be attached.

To test the efficiency of the glycoconjugates, they were than incubated at 75% and 96% RH were most detrimental changes were observed. It was found that glycosylation was efficient in reducing the moisture-induced aggregation of α-CT (Figures 5, 6). It is evident that this effect was more pronounced at 75% RH (~25% reduction) than at 96% RH (~20% reduction). Overall both sugars were similarly efficient in preventing aggregation despite of the major difference in molecular weight. Similarly, glycosylation was also efficient in preventing activity loss (Tables 4 and 5).

Since it is evident from the results that stability was imparted regardless of the size of the glycan attached to the protein, stability improvements were probably not due to the glycans working as molecular spacers preventing aggregation by keeping (unfolded) protein molecules at a distance. In contrast to liquid storage stability under accelerated conditions in which dextran conjugates were much more efficiently protected from aggregation than lactose conjugates [22], no difference was found in the improvement of solid-phase storage stability herein.

However, glycans interact with the protein surface and this interaction is limited to a small portion of the glycan molecule close to the protein surface [24]. The glycan- protein interaction leads to a shielding of protein surface area from water and thus the protein dynamics and structure becomes more conformationally restricted [22,24]. The most likely mechanism by which glycans improve the solid-state stability of lyophilized α-CT involves this mechanism. In this work, we have presented evidence to support that moisture sorption leads to an increase in protein structural dynamics which causes structural changes of the protein in the lyophilized state. These structural changes then trigger partial/complete unfolding and aggregation of the protein. Reduction in the structural dynamics by glycosylation should therefore reduce these deleterious pathways.

α-CT (EC 3.4.21.1, type II from bovine pancreas) and Dextran (10 kDa) were purchased from Sigma-Aldrich (St. Louis, MO); mono-(lactosylamido)-mono- (succinimidyl) suberate (SS-mLAC); 2,4,6-trinitrobenzene sulfonic acid (TNBSA) were from Pierce (Rockford, IL) and succinyl-Ala-Ala-Pro-Phe-p-nitroanilide was from Bachem (King of Prussia, PA). All other chemicals were of analytical grade or purer and were obtained fromvarious commercial suppliers.

α-CT was dissolved in deionized water (0.5 mg/mL, pH 7.3). These samples were frozen in liquid N₂ and lyophilized for 48 h using a Labcono FreeZone 6L freeze drier at a condenser temperature of -45°C and a pressure of <60 μm of Hg. The final moisture content of the lyophilized powders was 5.4 ± 1.5% (w/w) as determined by Karl Fisher titration with a Metrohm 831 KF Coulometer. After lyophilization and redissolving the samples, no change in pH was noted.

Synthesis of mono-(dextranamido)-mono-(succinimidyl)suberate (10 kDa) (SS- mDEX) was carried out as previously described by Solá et al. [23] In brief, dextran (10 kDa) (10.0 g, 1.0 mmol) and ammonium carbonate (2.0 g, 21.74 mmol) in nanopure water (50 mL) were stirred at 10°C for 5 days. Afterwards ammonium carbonate was removed by dialysis and the solution was lyophilized to afford 1-amino-dextran (10 kDa). To achieve dextran succinnylation, 1-hydroxy benzotriazole (0.675 g, 5.0 mmol) and disuccinimidyl suberate (1.841 g, 5.0 mmol) were dissolved in DMSO (60 mL) and heated to 80°C for 5 min. After cooling, 1-amino-dextran (10 kDa) (5.0 g, 0.5 mmol) was added and the reaction maintained at 20°C for 24 h. The product was precipitated by addition of CH₂Cl₂ (200 mL) followed by centrifugation (1,500 rpm) at 48°C for 15 min. The resulting white precipitate was lyophilized to afford SS-mDEX in quantitative yield. Reaction products were characterized by colorimetric ninhydrin and orcinol tests [22,23].

Covalent modification of α-CT with lactose and dextran conjugates was performed by chemical glycosylation with SS-mLAC (500 Da) and SS-mDEX (10 kDa) as described by Solá et al. [23]. To achieve protein conjugates with varying glycan molar content different amounts of SS-mLAC and SS-mDEX were added to α-CT solution to achieve molar ratios of 4.5 and 7.1 mol of reagent per mol of protein in 0.1 M borate buffer, pH 9.0. Reaction mixtures were stirred at 4°C for 2 h. Unreacted glycans and buffer salts were removed by dialyzing against deionized water. The glycoconjugates were subsequently lyophilized and stored at -20°C until use. The degree of protein modification was determined by colorimetric titration of unreacted amino groups with TNBSA [23].

To study the effect of moisture on α-CT and glycoconjugate solid-phase stability, we exposed the lyophilized powder (0.5 mg) in an Eppendorf tube to accelerated storage conditions, i.e., controlled humidity and high temperature. Incubation of the lyophilized powders at constant levels of RH was performed by vapor equilibration in dessicators over salt slush: LiCl (11% RH), NaBr (51% RH), NaCl (75% RH), MgCl₂ (81% RH), and K₂SO₄ (96% RH) [13]. The desiccators were kept in a LabLine incubator at a constant temperature of 50°C. These chambers were left for 48 h closed to reach equilibrium. Following the desired length of time, the incubated protein sample was removed and 1 mL of phosphate buffered saline (PBS, 10 mM pH 7.3) and a magnetic stirrer was added followed by 2 h of gentle stirring to ensure dissolving the soluble fraction of the sample. Soluble and aggregated protein were separated by centrifugation in a microcentrifuge at 9000 rpm for 10 min. To determine the amount of aggregated protein 1 mL of 6 M urea was added to the pellet and left stirring overnight. The protein concentration in the resulting clear solution was determined by absorption measurement at 280 nm. The experiments were performed in triplicate and the results averaged. Error bars in the figures are the calculated standard deviations.

The enzymatic activity of α-CT and α-CT glycoconjugates was determined by measuring the rate of hydrolysis of the substrate succinyl-Ala-Ala-Pro-Phe-p- nitroanilide [23]. Product formation (p-nitroaniline) was followed by measuring the formation of p-nitroanilide at 410 nm (ε410 = 8.8 mM-1 cm-1). Reactions were carried out in 10 mM potassium phosphate buffer (pH 7.1, 25°C). All reactions were started by addition of 60 μL of enzyme ([E]₀ = 0.8 μM) to 240 μL substrate solution ([S]₀ = 0.8 mM) in a final volume of 1 mL. Initial velocities were obtained from the linear portion of plots of the concentration of product formed vs. time.

Lyophilized samples of α-CT (0.5 mg) were prepared at pH 5 and stored in the humidity chambers at a constant temperature of 50°C under accelerated storage conditions. These samples were removed at various times and dissolved in 10 mM PBS at pH 7.3, and filtered through a Millex-GV filter (Millipore) of 0.22-μm pore size to remove insoluble materials, and then subjected to SEC-HPLC using a Perkin Elmer ISS 200 HPLC and an Applied Biosystems 759A detector. The column used was a Tosoh Bioscience TSKgel SuperSW3000 (4.6 mm × 30 cm, 4.6 μm) maintained at 25°C. Elution was performed using 50 mM NaH₂PO₄ containing 100 mM NaCl, pH 7.0, at a flow rate of 0.5 mL/min. The detection wavelength was 280 nm (see Figure 1 of the Supporting Information). The soluble fraction was determined from the monomer peak area before and after different incubation.

Near-UV circular dichroism measurements were acquired with an Olis DSM-10 UV-Vis CD spectrophotometer. The protein concentration was adjusted to 0.6 mg/ml in 10 mM potassium phosphate buffer at pH 7.3 and 25°C. Spectra were recorded from 260 to 320 nm using a 1.0 cm path length quartz cell. Each spectrum was obtained by averaging six scans at 2 nm resolution. Solvent reference spectra were digitally subtracted from protein CD spectra.

FTIR-microscopy studies were performed using a Nicolet Magna-IR 870 bench [1] and a Thermo Spectra-Tech Continuum microscope using a diamond cell (Thermo Spectra-Tech, Diamond windows) to avoid exposing the moist protein powders to the high pressure encountered during the formation of KBr pellets usually employed in such studies [13,33]. Spectra were recorded in the transmission mode at 4 cm^-1 resolution and 256 scans were averaged to obtain one spectra. To avoid artifacts by water vapor and sorbed water, the amide III region (1220-1330 cm^-1) was used for the analysis of protein secondary structure since the absorbance of water is negligible in this spectral region [34]. It should be noted that the use of the amide III region to investigate the secondary structure of hydrated and dehydrated proteins has been validated by simultaneous analysis of the amide I and III region [1,37,38].

Spectral changes caused by moisture were quantified by calculating the spectral correlation coefficient (SCC) [34,38]. The SCC value reflects differences of two spectra: for identical ones it is 1, for those with nothing in common it is 0. The second-derivative spectra in the amide III region (1220-1330 cm-¹) were calculated and saved for the spectral range on identical wavenumber scales and with identical data spacing. The data for the reference spectrum (in this case the lyophilized powder measured by FT-IR microscopy) and the spectrum with the varied condition (e.g., RH and incubation time) were imported into the program SigmaPlot and the correlation coefficient was calculated [34,38].

Measurements of protein-bound water were conducted by Karl Fisher titration using hydranal solvent with a Metrohm 831 KF coulometer. This method has been routinely used to measure the water content of protein powders. Samples (0.5 mg) each were stored at various RH. Following storage, 1.0 mL of anhydrous dimethylsulfoxide was added to each sample. This solution was then sonicated for 30 sec and injected into the coulometer. All water contents are reported as percentages (w/w). Equilibrium uptake of water by the lyophilized powder was achieved and no additional water was absorbed after a maximum of 8 h at 50°C for all RH.

All statistical analyses were done in Minitab 14 (Minitab Inc., State College, PA). Statistical analysis of the effect of the water content (11%, 51%, 75%, 81% and 96% RH) on the quantity of insoluble aggregates formed after 24 h, 48 h, 72 h, 96 h, 1 week, and 2 weeks was performed using the Kruskal-Wallis test and the individual differences between the various samples. Post-hoc comparisons of the means of individual groups were performed using Dunn's test. Statistical significance was accepted at the p ≤ 0.05 level. All experiments were performed at least in triplicate, the results averaged, and the standard deviations calculated.