Analysis of the Adsorbed Vaccine Formulations Using Water Proton Nuclear Magnetic Resonance—Comparison with Optical Analytics

Abstract

Purpose

To evaluate wNMR, an emerging noninvasive analytical technology, for characterizing aluminum-adjuvanted vaccine formulations.

Methods

wNMR stands for water proton nuclear magnetic resonance. In this work, wNMR and optical techniques (laser diffraction and laser scattering) were used to characterize vaccine formulations containing different antigen loads adsorbed onto AlPO₄ adjuvant microparticles, including the fully dispersed state and the sedimentation process. All wNMR measurements were done noninvasively on sealed vials containing the adsorbed vaccine suspensions, while the optical techniques require transferring the adsorbed vaccine suspensions out of the original vial into specialized cuvette/tube for analysis. For analyzing fully dispersed suspensions, optical techniques also require sample dilution.

Results

wNMR outperformed laser diffraction in differentiating high- and low-dose formulations of the same vaccine, while wNMR and laser scattering achieved comparable results on vaccine sedimentation kinetics and the compactness of fully settled vaccines.

Conclusion

wNMR could be used to analyze aluminum-adjuvanted formulations and to differentiate between formulations containing different antigen loads adsorbed onto aluminum adjuvant microparticles. The results demonstrate the capability of wNMR to characterize antigen-adjuvant complexes and to noninvasively inspect finished vaccine products.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11095-023-03528-7.

Introduction

Since their discovery and the first use in 1932 [1], aluminum-adjuvanted vaccines have a niche of their own among the diverse systems used to formulate pharmaceutical suspensions. In contrast to other applications of suspension dosage forms, such as delivery of insoluble drugs, protection of unstable drugs from hydrolysis or oxidation, etc. [2], aluminum adjuvants enhance the immunogenicity and efficacy of many vaccines. While the precise mechanisms underlying this enhancement are not well understood nor universally accepted, it has been postulated to proceed through a confluence of several factors [3]. One of the factors discussed in the literature was the formation of a depot at the injection site resulting in gradual antigen release. However, in a more detailed study [4], it was shown that this “depot effect” is not necessary for the activity of aluminum adjuvants. Several direct effects of aluminum adjuvants on the immune system in general have also been hypothesized, including the promotion of antigen uptake and immune presentation by cells, specific interactions between aluminum adjuvants and cellular receptors, production of inflammatory cytokines induced by aluminum adjuvants, and others [5]. The inflammatory hypothesis was corroborated by demonstration of the enhancement of adaptive immunity resulting from localized tissue damage by aluminum adjuvants [6]. Aluminum adjuvants may also amplify immune response to antigens through antigen-adjuvant interactions in the form of antigen-adjuvant complexes [7, 8]. Several recent reviews [3, 9] present a wide range of opinions on the underlying mechanisms of action of aluminum adjuvants which are still intensely debated.

Despite the unanswered mechanistic questions, aluminum adjuvants have been widely and successfully used for decades in many licensed vaccines, and continue to be included in novel, investigational vaccine formulations. For example, aluminum adjuvants were recently used in novel vaccine candidates containing attenuated SARS-CoV-2 virus [10], and recombinant subunits of the SARS-CoV-2 spike receptor binding domain antigen [11].

For vaccine development, process monitoring, and quality control of the finished drug products, vaccine manufacturers have used a spectrum of analytical techniques [12] to characterize the aluminum adjuvants and antigen-adjuvant complexes. However, the heterogeneous size and shape of antigen-adjuvant complexes formulated as suspensions limits the applicability of many analytical approaches. Some amenable techniques are particle counting and sizing, and optical techniques such as laser diffraction, dynamic light scattering (DLS), focused beam reflection measurement (FBRM) [13, 14]. Intrinsic fluorescence and infrared spectroscopy (e.g., FTIR) were also shown to provide important data on antigen-adjuvant adsorption since the structural changes of the adsorbed antigen result in shifts in the fluorescence and IR spectra [15]. Combination of particle tracking techniques (FRBM) with IR spectroscopy allowed monitoring of the process of antigen adsorption on different aluminum adjuvants under manufacturing conditions in a small-scale bioreactor [16].

Additional quality attributes of aluminum-adjuvanted vaccines include the parameters characterizing the stability of dispersion (sedimentation rate), and the density and compaction of a sediment (the sedimentation volume ratio, SVR) [2]. Small SVR values may point to potential difficulty to redisperse a vaccine prior to administration, and incomplete redispersion could be a reason for vaccine recalls [17]. Measurement of the sedimentation rate and SVR are often performed using optical scanners [18, 19] and laser scattering analyzers [20], requiring relatively high sample volumes (2 mL up to 20 mL) to be transferred to a method-specific optical cuvette or tube. Of note, both the sedimentation rate and SVR are often considered as indirect characteristics of the antigen-adjuvant complex. For example, higher doses of an antigen adsorbed onto adjuvant particles have been shown to increase the sedimentation rate [20] and also affects the density and compaction of the sediment [21].

These analytical techniques have a common and important drawback which restricts their application to finished products: they are invasive, i.e., sample transfer to a method-specific cuvette is often needed, hence they cannot be applied to drug products in its original containers, such as sealed vials, injection pens, or prefilled syringes. After the drug content is transferred out of the original container, additional sample preparation steps are often required, such as dilution, heating, etc. Sample preparation steps bring additional measurement errors and are hard to automate. Current analytical techniques often require sophisticated precision optical equipment as well as a significant level of expertise to analyze complex spectral information.

Another extensively used analytical tool, nuclear magnetic resonance (NMR) spectroscopy, has been successfully applied in pharmaceutical research for biophysical characterization of structural and dynamic aspects in drug product formulations necessary to achieve the stability of biologics [22]. In recent years, we have been developing the water proton NMR (wNMR) technology [23, 24] based on the analysis of the relaxation dynamics of the water proton signal ¹H₂O, which is sensitive to the organization and structure of solutes and particles present in biopharmaceutical formulations, including vaccines. The high concentration of water in biopharmaceuticals and vaccines (~ 90% and above, over 50 M) allows monitoring of the ¹H₂O signal using compact low-field benchtop NMR relaxometers with a wide bore which can accommodate a drug product in the original container. Hence, all analyses are done noninvasively and do not compromise a drug product which can still be used after measurement, including for patient administration [25]. Data collection is fast (from several seconds to 1–2 min at most), and data analysis is straightforward, easily automatable, and does not require high-level expertise. wNMR may be also implemented in the flow mode [26] as a noninvasive technique for real-time in-line monitoring of biomanufacturing unit operations, such as transfer and filling unit operations.

Our previous research has demonstrated the capability of wNMR to provide important information on the dynamic behavior of pharmaceutical suspensions containing aluminum adjuvants. We explored the concentration-dependent differences in the sedimentation kinetics of two commercially available aluminum adjuvant suspensions, Alhydrogel® and Adju-Phos® [27]. wNMR was successfully used to analyze freezing susceptibility and to detect prior freezing history of aluminum-adjuvanted vaccines in their original containers [28]. We have also demonstrated that wNMR can noninvasively assess the stability of aluminum adjuvants and antigen-adjuvant complexes under physical stresses, including gravitation, shear and freeze/thaw [29].

In the present paper, we analyzed fully dispersed investigational vaccine formulations with high-dose (HD) and low-dose (LD) of antigen adsorbed onto aluminum phosphate (AlPO₄) adjuvant particles using wNMR. We then monitored the sedimentation process of HD and LD formulations. Results from wNMR are compared with results from two commonly used optical techniques, laser diffraction and laser scattering (Turbiscan^LAB).

Materials and Methods

Materials

Two different investigational vaccine formulations of the same protein antigen adsorbed to identical amounts of AlPO₄ adjuvant particles were studied by means of wNMR and optical analytical techniques. We explored intra-batch and inter-batch variability of the formulations from two batches with different antigen loads (high dose, HD; and low dose, LD). The high-dose (HD) batch had a total antigen concentration of approximately 0.35 mg/mL, while the low-dose (LD) batch had a total antigen concentration of approximately 0.15 mg/mL.

Four samples of two investigational vaccine formulations were filled in identical vials and sealed identically (identical aluminum caps and septa). Two pairs of samples from each batch were studied, each pair representing the same batch—samples HD-A and HD-B from the HD batch, and samples LD-A and LD-B from the LD batch (see Table 1). All samples were formulated as a whitish aqueous suspension, and the samples had identical volumes (0.5 mL), formulated in the same buffer, and have identical AlPO₄ concentration, 0.66 mg/mL of Al(III).

Table 1.

Samples of HD and LD investigational vaccine formulations adsorbed to AlPO₄

Batch	Sample
HD	HD-Aa
	HD-B
LD	LD-Aa
	LD-B

^aHigh antigen content, high dose (HD); low antigen content, low dose (LD)

Methods

Fully Dispersed Samples

wNMR Measurements

Measurements of fully dispersed samples were performed at 25°C using MQC + (Oxford Instruments, plc., Oxford, UK), at 23.8 MHz ¹H resonance frequency, magnet bore size 26 mm, magnet and probe temperature of 25°C. The temperature of the magnet needs to be controlled within a range of ± 0.01°C to ensure the stability of the permanent magnet. Prior to data collection, the sample vial is attached to a sample holder (Figure S1, Supplementary Information), which was then placed inside the magnet probe for temperature equilibration for 40 min. The sample holder ensures that the sample vial is centered inside the magnet bore and that different vials inside the bore are positioned identically. Measurements of the water proton transverse relaxation rate R₂(¹H₂O) were performed in triplicate for each of the sample using the CPMG pulse sequence [30] with an interpulse delay of 0.5 ms; 5000 echoes were collected. Prior to each of these triplicate measurements, the sample vial was vigorously shaken by hand to achieve uniform full dispersion. For each measurement, 8 transients were accumulated with a relaxation delay of 5 s (total duration ~ 1 min 20 s). Each vial contains 0.5 mL sample. All measurements were performed at 25°C.

Laser Diffraction Measurements

All measurements of the particle size distribution of the adsorbed investigational vaccine formulations were performed using a Mastersizer 3000 instrument (Malvern Instruments Ltd., Westborough, MA, USA), with an operating range of 0.01 to 3,500.00 µm. About 0.8 mL to 1.0 mL of the sample was aliquoted into the dispersion unit of the instrument and 125 mL of the dispersant (ultrapure H₂O, Milli-Q®) was added, resulting in 125 × to 150 × dilution of the vaccine stock solution. The dispersion unit was connected to the thermostat maintaining constant temperature at 25°C (± 0.05°C). The resulting dispersion was circulated through the detector cell at ambient temperature to measure the particle size. Particle size distributions in dispersions were determined by measuring the angular variation in the intensity of light scattered from a laser beam (λ = 633 nm) passing through the dispersed particulate sample. The reportable value is the derived diameter (Dv), which is the particle size (in μm) for a specific percentile of the cumulative size distribution. That means that for a specific value of Dv(N), N% of the sample particles have a diameter of Dv(N) or lower. Particles were measured at room temperature using the built-in “non-spherical” option within the instrument’s software, and the average Dv(10), Dv(50), and Dv(90) values of 5 measurements were reported in µm to one decimal point. The coefficient of variation for the qualified laser diffraction assay ranged from 5 to 7% for the adsorbed antigens.

Sedimentation Kinetics

wNMR Measurements

For sedimentation kinetics monitoring, one vial from each batch was studied: sample HD-A from the HD batch and sample LD-A from the LD batch. Like the above experiments, the sample vial attached to the sample holder was equilibrated inside the magnet bore for 40 min. The sample vial along with the holder was then taken outside the magnet bore and vigorously shaken by hand to fully disperse the sample. Afterwards, the sample vial along with the holder was put back inside the magnet bore and data collection ensured immediately, at a cadence of once per 5 min for 4 h. At each time point, 8 transients were accumulated with a relaxation delay of 10 s, an interpulse delay of 0.5 ms and 10,000 echoes were collected (total duration of a single measurement was approximately 2 min 40 s). The relaxation delay is longer for monitoring sedimentation than for measuring a full dispersed sample (10 s vs. 5 s) is because the water proton transverse relaxation slows down as sedimentation proceeds, hence requiring longer relaxation delay. wNMR monitors the sedimentation of 0.5 mL suspension sample (sample height ca. 5 mm) in sealed vials. All measurements were performed at 25°C.

Dispersion Stability Analysis—Turbiscan^LAB

All measurements of the sedimentation rate of microparticles in a dispersion were performed using the Turbiscan^LAB instrument (Formulaction, Worthington, OH, USA) with the temperature-controlled cell (from ambient temperature to 60°C). An aliquot of 1.6 mL dispersed sample was loaded into a 4 mL glass vial and sealed with a Teflon lined cap. At a constant temperature of 28°C, the instrument scans the sample from top to bottom every 30 s for a duration of two hours. In each scan, it detects the transmission and backscattering of the laser (λ = 880 nm) through the sample. Once scanning is completed, the operator defines “the bottom of the vial” and “the meniscus of the sample” positions in the software by looking at the changes over time in the transmission and backscattering spectra. As the sample settled, backscattering increased near the bottom and decreased at the top and vice versa for transmission (Figure S2, Supplementary Information). Then, the largest peak in the appropriate spectrum is used to calculate the sedimentation rate. The thickness of this peak, which is the distance between the 2 intercepts of the curve with the threshold line, changes over time. The sedimentation rate is the slope of the initial linear portion of the peak thickness vs. time graph and is measured in mm/h (Figure S3, Table S1, Supplementary Information).

The sedimentation volume ratio (SVR) is calculated as:

$$SVR=\frac{H_{\mathit{sed}}}{H_{\mathit{disp}}}$$ 1

where H_sed is the height of the sediment layer after the completion of the sedimentation, and H_disp is the height of the fully dispersed sample (Table S1, Supplementary Information). During all sedimentation monitoring, the temperature was maintained at 28°C controlled by a Peltier cell (± 0.01°C).

wNMR Data Processing

Fully Dispersed Samples

The spin-echo signal intensity decay data of fully dispersed samples were processed using the single-exponential fitting function to extract the water proton transverse relaxation rate R₂(¹H₂O)

$$I\left(t\right)=I_0\times \exp \left[-,\times ,R_2,{(}^1,{\mathrm{H}}_2,\mathrm{O})\right]$$ 2

where I(t) is the observed echo signal intensity over the echo decay time t, and I₀, the pre-exponential factor, is the echo signal intensity at t = 0 (Figure S4, Supplementary Information). Three R₂(¹H₂O) values from triplicate measurements for each vial were arithmetically averaged. For batch averaging, i.e., averaging the values observed for two vials within a batch, the resulting mean and standard deviation (SD) values for each batch (HD batch with samples HD-A, HD-B and LD batch with samples LD-A, LD-B) were used to perform weighted averaging where the aforementioned SD values were used as weights.

Sedimentation Kinetics

Preliminary analysis of the experimental data demonstrated that the sedimentation process was mostly completed within 35–40 min for both HD-A and LD-A. Therefore, detailed data on sedimentation kinetics are presented for the first 90 min of the sedimentation process.

Note that during the sedimentation process, R₂(¹H₂O) is measured repeatedly at a cadence of every 5 min using the CPMG pulse sequence. For each measurement, the ¹H₂O echo signal intensity I decay is a function of two time-variables, the echo decay time t within each CPMG pulse sequence, ranging from 0 to 10 s, and the sedimentation time T, ranging from 0 to 90 min. Hence, we denote the echo decay signal intensity as I(t, T).

At the beginning of the sedimentation process, the aluminum microparticles are evenly dispersed and there is only one phase in the sample. Correspondingly, at the early stage of the sedimentation process, the spin-echo intensity decay data I(t, T) can be adequately fitted using a single-exponential function to extract the water proton transverse relaxation rate R₂(T) at time T of the sedimentation process as:

$$I\left(t,,,T\right)=I_0(T)\times \exp \left[-,t,\times ,R_2,(T)\right]$$ 3

Here, similar to Eq. 2, I₀(T) is a pre-exponential factor, the echo signal intensity at t = 0 of the sedimentation time T. The R₂(T) vs. T plot shows the kinetics of sedimentation process.

As the sedimentation proceeds (i.e., as T increases), two layers gradually emerge in the sample, one supernatant layer and one sediment layer. Correspondingly, single-exponential fitting of the echo decay data I(t, T) gradually becomes inadequate. Instead, double-exponential fitting is required, with one slower relaxing component attributed to the supernatant layer and one faster relaxing component attributed to the sediment layer. The water in the sediment layer relaxes faster than water in the supernatant layer because the sediment layer has much higher concentration of aluminum adjuvant microparticles, which facilitate water proton relaxation.

At larger T, the echo signal decay data I(t, T) are fitted by a double-exponential function as:

$$I\left(t,,,T\right)=I_0^s\left(T\right)\exp \left[-,t,\times ,R_2^s,(T)\right]+I_0^F(T)\exp \left[-,t,\times ,R_2^F,(T)\right]$$ 4

where I₀^S(T) is the echo signal intensity at t = 0 of the slow-relaxing component rate R₂^S(T) attributed to water in the supernatant layer, I₀^F(T) is the echo signal intensity at t = 0 of the fast-relaxing component rate R₂^F(T) attributed to water in the sediment layer (Figure S4, Supplementary Information).

Note that the pre-exponential factors in Eq. 4, I₀^S(T) and I₀^F(T), define the quantitative contributions of the supernatant and the sediment, respectively, to the total spin-echo intensity at each sedimentation time point T. This allows one to extract the sedimentation volume ratio SVR from the wNMR data at any sedimentation time point T, SVR^NMR(T), as a fraction of the fast-relaxing component’s contribution to the sum of the contributions of both components

$${\mathit{SVR}}^{\mathit{NMR}}\left(T\right)=I_0^F(T)/\left[I_0^F,\left(T\right),+,I_0^S,(T)\right]$$ 5

At the sedimentation completion time point (T = 90 min), the plateau value of SVR^NMR(T) is compared with the SVR from the Turbiscan^LAB data—the ratio of the height of the sediment H_sed after the completion of sedimentation process to the height of fully dispersed sample H_disp before sedimentation (see Eq. 1). Sedimentation monitoring with Turbiscan^LAB was performed for 120 min; however, after 90 min there are essentially no changes in the peak thickness data (Figure S3).

The optical sedimentation monitoring techniques used in this study (Turbiscan^LAB) traces the increase of the backscattering peak of the sediment layer over time to derive the sedimentation rate. Therefore, for comparison/correlation between wNMR and Turbiscan^LAB, it would be prudent to use the rate of the sediment formation captured by the increase in the fast-relaxing component R₂^F(T) obtained by wNMR over the sedimentation time T which is also attributed to the sediment layer.

The experimentally observed increase of R₂^F(T) vs. T during the sedimentation process of HD and LD formulations (Figure S5, Supplementary Information) could be modeled using the formalism of the kinetics of first-order reactions as follows,

$$R_2^F\left(T\right)=R_2^F\left(0\right)+\mathrm{\Delta }{R}_2^F\left[1,-,\exp ,(,-,T,\times ,k_{\mathit{sed}}\right]$$ 6

here k_sed is the sedimentation rate constant in min⁻¹, ΔR₂^F = R₂^F(∞) − R₂^F(0) ≈ R₂^F(T_end) − R₂^F(0), R₂^F(0) and R₂^F(T_end) are respectively the water proton transverse relaxation rate of the faster-relaxing component at the beginning and the end of the sedimentation process. Physically, R₂^F(0) is the water proton transverse relaxation rate of the fully suspended sample (only one layer), while R₂^F(T_end) is the water proton transverse relaxation rate of the sediment layer at the end of sedimentation (two layers now). R₂^F(∞) is the water proton transverse relaxation rate of the sediment layer after infinite sedimentation time (T = ∞) and hence is purely a mathematical parameter. In this work, T_end = 90 min and R₂^F(T_end) is the asymptotic approximation of R₂^F(∞).

Within the framework of the same first-order kinetic model, the sedimentation rate v_sed(T), which defines the instantaneous sedimentation rate at time T, could be calculated using the differential form of first-order rate equation as follows,

$$v_{\mathit{sed}}=\frac{{\mathit{dR}}_2^F(T)}{\mathit{dT}}=k_{\mathit{sed}}{\mathrm{\Delta }R}_2^F\exp (-\mathrm{T}\times k_{\mathit{sed}})$$ 7

At the beginning of the sedimentation (T = 0)

$$v_{\mathit{sed}}\left(0\right)=k_{\mathrm{sed}}\mathrm{\Delta }{R}_2^F$$ 8

k_sed (in min⁻¹) and ΔR₂^F (in s⁻¹) are obtained by fitting the experimental R₂^F(T) vs. T data with Eq. 6 (Figure S5, Table S2, Supplementary Information). From k_sed and ΔR₂^F, v_sed(0) (in s⁻¹/h) can be calculated. v_sed(0), the initial sedimentation rate from wNMR, is used to compare with the initial sedimentation rate from Turbiscan^LAB.

RINMR ver. 7.0 software (Oxford Instruments Industrial Analysis, Abingdon, UK) was used for wNMR data phasing (tuning the phase of the imaginary part of the data) and exporting to the tab delimited text format for further processing. Model fitting of these data was performed using the Origin 2019b software (OriginLab Corp., Northampton, MA).

Results and Discussion

Fully Dispersed Samples

Figure 1 compares R₂(¹H₂O) values observed for all four samples of the HD and LD investigational vaccine formulations. Differences between the samples of two batches (HD-A, HD-B and LD-A, LD-B) were obvious. Note that vials from both the HD batch (HD-A and HD-B) and the LD batch (LD-A and LD-B) demonstrated very close intra-batch R₂(¹H₂O) values (Fig. 1 and Table 2).

Fig. 1.

Comparison of R₂(¹H₂O) values for the full set of four samples from two batches (HD-A and HD-B from the HD batch, and LD-A and LD-B from the LD batch). Arithmetic averages and the error bars reflecting the SDs from three consecutive measurements are shown (see also Table 2).

Table 2.

R₂(¹H₂O) Values for Fully Dispersed HD and LD Investigational Vaccine Formulations

Batch	Sample	aR2(1H2O), s−1	bR2(1H2O), s−1
HD	HD-A	3.917 ± 0.035	3.872 ± 0.012
	HD-B	3.865 ± 0.013
LD	LD-A	3.641 ± 0.038	3.604 ± 0.001
	LD-B	3.604 ± 0.001

^aArithmetic average of triplicate measurements of each vial

^bWeighted average of the results for two vials from each batch

To emphasize the intra-batch similarity and inter-batch differences, Fig. 2 compares the data for the HD batch and the LD batch resulting from the weighted averaging of the R₂(¹H₂O) values obtained for individual samples (Fig. 1 and Table 2). Similar to Fig. 1, the difference between two batches is evident.

Fig. 2.

Comparison of R₂(¹H₂O) values for the HD batch (samples HD-A and HD-B) and the LD batch (sample LD-A and LD-B). Data for the HD batch and the LD batch and SDs are the result of weighted averaging of the measurements of individual samples (see also Table 2).

Laser diffraction results for fully dispersed samples demonstrated rather wide particle size distributions for both batches (Table 3). Note that the HD batch showed somewhat lower particle size on average, with derived diameters Dv(50) and Dv(90) smaller compared to those observed for the LD batch. However, these differences are well within the range of the coefficient of variation of the technique (5–7%). Also, as seen from Table 3, formulations of both HD batch and LD batch have distribution span greater than 2, with the LD samples showing slightly larger span (wider particle size distribution) compared to the HD samples. However, the distribution span difference between the HD batch and the LD batch is also within the range defined by the above coefficient of variation. In general, one may conclude that laser diffraction measurements show very minor, if any, differences between two formulations.

Table 3.

Particle Size Distribution Parameters of Fully Dispersed HD and LD Formulations^a

Sample	Dv(10), µm	Dv(50), µm	Dv(90), µm	Span, Sb
HD	7.65 ± 0.54	18.0 ± 1.3	44.8 ± 3.1	2.06 ± 0.23
LD	5.75 ± 0.40	19.2 ± 1.3	49.3 ± 3.5	2.27 ± 0.24

^aErrors calculated based on the coefficient of variation of 7%

^bCalculated as S = [Dv(90) – Dv(10)]/Dv(50)

In sum, wNMR analysis of the two fully dispersed formulations demonstrates the consistency of drug products within the same batch. More importantly, wNMR can clearly distinguish high- and low-dose batches. In contrast, laser diffraction was unable to detect statistically significant differences between HD and LD batches. Apparently, the impact of different antigen content on the size of AlPO₄ microparticles is rather modest. On the other hand, different antigen content, ca. 0.35 mg/mL (HD) vs. ca. 0.15 mg/mL (LD), has a significant impact on water proton relaxation.

Sedimentation Kinetics

To further explore potential differences between high- and low-dose batches, we studied the sedimentation kinetics of one sample from each batch—sample HD-A from the HD batch and sample LD-A from the LD batch. Sedimentation kinetics proves to be a valuable tool to study the stability of dispersion and indirectly characterizes the extent of antigen adsorption on the surface of the aluminum adjuvant microparticles. Another important parameter derived from the sedimentation kinetics, the sedimentation volume ratio (SVR), indicates the sediment compactness, which relates to the redispersion of aluminum-adjuvanted vaccines; sediments with smaller SVR are more compact and hence might be more difficult to redisperse.

As has been shown previously [27], the water proton transverse relaxation rate R₂(¹H₂O) consistently decreases during the sedimentation of aluminum adjuvant suspensions.

Indeed, as seen in Fig. 3, the water proton transverse relaxation rate R₂(T) of both HD batch HD-A and LD batch LD-A samples steadily decreased over the sedimentation time T. However, the results of single-exponential fitting of the wNMR data at each time point T do not show apparent differences between the sedimentation profiles of two batches (Fig. 3).

Fig. 3.

Sedimentation kinetics monitored via the water proton transverse relaxation rate R₂(T) derived from single-exponential fitting of the spin-echo decays taken every 5 min during the sedimentation process. Standard deviation is 0.001 s⁻¹ (fitting error). Solid line shows the data smoothed with B-spline.

As it has been mentioned above, the decay of spin-echo signal intensity I(t) observed during the sedimentation process is comprised of two components—slower relaxing supernatant and faster relaxing sediment (see also Figure S4, Supplementary Information). Therefore, we also used the double-exponential fitting function (Eq. 4) to analyze spin-echo decays taken at different sedimentation time points T. This approach allowed one to extract the water proton transverse relaxation rate of the supernatant (slower component) R₂^S(T) at time T, the water proton transverse relaxation rate of the sediment (faster component) R₂^F(T) at time T, as well as their respective contributions, I₀^S(T) and I₀^F(T), to the total spin-echo intensity I(T) at time T. Figure 4 presents changes in the water proton transverse relaxation rate of the slower relaxing supernatant component R₂^S(T) over sedimentation time for the two batches.

Fig. 4.

Sedimentation kinetics monitored via the water proton transverse relaxation rate of the slower-relaxing component (supernatant layer) R₂^S(T) over time. Standard deviation is 0.001 s⁻¹ (fitting error). Solid line shows the data smoothed with B-spline.

As seen in Fig. 4, the kinetic traces R₂^S(T) of the slower relaxing supernatant layer for HD and LD are also very similar. After completion of sedimentation, both profiles level off at nearly identical R₂^S(T) values. This suggests similar chemical compositions of the supernatant of HD and LD samples, which is predominantly the formulation buffer with very few antigen-adjuvant microparticles left in the supernatant layer at the end of sedimentation.

Stark differences between the HD and LD batches were observed when monitoring the faster-relaxing sediment layer, R₂^F(T) vs. T. As seen in Fig. 5, the HD sample shows notably faster sedimentation compared to the LD sample (v_sed(0) is 29.1 s⁻¹/h for HD-A and 23.4 s⁻¹/h for LD-A). Additionally, the higher plateau value of R₂^F(T) for the HD sample after completion of sedimentation suggests higher compaction of the sediment layer compared to the LD sample (the two batches have the same amount of aluminum adjuvant microparticles, at 0.66 mg/mL of Al(III)). Indeed, in more compact aluminum sediment, the concentration of AlPO₄ microparticles will be higher, and higher concentrations of aluminum adjuvants are known to result in much higher values of the water proton transverse relaxation rate R₂(¹H₂O) [27, 31].

Fig. 5.

(A) Sedimentation kinetics monitored via the water proton transverse relaxation rate R₂^F(T) of the faster-relaxing component (sediment layer) obtained from wNMR. Standard deviation is 0.080 s⁻¹ (fitting error). (B) Sedimentation kinetics monitored via the sediment layer peak thickness obtained from laser scattering (Turbiscan^LAB). Solid line shows the data smoothed with B-spline.

Monitoring the changes of R₂^F(T) in the sediment layer provides a good means for correlating wNMR data with Turbiscan^LAB data. Turbiscan^LAB monitors the increase in laser backscattering of the sediment layer, and the profile of such increase vs. sedimentation time T shows a pattern similar to the wNMR R₂^F(T) vs. T profile (Fig. 5). Turbiscan^LAB software reduced the estimates of sedimentation rate down to the fits of the initial linear segment of the graph depicting thickness of backscattering peak vs. sedimentation time (Figure S3, Supplementary Information). The resulting analysis showed a clear difference between the sedimentation rates of HD and LD samples. Based on the linear approximation of the Turbiscan^LAB data, the sedimentation rate (in mm/h) of the HD sample is 1.52 times greater than that of the LD formulation (Table 4).

Table 4.

Comparison of Sedimentation Parameters from Turbiscan^LAB and wNMR

Parameter	Sample
	HD	LD
Sedimentation Rate (TurbiscanLAB), mm/h	43.6	28.7
Sedimentation Rate (wNMR), s−1/h	29.0	23.5
HD rate/LD rate (TurbiscanLAB)a	1.52 (= 43.6/28.7)
HD rate/LD rate (wNMR)a	1.24 (= 29.0/23.5)
SVR(TurbiscanLAB)	0.36	0.45
SVR (wNMR)	0.30	0.38
HD SVR/LD SVR (TurbiscanLAB)a	0.80 (= 0.36/0.45)
HD SVR/LD SVR (wNMR)a	0.79 (= 0.30/0.38)

^aCalculated as the ratio of corresponding parameters for HD to LD

Comparable difference between the HD and LD formulations was obtained from fitting R₂^F(T) vs. T data (Fig. 5, and Figure S5, Supplementary Information) using first-order kinetics model (Eq. 6), and the extracted parameters of the fit to estimate the initial sedimentation rate v_sed(0) (in s⁻¹/h) from wNMR data (Eq. 8). As seen in Table 4, based on the wNMR data, v_sed(0) of the HD sample was 1.24 times faster compared to the LD sample.

Although they showed similar trends, the sedimentation rate ratios of the HD and LD formulations obtained from Turbiscan^LAB and wNMR are significantly different (1.52 vs. 1.24). This could be due to differences in data processing. As already mentioned, Turbiscan^LAB uses the initial linear segment of the sedimentation curve, while wNMR analysis is performed on the entire sedimentation curve, not just the initial linear sedment. Another possible source of this discrepancy could be the difference in sample volumes and vial dimensions used in each technique. Recall that Turbiscan^LAB analysis was performed with 1.6 mL of a sample, while the sample volume used for wNMR measurements was 0.5 mL. In any case, the two techniques both show that higher antigen dose leads to faster sedimentation of the aluminum adjuvant microparticles.

Double-exponential analysis of the wNMR data on sedimentation kinetics allows one to extract the quantitative contributions of the supernatant, I₀^S(T), and the sediment, I₀^F(T), to the observed total spin-echo intensity at different time points T of the sedimentation process (Eq. 4). These values make it possible to calculate the sedimentation volume ratio from the wNMR data, SVR^NMR(T), during the entire sedimentation (Eq. 5).

SVR^NMR(T) adequately reflects the gradual compaction of the adjuvant microparticles in the sediment layer during the sedimentation process. Note that the observed changes of SVR^NMR(T) vs. T (Fig. 6) corroborate the conclusion of a more compacted sediment layer of the HD sample compared to the LD sample which is based on faster relaxation rate of the sediment layer R₂^F(T) (Fig. 5). Indeed, the plateau value of SVR^NMR(T) of HD was significantly smaller compared with that of LD (0.30 vs. 0.38, Table 4), suggesting greater compaction of the HD sediment.

Fig. 6.

Changes in the sedimentation volume ratio SVR^NMR(T) during the sedimentation process. Standard deviation is 0.01 (fitting error). Solid line shows the data smoothed with B-spline.

SVR was also obtained from Turbiscan^LAB data using Eq. 1 (Table 4), although only at the end of the sedimentation. While the absolute SVR from the two techniques, Turbiscan^LAB and wNMR, differ to some extent (0.36 vs. 0.30 for HD and 0.45 vs. 0.38 for LD), the SVR ratio between HD and LD samples obtained by Turbiscan^LAB and wNMR were almost identical (0.80 vs. 0.79). Again, like the above discussed case of sedimentation rates, the difference in the absolute values of SVR could be due to the use of different sample volumes and different vial geometry between by the two techniques. Considering that wNMR is performed on intact single-dose vials, its results provide better measure of the finished drug products.

To summarize, both Turbiscan^LAB and wNMR consistently showed faster sedimentation and greater compaction of the sediment layer for the HD formulation than the LD formulation. This difference between high- and low-dose formulations could be explained based on the difference in electrostatic repulsion effects between AlPO₄ particles. In the absence of adsorbed antigen, repulsion between similarly charged adjuvant microparticles would result in rather slow sedimentation which takes multiple days to reach plateau [27]. Antigens and adjuvants typically are oppositely charged in a formulation. The larger the load of adsorbed antigen, the more charge reduction between adjuvant microparticles would occur, leading to faster sedimentation and more tightly packed sediment.

Conclusions

In fully dispersed formulations of an investigational vaccine, wNMR demonstrated the capability to distinguish between high and low antigen loads adsorbed onto AlPO₄ adjuvant microparticles. wNMR consistently detected close similarity between samples of the same batch with identical antigen load and revealed clear distinction between the samples from two different batches with high and low antigen loads, respectively. By way of comparison, optical methods analyzing particle size distributions such as laser diffraction, were not able to detect the differences between the batches with high and low antigen loads adsorbed onto AlPO₄ adjuvant microparticles.

Comparison of wNMR and optical techniques for analysis of dispersion stability (Turbiscan^LAB) showed the potential of wNMR to provide reliable estimates of both the sedimentation rates of investigational vaccine formulations and the sedimentation volume ratio in fully sedimented samples. Both are important quality characteristics of aluminum-adjuvanted vaccines. wNMR results showed consistency with the Turbiscan^LAB observations, and clearly differentiates between the formulations with high and low antigen loads adsorbed onto AlPO₄ adjuvant particles. The method has potential application to differentiate between high- and low-dose formulations.

Note that unlike optical analytical methods, all wNMR measurements are noninvasive, performed in the original drug product container without sample transfer and dilution. Much smaller sample volumes are needed for measurement compared to optical techniques such as Turbiscan^LAB. Additionally, wNMR measurements do not require optical transparency of the drug product container.

The capability of wNMR to distinguish between the formulations containing different loads of antigen adsorbed onto aluminum adjuvant microparticles paves the way for applying wNMR to monitor the formation of antigen-adjuvant complexes and to determine the degree of antigen adsorption onto the adjuvant particles. As such, wNMR could be advantageously used to facilitate vaccine product development and final product inspection.

Supplementary Information

Below is the link to the electronic supplementary material.

Data Availability

The authors hereby confirm that all the relevant data are included in the manuscript and the Supplementary Information, subject to any legal, ethical, or proprietary restrictions. Requests for access to the data may be considered on a case-by-case basis, and interested researchers can contact the corresponding authors for further information.