Publisher Correction: Characterisation of factors contributing to the performance of nonwoven fibrous matrices as substrates for adenovirus vectored vaccine stabilisation

The global network of fridges and freezers known as the “cold chain” can account for a significant proportion of the total cost of vaccination and is susceptible to failure. Cost-efficient techniques to enhance stability of vaccines could prevent such losses and improve vaccination coverage, particularly in low income countries. We have previously reported a novel, potentially less expensive thermostabilisation approach using a combination of simple sugars and glass micro-fibrous matrix, achieving an excellent recovery of vaccines after storage at supraphysiological temperatures. This matrix is, however, prone to fragmentation and currently not suitable for clinical translation. Here, we report an investigation of alternative, potentially GMP compatible, fibrous matrices. A number of commercially-available matrices permitted good protein recovery, quality of sugar glass and moisture content of the dried product but did not achieve the thermostabilisation performance of the original glass fibre matrix. We therefore further investigated physical and chemical characteristics of the glass fibre matrix and its components. Our investigation shows that the polyvinyl alcohol present in the glass fibre matrix assists vaccine stability. This finding enabled us to develop a custom-produced matrix with encouraging performance, as an initial step towards a biocompatible matrix for clinical translation. We discuss the path to transfer of the technology into clinical use, including potential obstacles.


54
Vaccine immunogens are composed of complex biological macromolecules. Good immunogenicity 55 requires stability of these molecules throughout the lifespan of a product, from production and 56 formulation, through transportation and storage to delivery to the recipient. Extrinsic factors such 57 as light, pH, agitation and oxidants combine with temperature fluctuations to challenge product 58 stability. Therefore, most vaccines need to be continuously stored in fridges or freezers. Maintaining 59 the cold chain and associated logistics in vaccination campaigns can contribute up to 45% of the 60 total cost of vaccination 1 . On the other hand, damage as a result of cold chain breakages costs 61 several million dollars annually 2 . These issues affect not only low income countries, where the cold 62 chain is regarded as being least reliable, but also developed countries 3 4 5 . The vaccines responsible 63 for eradication of smallpox and rinderpest (the only two diseases eradicated by immunisation) were 64 both thermostable, a factor believed to contribute to their success 6 . Development of technologies 65 to enhance vaccine thermostability has therefore been a major focus of research effort 7 . 66 Viral vectored vaccines, in particular adenoviruses, are versatile platforms for development of novel thermostabilisation performance, advantages of SMT include a relatively short process duration (in 79 some cases as little as 18 hours) and avoidance of the stress of extreme in-process temperature 80 exposure involved in lyophilisation and spray drying. 81 Our previous published SMT work used a glass fibre matrix and a polypropylene matrix; we reported 82 that the glass fibre matrix achieved better stability than the polypropylene matrix 26 . Glass fibre 83 matrix is however prone to shedding of non-biocompatible fibres, and so not suitable for clinical 84 translation. Here, we present results from an effort to characterise the factors contributing to the 85 performance of glass fibre matrix and to test alternative matrices more suitable for clinical use. 86 87 Viruses and infectivity titration 89 Simian adeno virus vectors ChAd63-METRAP and ChAdOx2-RabGP were prepared, purified and 90 tested for quality by the Jenner Institute Viral Vector Core Facility, as previously described 28,29 . 91 Viruses were dialysed against either a previously used storage buffer (10mM Tris, 7.5 % w/v sucrose, 92 pH 7.8) or unbuffered 0.5M trehalose and sucrose and stored at -80 o C as stock. Typical preparations 93 were supplied and stored at a titre of c. 1x10 12 virus particles (VP) per mL, corresponding to c. 1x10 10 94 infectious units (IU) per mL and hence a particle: infectivity ratio of c. 100. 95 For infectivity titration, duplicate fivefold serial dilutions were prepared in complete DMEM (10% 96 FCS, 100 U penicillin, 0.1 mg streptomycin/ml, 4 mM L-glutamine) and used to infect 80-100 % 97 confluent HEK293-TRex cells (ThermoFisher) grown in 96-well plates (BD Purecoat Amine, BD 98 Biosciences, Europe). Infected cells were immunostained and imaged as previously described 30 . 99 Wells containing 20-200 spots were used to back-calculate recovered infectious units. For experiments involving reconstitution of dried samples, this was performed by addition of 117 phosphate buffered saline (Sigma), followed by brief vortexing of the vial (1±0.5 seconds, three 118 times). Virus infectivity after reconstitution was assayed as described above. Recovery of infectious 119 virus was quantified by comparison to a sample of the starting material, included on the same assay 120 plate. Between the set-up of an experiment and the assay of recovered infectivity, such comparator 121 material was stored at -80°C in aqueous buffer (under which conditions loss of infectivity is known to 122 be negligible). 123 Recovery was calculated in terms of log10-fold loss in the total infectious virus content of the matrix 124 i.e. log10-fold loss = log10(infectious units dried on matrix based on -80 stored sample) -125 100 mm 2 pieces of glass fibre (S14) were cut, autoclaved at (121 o C, 15 minutes) and loaded with 50 138 µL of sugar solution prior to drying in the glove box, vialling and reconstitution as described above. 139 Reconstituted solution from 10 vials was aspirated using a syringe with a conventional 20G needle 140 (BD Biosciences), to produce a single pooled unfiltered sample. Reconstituted solution from a 141 further 10 vials was aspirated using a 5-micron filter needle (BD Biosciences), to produce a single 142 The glass transition temperature (Tg) of sugar glass in matrices was measured immediately after 156 drying had completed using Differential Scanning Calorimetry (DSC). The melting point of the binder 157 in untreated glass fibre (S14) was measured using DSC (Q2000, TA instruments). The instrument was 158 purged with dry nitrogen (50 mg/mL) continuously during sample measurement. Calibration was 159 performed prior to measurements using a certified reference material (Indium) for temperature and 160 heat flow accuracy.
(50:50). Discs were weighed and, for each matrix type in turn, a total mass of 5-15 milligrams was 163 loaded into Aluminium DSC pans (TA Instruments) and hermetically sealed. Samples were subjected 164 to a temperature ramp from -20 o C to 180 o C at a heating rate of 10 o C per minute. Measurements 165 on all samples were performed in duplicates. Thermograms relating heat flow (W/g) to temperature 166 ( o C) were analysed using Trios software (TA Instruments) for identification of the glass transition (Tg) 167 onset temperature. 168 For the measurement of enthalpic recovery, which manifests as an endothermic peak at the glass 169 transition, modulated DSC (temperature modulation ±0.50°C every 60 seconds and ramp rate 170 3 o C/min from -20 o C to 100 o C) was employed. The enthalpic recovery was estimated by linear peak 171 integration in the thermograms plotted between nonreverse heat flow (W/g) and temperature ( o C) 172 using Universal Analysis software (TA Instruments). 173 174 Protein recovery 175 A 10mg/mL solution of lysozyme (Sigma-Aldrich) was made in 0.5M trehalose sucrose. 25 µL was 176 loaded into each matrix in triplicates. Protein was reconstituted from the matrices after desiccation 177 overnight and recovery was quantified using EnzChek Lysozyme Assay Kit (ThermoFisher Scientific). 178 Fluorescence measurements were performed in triplicates for each sample and protein recovery 179 calculated by interpolation on a standard curve, using GraphPad Prism. 180 181 Thermogravimetric analysis 182 Degradation temperatures of matrix constituents were measured by thermogravimetric analysis 183 using a TGA Q500 (TA instruments). Samples loaded into a tared platinum pan just prior to 184 measurement were subjected to a temperature ramp at 5 o C per minute from ambient to 550 o C in a 185 flowing nitrogen atmosphere (100ml/min). The gas was switched to air at 550 o C (100 ml/min) and heat was continued at the rate of 5 o C/min to 730 o C. Data was analysed using Universal Analysis software (TA Instruments). 188 plotted between measured hydrodynamic diameter and known average molecular weight (KDa) to 212 interpolate size of the PVA in the extract. 213 Nuclear magnetic resonance (NMR) spectroscopy 214 The degree of hydrolysis of the PVA binder present on the glass fibre matrix (S14) was estimated by 215 1 H NMR measurement as the intensity of the peak attributable to the acetyl group present in non-216 hydrolysed PVA. A 5 mg/mL aqueous solution of PVA extracted from the glass fibre sample (S14) was 217 prepared in deuterium oxide. Reference spectra for PVA with varying degrees of hydrolysis were 218 obtained by mixing >99% hydrolysed PVA and 80% hydrolysed PVA in appropriate proportions to 219 produce standards containing c. 0%, 2%, 5%, 10% and 20% acetyl groups, again at 5 mg/mL in 220 deuterium oxide. An AVIII 700 instrument (Bruker Biospin) was used to generate 1 H 1D spectra 221 (employing a quantitative 1D NOESY (Nuclear Overhauser Effect Spectroscopy) presaturation 222 sequence with recovery delay d1 =30s) and 2D 1 H-13 C HSQC (heteronuclear single quantum 223 coherence spectroscopy) plots. 224 Application of PVA to matrices 225 Aqueous solutions of each type of PVA to be investigated were prepared at 10 mg/mL. Matrices 226 were cut into approximately 100 mm 2 pieces and loaded with until the matrices were saturated by 227 the solution and air-dried overnight. The polyamide matrix (33100L) required surface modification 228 by washing in 100% ethanol for 20 minutes and air drying before PVA could be loaded. Following 229 PVA application, vaccines were dried on the matrices and thermostability assessed as described 230

above. 231
Study of physical characteristics of matrices 232 The physical characteristics of matrices were studied using NWSPs (Nonwoven Standard Procedures) 233 prescribed by the EDANA, the international trade association for the nonwoven industry.

256
A glass fibre matrix ( Figure 5B) was custom made by a wet-lay process using commercially available 257 glass fibre and polyvinyl alcohol aiming for similar porosity, thickness, areal density and wetting 258 behaviour as the glass fibre sample (S14). A matrix of glass fibre type 475 with diameter 4 µm (Johns 259 Manville) was prepared and 1 g/L PVA solution (>99% hydrolysed, Mw 146000-186000 kDa, Sigma) 260 was applied to each side using a spray gun before drying at 110 o C for 15 minutes. 261 262 Glass fibre matrix achieves good stability but is not suitable for clinical 264 development 265 In our previously published work, we had used low vaccine doses (1.1 x 10 10 viral particles per 266 matrix, as compared to a typical human dose of 5 x 10 10 viral particles), applied to the glass fibre 267 matrix 'Standard 14' (S14, GE Healthcare, Figure 1A) 26  In our previously published work, we reported that the S14 matrix achieved better vaccine 289 thermostability than an alternative commercially-available polypropylene-based matrix (HDC ® II J200, 290 Pall Corporation). We therefore sought to identify alternative commercially-available matrices which 291 might offer thermostability equivalent to or better than that achieved with S14, but without the 292 problem of shedding of non-biocompatible fibres. A set of seven matrices were selected based on 293 manufacturers' product specifications claiming low fibre shedding, low chemical leaching and 294 compatibility with sterilisation either by dry heat, steam or gamma-radiation sterilisation ( Table 1). 295 Henceforth matrices are referred to, for clarity of identification, in terms of their fibre material and 296 the manufacturer's product name. 297 We initially screened all matrices for suitable loading capacity (>20 µL of deionised water/cm 2 ). 298 Matrices which did not absorb 20 µL/cm 2 (without visible remaining beads of water within two 299 seconds) were re-tested after treatment with 2% polysorbate 20 solution. Matrices which did not 300 absorb 20 µL/cm 2 after detergent treatment were not studied further. We proceeded to further 301 study of the remaining five matrices along with the two previously tested matrices (glass fibre S14 302 and polypropylene J200). 303 The architecture of the selected matrices was characterised by scanning electron microscopy ( Figure  304 2). Fibre diameters estimated for each matrix type using Image J analysis are presented in table 1. 305 Fibres in the polypropylene (J200) matrix (Fig 2A) and glass fibre (conjugate pad) matrix ( Fig 2F)  306 shared the straight, rod-like fibre morphology seen in the original glass fibre (S14) matrix (Fig 1A-B) 307 but had larger fibre diameters of 10 -20 μm (as compared to 4 μm in the S14 glass fibre sample). 308 The remaining four matrices (Fig 2B-E) exhibited a greater degree of curl along their length. A film, 309 potentially a binder, was apparent on the glass fibre matrix (conjugate pad) (Fig 2F). Matrices loaded 310 with 0.5 M sugar solution and dried at room temperature for 24 hours also showed differences in the distribution of the sugar glass intercalated between the fibresError! Reference source not 312 found. (Figure 2). Distinct films of sugar glass between the fibres were visible on the two glass fibre 313 matrices, but not in the polyamide (33100L), polyester (leukosorb) or polypropylene (J200) matrices. 314 The sugar loaded polyester (23100) matrix and cellulose (31 ET CHR) matrices showed a glazing 315 effect on the fibres, with discrete sugar glass films being apparent. 316 matrices 319 Adenovirus vaccine vectors were formulated in 0.5M TS and dried on the five selected 'new' 320 matrices, with glass fibre (S14) and polypropylene (J200) matrices as comparators of known 321 performance. Dried matrices were thermochallenged for a week at 45 o C prior to reconstitution and 322 infectivity titration. Marked thermostabilisation performance differences between the matrices 323 were apparent (Table 1). 324 As previously observed, the glass-fibre matrix (S14) showed minimal (less than 0. Standard sugar formulation (0.5M trehalose sucrose) was loaded into each matrix and dried at room 336 temperature and < 5% relative humidity. Dried samples were then subjected to modulated 337 differential scanning calorimetry (DSC) to measure glass transition temperature (Tg) onset 338 temperature and enthalpic recovery of the sugar glass formed on the matrices. The glass transition 339 temperature (Tg) indicates the temperature at which a low mobility sugar glass changes to a highly 340 mobile rubbery state and is known to be related to product stability in dry formulations 32 . Enthalpic 341 recovery is a measure of energy dissipated as a glass progresses through equilibrium and can reflect 342 molecular rearrangement during storage or physical ageing 33 . not correlate with thermostability ( Figure 3A). A possible correlation was observed between 345 thermostabilisation performance and high enthalpic recovery ( Figure 3B, r 2 =-0.70, p=0.02), as seen 346 with glass fibre matrices conjugate pad and S14 followed by polypropylene (J200) based matrix. 347 There was substantial variation in the residual moisture content of products dried under the same 348 conditions on different matrices ( Figure 3C), with lowest residual moisture in the best performing 349 matrix, S14. Recovery of a model protein (lysozyme) after desiccation and reconstitution did not 350 predict thermostabilisation performance ( Figure 3D). 351

352
Characterisation of S14 glass fibre matrix 353 Given our inability to identify a commercially-available matrix suitable for clinical application, we 354 turned our attention to detailed characterisation of the best-performing S14 matrix, with a view to 355 future production of a similar matrix from biocompatible materials. 356 We initially performed a more extensive characterisation of the physical properties of the matrix, 357 with results as shown in Table 2. 358 Use of chemical binders is common in the production of nonwoven fabrics such as S14 to provide 359 strength and desirable surface properties. In view of the possibility that a binder might be 360 contributing to S14's thermostabilisation performance, we investigated whether such a binder could 361 be identified in the matrix. 362 Scanning electron microscopy demonstrated film-like material which could represent binder 363 covering fires in some sections of the S14 sample ( Figure 4A). Differential scanning calorimetry and 364 thermogravimetric analysis (Figures 4B-C) demonstrated the presence of a material with a melting 365 point of 220 o C and a degradation temperature at 260 o C. infrared (FTIR) spectroscopy of the extract provided a fingerprint spectrum, which matched closely 368 with the expected spectrum of polyvinyl alcohol (PVA) ( Figure 4D) 34,35 . 369 The solubility and other properties of PVA vary widely according to molecular weight (MW) and 370 degree of hydrolysis, and so we sought to further characterise the presumed PVA extracted from 371 S14. We used dynamic light scattering (DLS) to compare the hydrodynamic radius of the PVA extract 372 to those of PVA samples of known MW. The results were consistent with a MW in the range of 35 373 kDa ( Figure 4E). 374 We then used 1H NMR to estimate the degree of hydrolysis of the polymer. The 1H NMR spectrum 375 obtained ( Figure 4F) was consistent with PVA, with a 2:1 ratio of the areas under the peaks at ~1.6 376 ppm and 3.9 ppm (corresponding to hydrogens in the CH2 and CH environments respectively). 377 Results obtained using a range of standards of varying percentage hydrolysis showed a clear 378 relationship of the size of a peak at 2.05 ppm to the acetyl group content (the presence of which, in 379 a sample of PVA, indicates incomplete hydrolysis). The spectra of the completely hydrolysed 380 standard and the S14 extract were similar, with only a trace of a peak in this area, and so we 381 concluded that the PVA extracted from S14 is likely to be completely hydrolysed. 382

384
Having identified PVA in the best-performing matrix, S14, we investigated whether the PVA may 385 function not only as a binder but might actually contribute to vaccine thermostabilisation. 386 We reasoned that treatment with PVA might enhance the thermostabilisation performance of 387 relatively poorly-thermostabilising matrices. We observed that application of PVA extracted from 388 glass fibre (S14) significantly improved thermostabilisation performance of two of the three tested 389 matrices, as assessed by infectious virus recovery after a four week 45 o C thermochallenge. Recovery 390 from the polyester matrix (Leukosorb) was enhanced by 1.3 log10-fold, while enhancement from the cellulose matrix (ET CHR) was enhanced by 2 log10-fold ( Figure 5A). Addition of PVA to a glass fibre matrix (Pall's conjugate pad) was not beneficial. This matrix already contains a binder (Fig 2F), 393 possibly PVA. 394 We proceeded to test whether the degree of hydrolysis of PVA had any impact on vaccine 395 thermostability at 45 o C for an extended period of 28 days. We tested polyester (Leukosorb), 396 polyamide (33100L) and a custom-made glass fibre-based matrix after treatment with PVA extracted 397 from S14 and two other commercially sourced polymers (Sigma) which were similar in size but 398 differed in degree of hydrolysis (30-70 KDa / 87-90% hydrolysed, and 31-50 KDa/ >99% hydrolysed). 399 We again observed substantial improvements in thermostabilisation performance, exceeding a 1 400 log10-fold increase in infectious virus recovery after thermochallenge, when PVA was applied to 401 matrices which did not contain PVA at baseline (polyamide and polyester) ( Figure 5B). The greatest 402 enhancement was seen with the S14 extract, followed by the fully hydrolysed PVA, with the least 403 enhancement seen with the incompletely hydrolysed PVA ( Figure 5B Having characterised the geometry and composition of S14, and the contribution of PVA to its 408 thermostabilisation performance, we proceeded to develop a bespoke matrix 'in-house'. Although 409 our ultimate aim is to produce biocompatible matrices, we sought as an initial step to test whether 410 the understanding we had gained would enable us to produce an 'in-house' wet-laid glass fibre 411 matrix which could replicate S14's thermostabilisation performance. The new matrix had structural 412 properties similar to those of S14 (area density 50.6±1.5 g.m -2 , thickness 0.56±0.02 mm, mean flow 413 pores 25±2.3 µm, absorption capacity 11.1±0.3 g/g, composed of borosilicate glass fibres with 414 diameter 3.6±1.8 and length 1.3±0.6 mm; see table 2 for data relating to S14), and a similar 415 appearance ( Figure 5C). Although thermostabilisation performance of the custom-made matrix did not exactly match that of S14 ( Figure 5D), it was closer than had previously been achieved with any of the other tested matrices (Fig 5B). As expected, further modification of the matrix with additional 418 PVA had no effect. 419

420
The starting point for the present study was the observation, in our previous work, that the non-421 biocompatible glass-based S14 matrix out-performed a polypropylene matrix 26 . This posed the 422 question of which properties of the matrix could be relevant for SMT performance, and whether a 423 more suitable matrix than S14 could be identified for clinical translation. 424 Our initial attempts to find a suitable commercially available matrix yielded disappointing results. 425 Adenovirus stability on the tested matrices was poor (table 1). Multiple variables differed between 426 each matrix, and we were therefore unable to perform experiments to clearly isolate the effect of a 427 single variable. There was not a clear relationship between characteristics of the matrices and their 428 stabilisation performance (Figures 2 and 3), with the possible exception of high enthalpic recovery of 429 sugar glass (which is not a parameter which can readily be 'designed in' to a new matrix). 430 Our ability to use analysis of commercially available matrices to draw conclusions to guide design of 431 new biocompatible matrices was thus limited. We therefore changed our strategy, seeking to 432 characterise S14 in detail and produce similar matrices 'in-house', allowing us to identify features of 433 S14 contributing to its stabilising performance. 434 Most significantly we found that PVA, present on the S14 matrix as a binder, appears to contribute 435 to the stability of adenovirus (Figures 4 and 5). It was shown that fully hydrolysed PVA, similar to 436 that we extracted from S14, was most beneficial in thermostabilisation ( Figure 5). Polyvinyl alcohol is 437 potentially suitable for use as an excipient in vaccine formulations: it is 'generally regarded as safe' 438 (GRAS) and is also a FDA approved inactive ingredient for parenteral use 36 . PVA has previously been 439 explored as an excipient in a number of studies of bio-macromolecular stability. It has been found to be beneficial in some formulations of proteins, including insulin 37 , but benefit has not been seen 441 consistently in other studies 38 39 40 . PVA may contribute to protein stability by hydrogen bonding of 442 hydroxyl groups in PVA to the proteins. In dried protein formulations, PVA has also been shown to 443 prevent deamidation more potently than another widely used polymeric excipient, polyvinyl 444 pyrrolidone 41 . 445 We now intend to develop the SMT method towards clinical application. There are two principal 446 obstacles to this goal: robust and GMP compliant execution of the process, and the availability of a 447 suitable matrix for GMP production. Robustness and GMP compliance of the process are clearly 448 interlinked. We believe several of the challenges of GMP execution of the SMT process can be 449 addressed by execution of drying within a lyophilizer without freezing or the application of vacuum: 450 existing large-scale GMP lyophilisation facilities could provide the necessary controlled temperature, 451 low humidity, aseptic environment. With respect to robustness, we have found in recent work that 452 the stability achieved by the process can be highly sensitive to deviations from the intended 453 conditions and, more troublingly, some unexplained inconsistency in performance can occur: work is 454 ongoing to address these issues. With respect to the matrix, GMP execution is likely to require 455 development of a new non-woven. In addition to replicating the stabilising performance of S14, such 456 a matrix needs to be biocompatible, to have good mechanical integrity (in particular, without 457 shedding fibres into the reconstituted product), and to be produced in line with the quality 458 requirements for a GMP raw material. We are now using the data provided by the present study in 459 order to develop such a matrix, mimicking the physical properties of S14 and making using of the 460 beneficial effect of PVA, but without the problematic use of glass.  Each panel relates the thermostabilisation performance of the various matrices (after one-week 485 thermochallenge at 45 °C, Y-axis, data shown in Table 1)  shown. 503

Figure Legends
Panel D shows the sample spectrum obtained from attenuated total reflectance Fourier transform 504 spectroscopy (ATR-FTIR) for the binder recovered from glass fibre (S14) (black), and a reference 505 library spectrum for PVA (grey). 506 Panel E shows estimation of the molecular weight of the binder extracted from glass fibre (S14) by 507 DLS. Points and solid line show a standard curve generated using PVA of known MW. Dashed lines 508 indicate the hydrodynamic radius of the PVA extracted from S14 (11nm) and the inferred MW (36 509 kDa). Points and error bars indicate the mean and range respectively of duplicate DLS 510 measurements. 511 Panel F shows 1H NMR spectra used to estimate the percentage hydrolysis of the binder extracted 512 from glass fibre (S14). The upper spectrum (labelled S14) is that of the extract, with unknown 513 percentage hydrolysis and hence an unknown percentage of monomers bearing acetyl groups. The 514 five spectra below were obtained using standards prepared by proportionately mixing 80% and 515 100% hydrolysed PVA to achieve a range of acetyl group content ranging from <0.1% up to 20% (as 516 per labels to left of panel). The X-axis indicates chemical shift measured in parts per million (ppm) 517 and the Y-axis shows relative intensity.