Purification of Nanoparticles by Size and Shape

Producing monodisperse nanoparticles is essential to ensure consistency in biological experiments and to enable a smooth translation into the clinic. Purification of samples into discrete sizes and shapes may not only improve sample quality, but also provide us with the tools to understand which physical properties of nanoparticles are beneficial for a drug delivery vector. In this study, using polymersomes as a model system, we explore four techniques for purifying pre-formed nanoparticles into discrete fractions based on their size, shape or density. We show that these techniques can successfully separate polymersomes into monodisperse fractions.

Purification by centrifugation. The initial step of polymersome purification by size involved removal of micelles from the solution using the KrosFlow filtration system. The polymersomes were centrifuged at 500 Rotational Centrifugal Force (RCF) for 20 minutes. The resulting pellet was removed and resuspended in PBS. This fraction contained the largest aggregate fraction. The supernatant was then re-centrifuged at 2000 RCF for 20 minutes and the pellet was removed and re-suspended, constituting fraction 1. This was repeated with further 20-minute centrifugations at 5000, 10000, 15000 and 20000 RCF.
Purification by GPC. For separation of polymersomes by GPC, micelles and aggregates were removed as described above and the remaining polymersome solution was concentrated to approximately 200 μ L using a 500 kDa MicroKros filter module. The solution was then placed in a glass liquid chromatography column containing Sepharose 4B. The fractions were collected in a 96-well plate. Dynamic Light Scattering (DLS) measurements were performed on a Zetasizer Nano ZS (Malvern Ltd.) as described previously 12 . Purification by density gradient centrifugation. Density gradient centrifugation was performed using solutions of 5, 10, 15, 20, and 25% w/v sucrose dissolved in PBS. Aliquots of 200 μ L of each solution were carefully layered one on the top of each other in the order from the densest to the least dense within a 1.5 mL microcentrifuge tube, while avoiding mixing between the layers. Finally, 150 μ L of the solution containing a mixture of rhodamine-labelled vesicles of different shapes (spheres and tubes) prepared by film rehydration was deposited at the top of the sucrose layers and the microcentrifuge tube was centrifuged at 20000 RCF for 2 hours. After centrifugation, 20 μ L of solution from each layer was collected and analysed by Transmission Electron Microscopy (TEM).
Polymersome characterisation. TEM analysis was performed using a FEI Tecnai G2 Spirit electron microscope and/or a JEOL 2100 operating at 200 kV equipped with a CCD camera Orius SC2001 from Gatan. Copper grids were glow discharged and the sample was adsorbed onto the grid. The sample was then stained with 0.75 wt% phosphotungstic acid (PTA) adjusted to pH 7.4 with NaOH. All the TEM analyses were carried out with dried samples.
DLS analyses were carried out using a Zetasizer Nano ZS (Malvern Ltd.) at a copolymer concentration of 0.25 mg/mL. DLS measurements were based on 12− 14 runs, 10-second sub-runs. Samples were analysed at 25 °C with a scattering angle of 173° and a 633 nm HeNe laser based on a material refractive index (RI) of 1.59, a dispersant refractive index of 1.330 and a viscosity of 0.89.

Results and Discussion
Why use polymersomes as a model for nanoparticle purification? In this study, we have chosen polymeric vesicles, polymersomes, as a nanoparticle model for demonstrating the efficiency of different purification techniques. We have previously shown that the average polymersome size is influenced by the preparation technique employed, but each method produces a wide distribution of sizes around their mean diameters 20 . The most energetically favourable geometry for a polymersome (and any other vesicle) is a sphere. However, the process of membrane wrapping can only occur above a critical radius typically controlled by the size of the single building block 21 . Above this critical radius, the energetic penalty required for an amphiphilic membrane to wrap into a vesicle is not dependent on the final diameter 19 . This means the self-assembly of spherical polymersomes is not associated with a strong size selective bias. The method of aggregate formation can also influence the assembled nanoparticle shape and structure 22 . The final morphology of nanoparticles depends on the distribution of molecules across the two monolayers. Membrane wrapping can result in different non-spherical shapes that are stabilised by the high molecular weight of block copolymers 23 . One way to improve homogeneity of vesicle size and shape is to provide additional energy to break and re-form the vesicles several times. For example, phospholipid-based vesicles, liposomes, are commonly homogenised by a combination of sonication and extrusion in order to break up larger structures and reduce the size diversity. However, these techniques are often infeasible for more mechanically robust nanoparticles such as polymersomes. The high molecular weight of polymers used in polymersome preparations ensures the mechanical resistance and therefore breaking of the polymersome membranes would require involvement of substantial forces 24 . Here we use pH-sensitive block copolymers PMPC-PDPA to form polymersomes using two methods: (i) solvent switch, exploiting the PDPA sensitivity to pH, and (ii) film rehydration.
The former method of self-assembly, referred to as pH-switch, is a version of a solvent switch method, where the solvent's pH is altered. It is a simple and quick procedure that facilitates the formation of polymersomes in less than an hour. The polymersomes are produced by dissolving polymeric unimers, and by subsequently changing the solubility conditions to drive self-assembly. Under aqueous conditions, with a pH above the polymer pKa, the unimers come together to form structures that shield their hydrophobic blocks from water. As we observed previously, this initial stage is relatively fast and initially leads to the formation of spherical micelles (monolayered polymeric aggregates without an aqueous interior) regardless of the copolymer hydrophobic/hydrophilic ratio 12 . For membrane-forming copolymers the transition from micelles to vesicles (double layered membranes enclosing an aqueous interior) occurs through the formation of different intermediate phases mostly via a growth mechanism 25 . However, the critical condition for such a growth is that there is an equilibrium between unimers and assemblies. When the pool of unimers in solution is exhausted, the self-assembly process arrests often leading to the formation of metastable structures such as micelles. Conversely, if a vesicle forms and there is still a pool of unimers in solution, the vesicle starts growing with a misbalance between the inner and the outer layer leading to the formation of high genus vesicles 22 .
For film rehydration methods, the formation of polymersomes occurs via the hydration of a pre-formed copolymer film. During the hydration process, the copolymer forms different structures depending on the copolymer/water ratio 10 . This process is kinetically complex and often leads to the formation of different metastable phases such as multilamellar aggregates 26 , as well as tubular polymersomes 27 .
This broad variety of structures make polymersomes an ideal platform to develop new methods of separation as their properties are highly dependent on their size and shape.
Cross flow-filtration. One method of purifying polymersomes based on size is filtration. Dead-end filtration is generally unsuccessful because nanoparticles concentrate at the pores resulting in a "filter cake" that blocks the filter. Cross flow-filtration (CFF) overcomes this problem by providing a flow of particles at a tangent to the pores under high pressure. The pressure allows particles smaller than the pore size to permeate the membrane while the tangential flow prevents the formation of a filter cake (see the representative scheme in Fig. 1A) 28 . After producing polymersomes by the pH-switch method, we used a CFF equipped with membranes with a pore size of 50 nm in diameter to efficiently purify spherical and genus polymersomes from micelles, as shown by DLS and TEM in Fig. 1B. However, the same approach evaluated with DLS and TEM was not applicable to the separation of polymersomes (vesicles) into distinct sizes and shapes. This could be due to the intrinsic soft nature of polymersomes, which can deform in the proximity of the pores if a specific threshold pressure is applied. Although DLS provides a rapid measurement of a sample size distribution, it is important to note that DLS results can be less accurate in samples with polydisperse or agglomerated nanoparticles. Similarly, a single TEM image may not be completely representative of the entire sample. Therefore, it is good practice to include both measurements when assessing particle size. In general, DLS data presented here were in agreement with TEM images. Polymersomes purified from micelles with filtration using the KrosFlo Research IIi System were observed as a polydisperse mixture of sizes in the TEM image, and similarly the DLS reported a relatively high polydispersity index (PDI) of 0.283. Differential centrifugation. Centrifugation is one of the most commonly used techniques in bioscience research, mostly for separation of mixtures by size and density. Differential centrifugation (DC) is used to separate multiple fractions within a sample. In this method, the mixture is centrifuged multiple times, and after each run the pellet is removed and the supernatant is centrifuged at a higher centrifugal force (Scheme in Fig. 2A).
In our experimental approach, polymersomes were formed by the pH-switch method. Figure S2 in supporting information shows a TEM picture of these polymersomes before carrying out any further purification steps. It is evident that the resulting sample was highly polydisperse in terms of both size and shape, and contained a large population of micellar structures. Micelles have been thus preliminary removed using the previously described CFF system. The remaining polymersomes were then separated into different size fractions by DC. TEM micrographs for each fraction after the DC, and their respective DLS analyses, can be seen in Fig. 2B. This method was quick, simple and resulted in a successful separation of the polydisperse sample into distinct, more homogenous size fractions without any loss of material. However, further experiments showed that DC method cannot be used to separate different shapes of polymersomes as the high-genus structures were observed in multiple fractions. Recently, we have observed that the low RCF fractions of polymersomes formed by film rehydration and purified by DC are enriched in tubular polymersomes 29 . However, the separation was not complete and, therefore, the level of purification was not satisfying. For this reason, we focused then on size exclusion chromatography.
Size exclusion chromatography. The next separation technique explored was size exclusion chromatography (SEC), which separates particles based on their hydrodynamic volume. As a mixture moves through the stationary phase of a SEC column, typically made of packed hydrophilic polymeric beads, smaller molecules meander in and out of pores within the gel, whereas larger molecules cannot penetrate into the pores, allowing them to pass through the stationary phase quicker and elute from the column earlier. SEC is commonly used to separate free small molecules from those encapsulated within nanoparticles, however, the use of the method for the separation of distinct sizes and shapes of nanoparticles can be more challenging 30 . To improve the resolution of SEC, the polymersome solution was concentrated into a small volume using the KrosFlo Research IIi System with a 10 kDa pore size membrane. This reduced the time required for the liquid to be absorbed by the stationary phase. The eluted liquid was passed twice into the same column, doubling the effective column length. The eluted sample was then collected into a 96-well plate and the separated fractions' mean diameter were calculated from the DLS measurements. The data is reported using each sample scattering intensity (indicative of the polymersomes size and concentration) and the mean size (diameter) as a function of the elution volume. The relative frequency size distribution of each fraction of polymersomes is plotted in Fig. 3A. The resulting chromatogram is shown in Fig. 3B that suggests the effective separation of the polymersomes into several sizes. This technique was effective at separating the sample into many discrete monodisperse size fractions. Examples of three fractions separated by this method, and analysed by TEM, are shown in Fig. 3C. The monodispersity of the fractions separated by SEC was improved as compared to the samples purified by the DC technique. However, this method is more time consuming and may result in a material loss within the SEC column.
We thus came to the preliminary conclusion that CFF can be an ideal tool for separating micellar structures from vesicles. At the same time, both DC and SEC were found to be optimal for purifying polymersomes according to their size, with their own advantages and disadvantages. However, none of these techniques allowed us to carry out a complete separation by nanoparticle shape. Each samples still contained high level of high genus structures (in the case of polymersomes preparation via pH-switch method), or tubular vesicles (in case of a film rehydration method preparation). We thus explored another technique, namely density gradient centrifugation (DGC), a method widely employed in biochemistry and molecular biology to successfully separate sub-cellular fractions and organelles, according to their density. Density gradient centrifugation. Finally, we tested DGC as a technique for purification of polymersomes according to their shape. Sucrose DGC allows a mixture to be separated into different fractions based on the particle density. Non-spherical polymersomes tend to have a larger density than spherical ones as the relative internal aqueous volume is smaller and hence they posses a larger polymer/water ratio. To test this, a sample of polymersomes formed by film rehydration was placed on a preformed discontinuous sucrose density gradient by layering successively lower sucrose densities into a 1.5 mL microcentrifuge tube. The layers contained 0% (i.e., the polymersomes solution in PBS), 5%, 10%, 15%, 20% and 25% w/v sucrose in PBS. The sample was then centrifuged for 2 hours at 20000 RCF yielding nanoparticles of different densities suspended in each layer. Polymersomes formed by pH-switch were recovered from the 5% w/v, and 20% w/v sucrose layers (all other fractions did not contain nanoparticles). The 5% w/v sucrose fractions, displayed in Fig. 4A (top), contained a mixture of spherical polymersomes with different sizes. Particles recovered from the 20% w/v sucrose layer contained predominantly large high genus structures (Fig. 4A, bottom). An additional electron micrograph of the 20% w/v sucrose fraction is shown in the Supplementary Information (Fig. S3). The shape and the fold of genus (1-fold, double fold, triple fold, etc.) of the nanoparticles in the 20% w/v sucrose layer varied. The differential density between the two populations can be easily explained by comparing the ability to encapsulate water with the spheres having larger volumes than the high-genus structure. This means that a 100 nm sphere has half the number of copolymer chains compared with a 100 nm torus (i.e. 1-fold genus) with a 10 nm hole. Thus, the particle density of spherical polymersomes is always lower than tubular and genus polymersomes.
In a similar way to the pH-switch, the film rehydration technique can result in the formation of a mixed vesicle population of different shapes. However, film rehydration of a polymeric film results in the formation of tubular vesicles instead of genus structures, in addition to spherical vesicles. The tubular vesicles are transitional structures that form during rehydration of the polymer film, and eventually "pearl" into smaller tubes and then further into spherical polymersomes 11 . The higher particle density of tubular polymersome membranes should enable these structures to be purified from spherical polymersomes using DGC. Therefore, this possibility was explored using DGC methodology as described earlier for the separation of genus and spherical particles, however, this time, nanoparticles were prepared using the film rehydration method. The post-centrifugation density gradient for rhodamine-conjugated PMPC-PDPA polymersomes prepared by film rehydration is shown in Fig. 4B. The method yielded four separate fractions. The fraction at 0% to 5% w/v sucrose concentration (Fig. 4B, top) contained the smallest spherical polymersome dispersion, the 10% w/v sucrose fraction (Fig. 4B, middle) contained monodisperse spherical polymersomes, the 15% w/v sucrose fraction (Fig. 4B, middle) contained a mixture of spherical and short elongated polymersomes, and finally the 20% w/v sucrose fraction contained long branched and un-branched tubular polymersomes. In order to quantify the size distribution of the individual fractions isolated using DGC, the TEM images were analysed using ImageJ software. In particular, the collected 0% fraction contained polymersomes having an average diameter range of 10-20 nm. The vesicles present in the 5% w/v fraction were 20-40 nm in diameter, those in the 10% w/v fraction were 40-60 nm, while in the 15% w/v sucrose band the polymersomes were 60-80 nm, although there was also a significant number of vesicles in the diameter range of 90 to 150 nm (see Fig. S4 in supporting information). The broad size distribution of the 20% w/v fraction is due to the presence of tubular vesicles (Fig. S4). The last graph of Fig. S4 displays an overview of all the analysed samples. The DGC has proven to be a very effective method that allowed to purify the film rehydrated nanoparticles into the distinct shape fractions. As shown in Supplementary Information (Fig. S4), these nanoparticles were initially a highly polydisperse mixture of tubular and spherical polymersomes with varying diameters. The DGC provided an easy and effective way of separating these structures with minimal loss of sample during the process, and enabled the sample to be concentrated at the interface between two different sucrose bands. This is because the particles get trapped at the density equilibrium.

Conclusions
Purifying nanoparticles into monodisperse fractions is essential to understand the physical properties of nanoparticles, such as size and shape, as well as to minimise the variability in biological applications. For purification of nanoparticles by size, other groups have explored methods such as magnetic field flow fractionation filtration 31 , size exclusion chromatography 32 , size selective precipitation 33 , density gradient centrifugation 34 , and cross-flow filtration 35 . Similarly, methods for purifying nanoparticles by shape include centrifugation 36 , gel electrophoresis 37 , and density gradient centrifugation 38 . However, these reports have predominantly focused on "hard" nanoparticles such as gold nanoparticles and carbon nanotubes. Moreover, many of these methods cannot be applied to "soft" nanoparticles of tens to hundreds of nanometres in diameter.
The interesting physico-chemical and mechanical properties of polymersomes have attracted several application-focused studies. However, the unique polymersome stability, mechanical robustness and the dynamic self-assembly processes are also the major obstacles in the preparation of homogenous samples. Formation of polymersomes by self-assembly inherently leads to the relatively broad size distribution and high variability in the morphology. Non-vesicular structures, such as micelles, as well as tubes and genus, form alongside spherical vesicles. With the aim of consistency and reproducibility in biological and medical applications, it is of paramount importance to introduce new approaches enabling the production of homogenous polymersome preparations. Here, we have compared four different, yet complementary, techniques to separate polymersomes according to their size and shape. CFF, DC and SEC provided a good solution for the size-dependent separation of particles with each technique bearing different advantages and disadvantages. The CFF method is efficient at separating micelles from polymersomes possibly due to a combination of size exclusion and differential fluid dynamics. Although it cannot be used to separate sub-populations of polymersomes according to their size, the CFF method does provide a means of concentrating a polymersome sample, and holds promises in producing consistent scaling up protocols. The DC and SEC-based methods enabled separation of polymersomes of different sizes into distinctive sub-populations of less than 10 nm difference in diameter. However, these purification methods result in a loss in sample concentration. While neither CFF, DC nor SEC allowed for shape-based separation, DGC achieved this goal. We have observed that the different shape of polymersomes correspond to changes in the density of membrane packing, providing the means for their separation by the DGC-based method.
In this study, block copolymer polymersomes have been used as a nanoparticle model for investigations of various purification techniques with the aim to efficiently separate particles by size and shape. Depending on the type of nanoparticles employed, the ideal purification protocol may combine two or more of these techniques. Once nanoparticle solutions are purified, each fraction can be quantified. Therefore, these techniques not only allow samples to be purified to extract the nanoparticles with the desired physical properties, but may also allow different nanoparticle assembly techniques to be assessed in order to determine which techniques produce nanoparticles with the desired physical properties.
Exploring various separation methods, we have demonstrated it is possible to obtain a homogenous polymersome preparation, both in terms of particle size and shape. This work could provide the basis for more consistent results in biological experiments and improve application development in medicine and drug delivery.