Introduction

Over the past 20 years, nanomedicines have attracted significant attention due to their potential for diagnosis, prevention, and treatment of various diseases [1, 2]. Contrary to bulk structures or biological and chemical molecules, these nanoscale materials present diverse advantages [3,4,5,6,7]. Among all varieties of multifunctional nano-based drug delivery systems, self-assembled spherical liposomes, characterized by a bilayer phospholipid membrane surrounding an aqueous core, have received considerable attention owing to unique physicochemical and biological properties as well as the tremendous potential for drug delivery applications using both hydrophilic and hydrophobic drugs [8,9,10]. These liposomal formulations change the pharmacokinetic properties of the encapsulated drugs (e.g., doxorubicin, daunorubicin, or vincristine) resulting in improved safety and increased therapeutic index [1, 8]. Over the past years, many liposome formulations (e.g., doxil, DaunoXome, Myocet, Marqibo, or Onivyde), comprised of various lipids have been approved by medical agencies [1, 11,12,13]. However, the optimization of liposome characteristics during drug development is still challenging despite great advances in formulation strategies.

According to previous reports, liposomes can be produced using a wide range of methods, including thin-film hydration, sonication, extrusion, reverse-phase evaporation, freeze-thaw process, or high-pressure homogenization for fabrication of multi- or unilamellar vesicles [13, 14]. All these methods suffer from similar drawbacks, including poor batch-to-batch reproducibility, risk for sample contamination, non-scalability, or lack of temperature control which increases the risk of lipid degradation [14]. To overcome these limitations, microfluidics-based methodologies have been introduced as an alternative method to produce liposomes based on nano-precipitation and self-assembly reactions during millisecond-controlled mixing at the nano-liter scale [15,16,17]. This reproducible and robust strategy offers the possibility to reduce both production time and the economic costs of developing advanced formulations [14]. However, several parameters influence the formulation optimization and characteristics of resulting liposomes. For example, it is necessary to optimize microfluidic flow rate settings to produce liposomes with sizes suitable for biomedical applications.

In this study, we present a systematic approach based on a design-of-experiment (DoE) model for the rapid bottom-up formulation of well-defined liposomes by microfluidic mixing using a staggered herringbone chaotic micromixer (SHM). The DoE was used to systematically optimize experiments, resulting in a minimal number of formulation runs. In addition, the DoE enabled us to assess the influence of factors important to the microfluidics process as well as their impact on the characteristics of liposomes. As a proof-of-concept, we selected the lipid compositions of five clinically approved liposomes (i.e., doxil, DaunoXome, Myocet, Marqibo, and Onivyde) and combined microfluidics with DoE to determine optimal flow rate settings for the production of liposomes with specific sizes similar to the marketed products. The developed microfluidics prediction models were finally used to incorporate hydrophobic gold nanoparticles (HGNs) into the lipid bilayer without affecting the size of resulting liposomes. Empty liposomes and HGN-loaded liposomal nanocomposites were characterized using dynamic light scattering (DLS), Fourier-transform infrared spectroscopy (FT-IR), and transmission electron microscopy (TEM).

Materials and methods

Materials

The lipid hydrogenated soy phosphatidylcholine (HSPC), 1,2-distearoyl-sn glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol)-2000] (DSPE-PEG2000), and 1,2-distearoyl-sn-glycero-3- phosphorylcholine (DSPC) were obtained from Avanti Polar Lipids (Alabaster, USA). Egg yolk phosphatidylcholine (EPC) and sphingomyelin (SM) were purchased from Lipoid AG (Steinhausen, Switzerland). Cholesterol, Dulbecco’s phosphate-buffered saline (D-PBS), absolute ethanol (EtOH), tetraoctylammonium bromide, dodecanethiol (DDT, 98%), uranyl acetate solution, gold (III) chloride trihydrate (HAuCl4-3H2O), sodium borohydride (98%), tetraoctylammonium bromide (98%), methanol, chloroform, and toluene were obtained from Sigma-Aldrich (Buchs, Switzerland).

Preparation of empty clinical liposomes using microfluidics

To prepare and optimize clinical liposome formulations, a microfluidics approach based on the NanoAssemblr™ benchtop instrument (Precision NanoSystems Inc., Vancouver, Canada) was used. The microfluidic cartridge had two inlets, which were merged to a micro-channel. One inlet was used for the lipid mixtures dissolved in EtOH. Initial lipid concentrations introduced to the micromixer were adjusted to increase dilution effect by increasing FRRs to result in final lipid concentrations of 2 mg/mL. Using this strategy, post-processing steps (i.e., sample concentrating) to match lipid concentrations of different runs could be avoided, making the process more time-efficient [16, 18, 19]. The other inlet contained an aqueous buffer (D-PBS, pH 7.4). Lipids and buffer were mixed under laminar flow in the SHM structure of the microfluidic cartridge in order to form unilamellar liposomes. Disposable syringes (1 mL for EtOH and 3 mL for aqueous solutions) suitable for the fluid inlet connectors were used. The NanoAssemblr™ allowed the control of TFR (4–12 mL/min) and FRR (1.5:1 to 5.5:1 D-PBS:EtOH ratio). Optimization of clinical liposome formulations (Table 1) using microfluidics was based on a full factorial DoE (Stavex 5.2, Aicos Technologies, Basel, Switzerland). The experimental design shown in Table 1 was used for the formulation of each clinical liposome composition listed in Table 2. In a last step, liposome formulations were purified using dialysis (12–14 kDa; Medicell membranes Ltd., London, UK) against Poncini water to remove residual solvent (room temperature and overnight stirring at 150 rpm).

Table 1 Clinical liposome formulations approved by medical agencies [13]. Liposome formulations are listed according to year of clinical approval. The major structural lipid of each liposome composition is italicized and the corresponding phase transition temperature (Tm) is given
Full size table
Table 2 Design-of-experiment (DoE)-based microfluidic flow settings. Different flow rate ratios (FRR) and total flow rates (TFR) used for the formulation development of each clinical liposome formulation are represented
Full size table

Physicochemical characterization of liposomes

Size and polydispersity index (PDI) of liposomes were determined by DLS using a Malvern Zetasizer Nano ZS (Malvern Instruments, Worcestershire, UK). Particle size and zeta potential were measured in aqueous solution at 20 °C.

Stability of liposomes

The stability tests were performed for 1 month. The samples were stored at 4 °C under a static condition (without stirring), and at specific time points, size and PDI were analyzed by DLS. All experiments were performed in triplicate.

DoE approach

To analyze the influence of TFR and FRR on the physicochemical characteristics of liposomes formulated using microfluidics, a DoE was used (MODDE software, Umetrics). Results from three independent experiments were analyzed using both multiple linear regressions (MLR) and partial least square (PLS) regression to optimize the prediction models. The collected data was used to determine the statistical significance between model coefficients. The prediction error was minimized using the least squares regression analysis. The analysis of variance (ANOVA) was carried out to validate models based on R2 and Q2 as the goodness of fit and prediction, respectively. In addition, the model validity and reproducibility were determined. The model and replicate errors were investigated by lack of fit (LOF) test.

HGN synthesis

HGNs were synthesized using the protocol previously reported by Rasch et al. [26]. Briefly, 20 mL of a 41.65-mM HAuCl4 solution in H2O was mixed with 80 mL of a 150-mM tetraoctylammonium bromide solution in toluene as an organic phase by stirring at 600 rpm for 60 min. The purple organic phase was collected by extraction, and the colorless aqueous phase was discarded. A 0.6-mL DDT was then injected into the toluene phase to achieve a 3:1 ratio of thiol groups to Au atoms. After 15 min, the toluene phase turned colorless and then 20 mL of a 10-mM sodium borohydride solution was added rapidly, resulting in a dark brown microemulsion. After 12 h of stirring, the brown toluene phase was extracted and centrifuged at 8000 rpm for 5 min along with methanol as anti-solvent (1:4 toluene to methanol ratio) to precipitate the gold nanocrystals. Gold nanocrystals were dispersed in chloroform and size-selective precipitation was performed by centrifugation at 8500 rpm for 10 min to collect any poorly dispersed or aggregated nanocrystals. The supernatant was gently mixed with 250 μl of methanol and then centrifuged at 8500 rpm for 5 min. The process of adding methanol, centrifuging, and separating the supernatant from precipitates was repeated several times, and the precipitates collected in steps 5–7 were used for nanocomposite formulations.

HGN characterization

The size and shape of HGNs were determined by TEM imaging (CM-100; Philips, Eindhoven, Netherlands) [27]. Since toluene is able to degrade carbon-coated copper grids, toluene was completely removed by evaporation under N2 and HGNs were dispersed in EtOH. After that, 5 μL of samples in EtOH was deposited on a 400-mesh grid after plasma treatment for 30 s. The diameter of at least 100 HGNs was measured for size. The qualitative analysis of synthesized HGNs was performed using a FTIR spectrometer to assess functional groups on the surface. The FTIR spectra in the range of 550–4000 cm−1 were obtained using a FTIR ATR-IR (Bruker Tensor 27).

HGN encapsulation into liposomes

Lipid stock solutions and HGNs were dissolved using 40% THF in EtOH at a lipid to HGN ratio of 0.107 (w/w) [28]. This lipid gold nanocrystal solution was rapidly mixed with an aqueous solution (D-PBS; pH 7.4) as described for preparation of empty clinical liposomes using the Nanoassemblr™ benchtop instrument. FRR and TFR were selected based on DoE prediction models. Nanocomposites were purified using dialysis (12–14 kDa) to remove residual solvent.

HGN-liposome nanocomposite characterization

TEM analysis was used to detect HGNs in liposomes as described above [27]. In brief, after sample deposition, grids were washed with water and then incubated for 10 s with a 2% uranyl acetate solution. The excess of uranyl acetate was removed and the samples were dried overnight at room temperature before imaging.

Statistical analysis

All experiments were performed in triplicate. Mean and standard deviation were then calculated. To access statistical significance for paired comparisons, a one-way ANOVA combined with Bonferroni’s multiple comparison was used. Significantly different results are indicated for p values lower than 0.001(*). All calculations were performed using Origin Pro 8.6 software (Origin Lab Corp., MA, USA).

Results and discussion

In recent years, several groups have demonstrated that microfluidics is a reproducible, scalable, cost-effective, and simple method for the production of various nanoparticles. Rotational flow and wrapping of fluid streams in microfluidic channels results in rapid mixing of aqueous and organic fluids [15]. Small unilamellar liposomes can be formed in milliseconds due to the polarity increase in the SHM channels that facilitates a nano-precipitation reaction followed by supersaturation and self-assembly of lipid molecules [16]. Importantly, key physicochemical properties of liposomes, such as size and PDI, are highly influenced by flow rate settings during microfluidic-based formulation. Thus, optimization of the entire process is difficult and time-consuming when a one-factor-at-a-time method is used.

Here, we propose a systematic approach based on DoE for the rapid optimization of liposome formulations. The complete optimization procedure includes five steps: (i) design of microfluidic settings based on a full-factorial DoE, (ii) formulation of liposomes using specific flow rate settings for each run, (iii) physicochemical characterization of resulting liposomes, (iv) determination of DoE prediction models, (v) formulation of liposomes with specific sizes using predicted flow settings. To demonstrate the validity of our strategy, five different liposome compositions based on clinically approved liposome formulations, i.e., doxil, DaunoXome, Myocet, Marqibo, and Onivyde [21,22,23,24,25], were selected (Table 1).

The DoE design space presented here is based on nine different runs with systematic variations in FRR and TFR (Table 2). This strategy offers the possibility to assess the effect of both FRR (from 1.5:1 to 5.5:1) and TFR (4 to 12 mL/min) on physicochemical characteristics of liposomes (size, PDI, and zeta potential) using a minimal number of runs. Notably, this design can be expanded to any other parameter of interest. After DoE-based specification of microfluidic settings (Table 2), different lipid mixtures based on clinically approved compositions were prepared in EtOH (Table 1, Figs. 1a and 2b). Microfluidic flow settings were adjusted for each run and liposomes were prepared by mixing of ethanolic lipid solution with D-PBS. Figure 1c schematically illustrates the design space to investigate the influence of FRR and TFR on liposome size based on a combined microfluidic-DoE approach.

Fig. 1
figure 1

Schematic representation of rapid microfluidic-based liposome formulation development. a Liposomes based on clinically approved lipid compositions are formulated using a staggered herringbone micromixer (Staggered herringbone micromixer adapted from Belliveau et al. [15]). b Chemical structures of different lipids used in clinical liposome formulations. c Schematic representation of design space investigated during microfluidic-based full factorial DoE approach. Influence of fFRR and TFR on the physicochemical characteristics of liposome was determined

Full size image
Fig. 2
figure 2

DoE-based optimization of clinical liposome formulations using microfluidics. TFR and FRR were altered to investigate their effect on hydrodynamic size and size distribution (polydispersity index, PDI) of resulting liposomes. DoE-based prediction models are represented as two-dimensional contour plots with color codes indicating liposome sizes from small (blue) to large (red) and PDI values from low (blue) to high (red). Specific sizes and PDI values are indicated in boxes in each design space

Full size image

Previous studies have shown that microfluidics is a robust method to precisely control TFR and FRR between aqueous and solvent inlet streams [16]. Interestingly, all procedures for the formation of the presented liposomes were performed in less than 1 h, demonstrating that microfluidics is highly time-efficient in contrast to other common liposome preparation methods, such as film rehydration-extrusion (Table 3).

Table 3 Expert opinion on the most important advantages and disadvantages of top-down versus bottom-up formulation strategies. Preparation of liposomes using conventional film rehydration-extrusion method as compared to rapid microfluidics mixing technique [18, 29, 30]
Full size table

In all experiments, zeta potential of resulting liposomes did not change significantly with varying TFR and FRR. This indicates that surface properties of different liposome compositions are maintained despite varying microfluidic settings. The zeta potential for each formulation is presented in Table S1. The predicted DoE models for flow setting-dependent size or PDI of liposomes with different lipid compositions are shown in Fig. 2. For all formulations, the smallest hydrodynamic diameter was achieved at medium FRR. Smaller FRRs caused an increase in liposomal size. Notably, design spaces could be grouped according to lipid-phase transition temperature and surface characteristics. Lipid compositions containing surface-modifying lipids (i.e., doxil, Onivyde) or lipids with low-transition temperature (i.e., Myocet) resulted in smaller liposomes at defined FRR as compared to other lipid compositions (i.e., Marqibo, DaunoXome). This result was also confirmed by resulting PDI values. Size distributions of doxil, Onivyde, or Myocet were always lower than 0.25, whereas PDI values of Marqibo and DaunoXome increased significantly at low FRRs. This might be explained by the reduced rate of diffusion within the micromixer channels at low FRRs [14]. Taken together, these results confirmed that FRR is a key parameter to control size and PDI of liposomes, as previously reported for other nanoparticles [14, 29, 31, 32].

In sharp contrast, TFR did not significantly influence the size of resulting liposomes (Fig. 2 and Fig. S1a) [16, 33]. However, there was a significant change in PDI with increasing flow rate for liposomes containing DSPC, i.e., Onivyde and DaunoXome (Fig. 2 and Fig. S1b). At medium FRR, increasing TFR resulted in smaller size distributions. The different PDI levels for DaunoXome and Onivyde can be explained by the presence of DSPE-PEG, which results in homogenous liposome formation. The constant PDI at varying TFR for doxil in comparison to Onivyde might be due to the usage of HSPC, which consists of a lipid mixture with varying fatty acid tail lengths, and, thus, results in homogenous particle formation even at low flow rates (i.e., 4 mL/min). These results confirmed that lipid compositions highly influence the features of liposome [34]. Importantly, the size and PDI for all liposomes remained constant for TFRs above 8 mL/min. This behavior of vesicle formation at increased TFR demonstrates the ability of microfluidics as high-throughput formulation method, which is one of the key aspects for large-scale manufacturing of liposomes.

The statistical analysis of the DoE prediction models indicated that the MLR model has a higher fit for all the tested compositions as compared to the PLS model. The optimized model was determined according to R2 and Q2 values presented in Table S2. A smaller difference than 0.3 between R2 and Q2 represents an index for a good model fit. The significant coefficients determined in the model are shown in Fig. S2. As an example, the FRR × FRR term in the Myocet size model confirmed the significant impact of FRR on the size of Myocet-based liposomes formed in the microfluidics channel. In addition, Table S2 shows the ANOVA analysis, demonstrating the statistical significance and validation of each model generated. All models confirmed the significant impact of FRR on liposome size and PDI, while varying TFR mainly resulted in a difference of production rate.

After demonstrating that a combined microfluidics DoE approach is a rapid and controllable strategy for the production and optimization of liposomes, we also assessed the long-term stability of liposomes formulated by microfluidic manufacturing at a storage temperature of 4 °C. Clinical liposomes were prepared by microfluidics at different FRR and TFR as described in Table 2 and stored at 4 °C for 1 month. All liposome formulations except doxil-based liposomes showed good size stability over the course of storage and liposome sizes remained mainly unaffected (Fig. 3). In the case of doxil, limit size liposomes were produced [35], which resulted in a minor size increment during storage and also alteration of size distribution. Notably, changes in PDI during storage were remarkable for samples prepared at lowest TFR (Fig. S3). According to the stability tests, homogenous liposomes (i.e., without significant differences in vesicle size) showed a higher long-term stability after preparation at higher FRR. This is in agreement with a report by Correia et al. [33] who showed improved stability of liposome formulations with increased FRR. In addition, it has previously been shown that liposomes with 1:1 lipid:cholesterol ratio have an ideal stability [36, 37] confirming our observations in the case of the Myocet formulation with a 1.1:0.9 EPC:Chol ratio. Other groups reported that a 2:1 ratio between lipids and cholesterol might increase liposome stability (e.g., in DaunoXome and Onivyde) [38]. Figure 3 illustrates the change in size for diverse lipid:cholesterol ratios and presence of PEG in lipid compositions over time.

Fig. 3
figure 3

Storage stability of clinical liposomes formulated by microfluidics. Liposomes were stored at 4 °C for 28 days. The bright and dark diagrams represent the sizes of liposomes after preparation (t(0)) and after storage at 4 °C for 28 days (t(28)). Run numbers indicate the microfluidics settings (i.e., FRR/TFR) based on Table 2

Full size image

Based on our results, both claims are valid and the optimal lipid-to-cholesterol ratio for storage stability of liposomes manufactured by microfluidics technology is between 1:1 and 2:1.

To validate our DoE-based formulation strategy, we used the flow rate settings identified during the DoE to incorporate HGNs into liposomes with a defined size (Fig. 4). Gold nanoparticles have attracted great attention in the field of diagnosis and therapy due to their unique physicochemical properties, facile synthesis pathway, and various functionalization capabilities [27, 39,40,41,42].

Fig. 4
figure 4

Validation of DoE prediction model. a Schematic representation of microfluidics-based encapsulation of hydrophobic gold nanoparticles (HGNs) into a clinical liposome formulation (Staggered herringbone micromixer adapted from Belliveau et al. [15]). b Representative transmission electron micrograph of HGNs. c Fourier-transform infrared (FTIR) spectrum of HGNs functionalized with dodecanethiol (DDT). Transmission electron microscopy (TEM) analysis of d empty liposomes and e HGN nanocomposites. Single HGNs are clearly visualized within liposomes (black arrows)

Full size image

Therefore, we synthesized HGNs with diameters between 1.5 and 3 nm for the incorporation into liposomes (Fig. 4a, b) [26]. The FTIR analysis of HGNs (Fig. 4c) showed peaks at 720 cm−1 and 2850–2900 cm−1, corresponding to C-S and CH2–CH2 symmetrical and unsymmetrical stretching, respectively. The S-H vibration peak at 2572 cm−1 was not observed, suggesting the formation of Au-sulfide bonds and replacement of H with Au in S-H groups of DDT. These results confirmed the successful synthesis of ideal HGNs for the incorporation into the liposomes.

HGN-liposome composites were formulated by rapid mixing of an ethanolic solution containing HGNs/lipids and an aqueous phase (D-PBS) at optimized FRR and TFR. Notably, HGNs were not dispersible in pure EtOH and addition of at least 40% THF was needed for adequate dispersion [28]. Resulting HGN-liposome nanocomposites had a similar hydrodynamic diameter as compared to empty clinical liposomes with sizes of 111 ± 17.3 nm and 109.3 ± 15.3 nm, respectively (Fig. 4d, e). Electron microscopy analysis confirmed the incorporation of HGNs (dark spot) into liposomes without formation of HGN clusters (Fig. 4e). Recently, it has been shown that hydrophobic alkyl chains of phospholipids and the presence of cholesterol can improve hydrophobic association of nanoparticles in the bilayer space and the entrapment efficiency [27, 43, 44]. Therefore, it is tempting to speculate that HGNs associate with hydrophobic lipid tails in the lipid bilayer due to a strong thermodynamic driving force [26]. The HGN-lipid ratio was adjusted to result in encapsulation of a single HGN per liposome. Thus, highly efficient mixing in the microfluidic channels allowed efficient loading of HGNs without affecting physicochemical characteristics of liposomes. Importantly, flow rate settings selected based on DoE models resulted in predicted diameters demonstrating the potential of our appraoch to formulate liposomes with specific physicochemical characteristics. This facilitates the process development resulting in reduced time and costs. In addition, this bottom-up formulation strategy is generally considered as less harsh in comparison to conventional top-down methods for the preparation of unilamellar liposomes based on mechanical disruption of multilamellar vesicles [16]. In our opinion, the key advantage of microfluidics is the continuous process which allows for a much easier scale up and parallelization when compared to the conventional methods (Table 3).

Conclusion

In this study, we demonstrate that combining microfluidics with DoE is a robust and rapid strategy to formulate liposomes with well-defined and reproducible size. DoE analysis confirmed the statistical significance of flow rate settings on the resulting size and PDI of liposomes. The determined models could efficiently be used to predict liposome characteristics as a function of microfluidic parameters (i.e., FRR and TFR). Thus, the size of liposomes formulated by microfluidics could be easily adjusted via simple changes of flow rate settings based on design spaces. This strategy presents an interesting alternative to time-consuming traditional methods for the formulation of liposomes. Advantages including simple screening and capability of facile scale-up make the presented approach suitable for many applications. In the future, this might facilitate the screening of various nanocarriers during preclinical development and may open up a path towards broad applicability of advanced technologies in the preparation of conventional drug delivery systems.