Quality by design approach identifies critical parameters driving oxygen delivery performance in vitro for perfluorocarbon based artificial oxygen carriers

Perfluorocarbons (PFCs) exhibiting high solubility for oxygen are attractive materials as artificial oxygen carriers (AOC), an alternative to whole blood or Haemoglobin-based oxygen carriers (HBOCs). PFC-based AOCs, however, met clinical translation roadblocks due to product quality control challenges. To overcome these issues, we present an adaptation of Quality by Design (QbD) practices to optimization of PFC nanoemulsions (PFC-NEs) as AOCs. QbD elements including quality risk management, design of experiments (DoE), and multivariate data analysis facilitated the identification of composition and process parameters that strongly impacted PFC colloidal stability and oxygen transport function. Resulting quantitative relationships indicated a composition-driven tradeoff between stability and oxygen transport. It was found that PFC content was most predictive of in vitro oxygen release, but the PFC type (perfluoro-15-crown-5-ether, PCE or perfluorooctyl bromide, PFOB) had no effect on oxygen release. Furthermore, we found, under constant processing conditions, all PFC-NEs, comprised of varied PFC and hydrocarbon content, exhibited narrow droplet size range (100–150 nm) and narrow size distribution. Representative PFOB-NE maintained colloidal attributes upon manufacturing on larger scale (100 mL). QbD approach offers unique insights into PFC AOC performance, which will overcome current product development challenges and accelerate clinical translation.


Results and discussion
Here, we present QbD-based development adapted to PFC-NE-based AOCs with two PFC molecules commonly reported from clinical studies: perfluoro-15-crown-5-ether (PCE 2,3 ) and perfluorooctyl bromide (PFOB 13,39 .) These two molecules have been extensively used in regenerative medicine preclinical studies and have the most well established biodistribution kinetics in vivo 18 . The present PFC-NEs are formulated with medium chain triglycerides (MCT), Miglyol 812, to impart colloidal stability and a solubilizer, transcutol, to impart drug loading capacity for later use.
Definition of product quality attributes and target specifications. As an essential preliminarystage action and part of the QbD guidelines, product quality was defined by identifying CQAs and assigning acceptable specifications for each attribute. This step implemented measurable metrics to assess the degree of success of the oxygen carriers. CQAs were selected to reflect nanoemulsion colloidal stability and oxygen delivery performance ( Table 1). The specifications or limits were chosen from a combination of prior knowledge and relevant literature reports 1,18,22,23,37,38 . For example, it is widely accepted that the larger a nanoparticle's size, the more readily it is taken up by monocytes and macrophages. This results in rapid clearance from the body, which is undesirable in oxygen carriers. In summary, development of an oxygen carrier with both stability and oxygen loading has been an ongoing challenge; thus, defining product quality in this stage facilitated the development www.nature.com/scientificreports/ of a stable product proficient in oxygen transport. Definition of CQAs and target specifications makes up only part of the much broader quality target product profile within the QbD methodology, which is not within the scope of this report.
Production, characterization, and quality control of PFC-NEs. Freshly prepared PFC-NEs were uniform, opaque liquids which ranged from 95 to 145 nm in diameter with polydispersity index < 0.2 and negative zeta potential. All formulations were periodically monitored for changes in size and size distribution for 30 days to measure colloidal stability during refrigeration. Additionally, samples were exposed to 50 °C for 2 weeks to simulate accelerated stability conditions, and size was recorded after exposure. The size change was converted to percent of initial size for analysis. This so-called thermal storage stability was of interest to verify stability was maintained during prolonged temperature rise. Ostwald ripening destabilization is governed by solubility, molar volume, and diffusion coefficient of the dispersed phase material 40,41 . All of these exhibit a temperature dependency that could accelerate mass transfer and droplet size growth. Following a detailed study reported in the literature of the colloidal behavior during autoclaving of triglyceride nanoemulsions stabilized by Poloxamers along with other reports of the prevalence of Ostwald ripening in PFC nanoemulsions 42 , it was expected that Ostwald ripening is the predominant mode of destabilization 43 . In the present study, three nanoemulsions exhibited size change beyond the specification for thermal storage. Samples were also sterile filtered through 0.22-μm-pore syringe filter and separate samples were centrifuged; in both cases samples retained droplet size and polydispersity. These two steps simulated sterilization and cell culture treatments, respectively. Table 2 provides all composition variables and processing conditions, and Table 3 tabulates CQA evaluations for all nanoemulsions. Supplemental Figures S1 and S2 give a graphical summary of the nanoemulsion characterization.
Evaluation of biphasic PFC-NEs as design controls to triphasic systems. Next, we evaluated the oxygenation performance of biphasic nanoemulsions (BNs), (Fig. 1B, right configuration). We formulated biphasic PFC-NEs as comparison systems to triphasic nanoemulsions (TNs), (Fig. 1B, left configuration). Oxygen release measurements were conducted using an experimental setup illustrated in Fig. 1C. As oxygen partitioned from the NE droplets, it passed through the release medium, where dissolved oxygen levels were measured, and finally equilibrated into the headspace of the flask. Figure 1D details the compartments and boundaries across which oxygen equilibrated. The resulting oxygen release data, Fig. 1E, reflected the initial accumulation and subsequent dissipation of oxygen in the release medium. Full details appear in Materials and Methods. While this experimental setup demonstrated oxygen transport in a closed system well enough to allow us to differentiate between samples having different oxygen affinities, it is important to acknowledge that oxygen would be consumed by a living system and shear forces in the vasculature space would likely affect the convective diffusion. Oxygen release was quantified and compared by extracting a maximum concentration from each dissolved oxygen release profile (C max ), illustrated in Fig. 1E for biphasic nanoemulsions. This overlay compares a biphasic nanoemulsion comprised of 14.3%vol PCE (BN-PCE) to a PFC-free nanoemulsion (internal phase comprised of MCT at the same volume fraction) and to deionized water (containing no internal phase). Water as a control was important as it simulated the effect of using UW solution, the current gold standard in organ preservation 44 . The BN-PCE exhibited an increased oxygen transport activity compared to PFC-free NE and deionized water controls. The BN-PCE had a C max of ~ 4 × and ~ 5 × those of PFC-free and water control. The 14.3%vol biphasic PFOB nanoemulsion (BN-PFOB) also demonstrated higher oxygen transport activity compared to the controls. These data support the potential of PFC-NEs in oxygen delivery. www.nature.com/scientificreports/ Further evaluation of the oxygen transport activity revealed the linear nature of the oxygen affinity for PFC material. Figure 1F shows this relationship among all 8 TNs with respect to theoretical volume fraction of PFC. Expanding the search radius to include both BNs and TNs (TN3-TN8 and BN1, Fig. 1G) and employing quantified PFC concentration demonstrated the linear relationship was preserved. This suggested that the hydrocarbon shell had little effect on the oxygen release. Taking the oxygen release findings at face value suggested the biphasic PFC nanoemulsion was the optimal formulation. However, evaluation of the colloidal stability highlighted the risk associated with the biphasic formulation. Table 3 and Supplemental Figure S1 present key stability issues observed with biphasic formulations compared to triphasic nanoemulsions. Taking stability to be indicated by droplet size change and polydispersity changes over time and in response to accelerated conditions, it was readily noticed that BN-PFOB destabilized rapidly during shelf-storage conditions at 4 °C. BN-PCE maintained size through one month of storage, but did not endure the thermal storage test. Because both BNs failed to meet all colloidal stability CQA specifications, an alternative formulation was needed.

Risk management and design of experiments.
Previously collected data 37 indicated that formulation modifications increased stability of PFC nanoemulsions. Lambert and Janjic 37 demonstrated that inclusion of MCT shielded PFC-NEs from destabilization in serum-and salt-rich media. Other reports demonstrated similar claims, including supplementing with an insoluble oil decreases the Ostwald ripening rate 45,46 . This information provided guidance in our formulation development of stable PFC oxygen carriers, and we subsequently modified the composition to include an additional hydrocarbon component. These triphasic PFC nanoemulsions have been considered for a variety of applications (e.g. small molecule anti-inflammatory and imaging reagent  www.nature.com/scientificreports/ carriers). Here triphasic nanoemulsions (TNs), comprised of a hydrocarbon phase (MCT and transcutol) and either PFOB or PCE, were investigated for their oxygen release performance in vitro. It was anticipated that adding a hydrocarbon oil could compromise the activity of the oxygen-rich PFC phase, so subsequent product development applied concepts of QbD to investigate the (i) colloidal stability and (ii) oxygen transport activity of this triphasic configuration (PFC-in-HC-in-water). The schematic in Fig. 2 illustrates the conceptual balance of these two opposing attributes. Data throughout the text reinforce that understanding and controlling this balance is crucial to the development of an AOC. To provide direction and structure within this formulation space in this stage of product development, risk management followed by statistical design of experiments played a central role in evaluating triphasic nanoemulsions.
A risk management strategy called failure mode, effects, and criticality analysis (FMECA) examined and ranked potential sources and mechanisms of CQA specification failure. Full details on FMECA risk calculations appear in Materials and Methods. This methodology identified that composition-related factors were the most critical failure modes. Figure 3 summarizes the initial risk assessment. Here, the average risk priority number (RPN) for all failure modes is collected for each CQA, highlighting the risk associated with the collective average of entries for each of four broad failure modes categories (composition, oxygenation, processing, or operator Step-by-step in vitro oxygen loading and release experiments. The oxygen carrier sample is loaded by bubbling oxygen and then passing several passes of oxygen over the headspace. Meanwhile, release medium is deoxygenated by bubbling N 2 and then passing several passes of N 2 over the headspace. Oxygenated sample is removed by needle and syringe and injected into sealed release medium. Dissolved oxygen levels are monitored in the release medium (ChemDraw). (D) Schematic showing simplified sequence of interfaces that oxygen equilibrates across in the in vitro oxygen release setup. After loading the PFC-NE with O 2 , the majority of the dissolved oxygen resides in the PFC phase. There is no O 2 in the release medium or the headspace of the round bottom flask. A concentration gradient drives the oxygen from the PFC core first into the hydrocarbon shell in the case of TN, then into the release medium, and finally into the headspace of the vessel. (ChemDraw) (E) Representative oxygen release data shows that PFC presence is essential for high levels of oxygen loading. (GraphPad Prism) (F, G) Dissolved O 2 maximum concentration (C max ) is linearly proportional to both theoretical PFC concentration, in percent volume, and PFC concentration quantified by 19  A key objective is to find a balance between these opposing quality attributes. Conceptual scatterplots (formed using JMP Pro) were formed by normalizing CQAs C max and Thermal stability Δsize into a number ranging from 0 to 1 by dividing each by the maximum respective CQA measurement (Balance graphic drawn in ChemDraw).  www.nature.com/scientificreports/ error). In the unabridged FMECA, each entry is broken down further by the specific failure mode causes. For example, inappropriate hydrocarbon: perfluorocarbon (HC: PFC) ratio (belonging to the broad failure mode inappropriate composition) causes oxygen release to fall out of specification because the hydrocarbon shell limits the oxygen transfer rate at high ratios.. To appreciate a more comprehensive look at the failure modes, cause and effect diagrams are shown in Supplemental Figure S6 for Thermal stability Δsize and C max in which sources of variability for both responses appear in expanded detail. FMECA results directly informed which factors appeared in the experimental design. Because FMECA assigned highest risk to composition-related factors, a d-optimal design of experiments selected five composition factors and one processing factor to study. Design factors and levels are provided in Table 2.
In this screening experimental design, PFC-NE composition was represented by five factors: PFC type, continuous media type, internal phase (IP) volume fraction, hydrocarbon: perfluorocarbon ratio by volume (HC: PFC ratio), and proportion transcutol in IP by volume. Three of these: IP volume fraction (X1), HC: PFC ratio (X2), and proportion transcutol (X4) reflected the absolute and relative concentrations of IP components. X1 was the volume of internal phase components divided by the total NE volume. This was a measure of emulsion droplet concentration. In all cases, surfactant amount was adjusted such that all emulsion formulations had identical surfactant to oil ratios.
X2 was a unitless quantity indicative of relative volumes of PFC (core compartment) and HC (shell compartment) components in the internal phase. The higher the ratio, the lower PFC amount in the formulation. This also meant the hydrocarbon shell surrounding the PFC phase was thicker, which possibly imparted resistance to oxygen release. X4 was the volume fraction of the internal phase that was comprised of transcutol, a versatile co-solubilizer. Our group previously used transcutol to increase loading of celecoxib by 10 times the dose of a nanoemulsion delivery platform which had effect treating rat chronic pain model 38,47 . When adjusting X4, it was added at the expense of the MCT oil. These three were chosen as they collectively reflect the concentration of internal phase components and could be varied in DoE to reduce multicollinearity in statistical regression analysis. PFC type (X5) and continuous medium (X6) were categorical formulation factors. Perfluorooctyl bromide (PFOB) and perfluoro-15-crown-5-ether (PCE) are nonvolatile perfluorocarbon oils (boiling points are 142 and 145 °C, respectively). Number of passes (X3) was the single processing parameter chosen to study based on our prior work 18,37,38 . It was anticipated that nanoemulsion droplet size would impact the oxygen release profile by affecting the surface area available for oxygen exchange. Adjusting the number of re-circulations on the microfluidizer was a direct way to manipulate droplet size while keeping composition and concentration constant. This 6-factor, 2-level DoE was suited to screen and identify factors active in impacting attributes. Table 3 shows that triphasic nanoemulsions (TNs) exhibited a small but consistent size reduction and increase in size distribution compared to biphasic nanoemulsions (BNs). Further, a discrepancy in size and PDI was seen between PFC types. This is highlighted in a scatterplot of size vs. PDI and cluster analysis which defined three distinct clusters (Fig. 1A). Zeta potential remained negative, but four TNs exhibited zeta magnitude lower than the CQA specification. Even so, it was evident that TNs exhibited improved colloidal stability compared to the BNs. All triphasic NEs maintained size in both centrifuge and filtration test. Only one TN failed to meet thermal stability Δsize specification and none failed to meet the day 30 Δsize specifications. TN3, the triphasic formulation that failed to meet thermal stability spec, consisted of the lowest amount of MCT. These observations suggested that electrostatic repulsion was not the principal mode of stabilization and that MCT acted as a stability enhancer.

Evaluation of Triphasic PFC nanoemulsions.
After observing the corresponding shift in colloidal stability with the introduction of MCT oil, oxygen transport activity of TNs was assessed. Presented in Fig. 4 are notable dissolved oxygen concentration-time comparisons. Maximum concentration (C max ) was attained on the order of one minute into the release for all samples. The C max numbers observed in this experiment matched those reported by others 50,51 in similar experimental setups. Panel A compares three TNs consisting of different levels of PFC concentration for each PFC type along with the BNs. The C max increased with increasing PFC amount, and all PFC-NEs had higher C max than water. Figure 4B-D each compare PFC type. Figure 4B contrasts PFC type for BNs. There was a difference in C max for PFC type for BNs. Figure 4C displays TN samples with different PFC types for 7.7%vol and Fig. 4D displays TN samples with different PFC types for 5%vol. PFC type had no detectable effect on the oxygen release of TNs. This apparent contradiction may be explained by considering the poor colloidal stability of BNs . Droplet size of BN-PFOB rapidly grew in storage, whereas BN-PCE maintained size for 1 month (see Fig. 5E-F as well as Supplemental Figure S3.). This size increase would decrease the surface area available for release and could account for the apparent difference in C max between PFC type in the BN comparison. Figure 4E-F compare samples of like PFC and like volume fractions, but different PFC concentrations. In each case, the sample with higher PFC concentration exhibited higher C max . This supports the idea that the hydrocarbon shell dissolved less oxygen and may have obstructed the release of oxygen from the PFC core. Figure 4G compares TNs comprised of different volume fractions. The oxygen release was significantly greater for the sample with higher volume fraction for nearly six minutes of the release and remained, on average, higher for the duration of the release. This effect was attributed to the higher amount of material in the NE. This trend is supported by Fig. 1F-G. Although the TN exhibiting the higher C max contained 1.8 × higher PFC concentration (5.0 vs. 2.8%vol), the C max was less than 1.4 × higher (1.546 vs. 1.162 mg/L) than the TN consisting of less PFC. This was because the aqueous continuous phase and hydrocarbon compartment of the internal phase were saturated with oxygen and contributed to the release. Figure 4H shows the TN-PFOB (5% vol) compared to HC biphasic nanoemulsion (PFC-free). These two NEs had the same volume fraction, and thus made a clear comparison between PFC and HC material. Figure 4I compares TN-PFOB (5% vol) to water. These comparisons showed a higher C max for the TN-PFOB and confirmed that PFC material was essential for oxygen transport. Overall, the oxygen release was www.nature.com/scientificreports/ highly affected by PFC content. This was observed in NEs containing higher overall volume fraction (Fig. 4G), and in NEs comprised of equal volume fractions but differing in concentration of PFC (Fig. 4H). It is worth noting that exchanging perfluorocarbon type had negligible effect, suggested by Fig. 4C-D the linear nature of the C max -PFC loading relationship (Fig. 1F-G). Further, the presence of salt in the continuous media did not impact the oxygen release profile. Evaluation of triphasic nanoemulsions revealed virtually identical oxygen release profile between PFC type (Fig. 4C-D) and slight discrepancies in thermal stability Δsize between PFC type (Fig. 5A) that appeared to be a result of low MCT amount. Additionally, it was observed that C max and thermal stability Δsize were positively correlated (Fig. 5B). This finding represented the key challenge and was central in supporting other reports of PFC oxygen carriers. Recognizing the stability differences between PFC types, we further evaluated colloidal stability of triphasic PFC-NEs. While bubbling gas through a nanoemulsion was a simple way to perform oxygen loading, we wanted to ensure this step would not mechanically disturb the emulsion and lead to destabilization. Figure 5C-D are representative size distribution measurements taken before and after the oxygen bubbling step. It was observed that both TNs maintained size distribution. While no significant difference between PFC types in C max was observed, we wanted to verify the PFC loading was not different. Figure 5I shows representative 19 F www.nature.com/scientificreports/ NMR quantification, suggesting PFC loading in triphasics was similar for two PFC types. NMR spectra corresponding to Fig. 5I are shown in Fig. 5G-H.

Regression model construction and evaluation.
Multiple linear regression (MLR) was utilized to fit models that described C max and Thermal stability Δsize in terms of the DoE factors. Reduced MLR models for both C max and Thermal stability Δsize appear in Table 4. The terms X1, X2, X3, and X4 were present in the reduced C max model. While the oxygen solubility governed how much oxygen can be loaded and released, PFOB and PCE had similar enough oxygen affinities to not see a significant difference between their oxygen transport activities. This was first suggested by data in Fig. 4C and corroborated by elimination of X5 from the model. Similarly, X6 was removed during regression analysis. While salinity affects gas solubility in water, the main carrier of oxygen in the nanoemulsion was the PFC phase. Therefore, the reduction in oxygen loading in the aqueous phase when adjusting for tonicity resulted in negligible oxygen-carrying loss. Estimates for X1 and X2, terms which represented the amount of PFC, had positive and negative signs, respectively. Consistent with data shared in previous sections, higher C max resulted from pairing higher internal phase fraction (positive X1) with lower HC:PFC ratio (negative X2), ultimately yielding more PFC content. Further, C max was dependent on X3. More passes resulted in more surface area available for oxygen transfer.  19  www.nature.com/scientificreports/ For the reduced thermal stability Δsize model, X1, X2, and X4 appeared as model terms. X2 and X4 collectively determined the amount of MCT. As anticipated, high X2, where MCT oil is high, suppressed size growth and high X4, where MCT oil is reduced, promoted size growth. Both models had these three terms in common, but they had opposing effects in each response's model, supporting data in Fig. 5B.
The formulation space described by regression models for C max and Thermal stability Δsize is summarized in Fig. 6 along with regression term estimates (B-C) and actual by predicted plots (D-E). Panel A shows a plot overlaying contours of both responses in formulation space X1 and X2. The shaded blue and red regions represent where C max and Thermal stability Δsize, respectively, were outside of specifications, highlighting this opposing nature of the two responses. Model goodness-of-fit was reflected by the appearance of the data in observed by predicted plots falling along the diagonal 1:1 line in Panels D and E.
In the triphasic formulations, the makeup of the dispersed phase (i.e. volume fraction, proportion transcutol, HC: PFC ratio) was the most critical feature for both responses. C max and thermal stability were both described by these factors, but with opposing outcomes. Importantly, the DoE regression analysis provided quantitative relationships describing both responses. This allowed simultaneous monitoring of both CQAs during product development to achieve a PFC nanoemulsion formulation that satisfied both CQA specifications while minimizing material and time resources. In addition, analysis highlighted X3 impacted C max but not Thermal stability Δsize. This key finding indicated processing could be tuned to enhance oxygen transport without affecting colloidal stability.

Scale-up of PFOB nanoemulsions.
To demonstrate scalability, one of the more PFC-rich TNs (TN2) and a BN of like PFC type (BN1) were chosen to be produced at 4 × scale (100 mL). Figure 7B-C shows that the scaled-up emulsions maintained monomodal size distribution. Because producing nanoemulsions in larger batches is essential for moving into pre-clinical and clinical phases of development, this finding was encouraging. Figure 7A shows the dependence of size on number of passes for 100-mL NEs. The biphasic formulation had a steeper dependence on processing time compared to the triphasic. This indicates the triphasic NE was produced at a specified size with less processing than the biphasic NE, which was important to acknowledge alongside the finding that more recirculations improves C max without affecting Thermal stability Δsize.

Conclusions
We have presented an implementation of systematic pharmaceutical product development methodology, QbD, to formulation of PFC-NEs as AOCs. Through this process, we identified critical quality attributes that reflected colloidal stability and oxygen delivery performance. Furthermore, quality risk management was implemented to enable rational formulation development where both composition and processing parameters were evaluated. The risk assessment guided creation of DoE used to construct regression models which described two competing quality attributes, oxygen release and colloidal stability. We found that the triphasic NE colloidal stability exceeded colloidal stability of biphasic NE, while the biphasic NEs demonstrated highest oxygen loading. Further, increase in PFC loading led to increase in oxygen loading. We showed that presented nanoemulsions remained stable under oxygen exposure in vitro (Fig. 5C-D). We also established scalable processing parameters which was deemed successful by producing 100 mL-scale nanoemulsions.
This study in its current form presents certain limitations. First, only two commonly used perfluorocarbons were used. Second, all oxygen release studies were performed in vitro under very controlled conditions. Future studies will expand to assessing presented PFC-NEs in organ preservation testing. Due to small sample size, we were not able to fully quantify the impact of hydrocarbon layer on the oxygen loading. Our data suggested that regardless of hydrocarbon presence, the loading capacity of presented nanoemulsions was directly linearly correlated to PFC content levels. www.nature.com/scientificreports/   Failure mode, effects, and criticality analysis. Formulation and manufacturing of nanoemulsions was examined one unit operation at a time. In each unit operation, all sources of variability of critical quality attributes were identified and treated as a failure mode. All failure modes in all critical quality attributes were evaluated by assigning a Risk Priority Number (RPN), which was calculated as 52,53 : Severity, frequency of occurrence, and detectability associated with a failure mode were scored on a scale of 1-5, low to high risk. Specifically, 'no appreciable consequence to batch quality' , 'has not happened' , and 'readily detected' corresponded to a score of 1 for severity, frequency of occurrence, and detectability, respectively. 'Total batch loss' , 'happens regularly' , and 'not detectable within current unit operation' corresponded to a score of 5 for severity, frequency of occurrence, and detectability, respectively. Accordingly, the higher the RPN, the more risk associated with the corresponding failure mode.

Design of experiments.
Five composition factors and one processing factor were each varied at two levels in a D-optimal screening design 54 of experiments (DoE) consisting of 8 experimental runs created using JMP Pro 14.3.0 (SAS, Cary, NC, USA). In this design, the impact of all 6 main effects on emulsion attributes was statistically evaluated. Composition factors included PFC type, continuous media type, internal phase (IP) volume fraction, hydrocarbon-to-perfluorocarbon ratio by volume (HC: PFC ratio), and proportion transcutol in IP by volume. The lone processing factor was number of recirculations (passes) through microfluidization interaction chamber. Data processing and regression model construction. Regression models were constructed using multiple linear regression. Using backwards stepwise variable selection and selecting P value < 0.05 as the significance threshold, model terms with P > 0.05 were removed one at a time to form reduced regression models. Coefficient of determination (R 2 ), root mean square error (RMSE), and leave-one-out cross validation (LOOCV) were used to assess model goodness-of-fit. ANOVA and model fit statistics are given in full detail in Supplemental Tables S1-S2.
Production of triphasic nanoemulsions. All nanoemulsions were produced following earlier published protocols 18,33,36,38 . Briefly, a micelle solution of blended non-ionic surfactants (2% Pluronic P-105, 3% Pluronic P-123 w/v) was made by following previously reported protocol with minor modifications 55  www.nature.com/scientificreports/ aliquoted and stored at 50 °C for two weeks. All colloidal stability tests were begun on samples 1 week or less after preparation. Supplemental Figures S1 -S5 show graphically the results of all tests. Individual unpaired, two-tail t-tests were done on the quality control tests and are annotated on the bar graphs in Supplemental Figures S1 and S2. DLS data were summarized by mean ± standard deviation of 3 measurements unless otherwise specified.
Oxygen release kinetics. 2.5 mL of sample (AOC or control) was dispensed into a 100-mL 3-necked round bottom flask under magnetic stirring. Oxygen gas was bubbled into sample through glass Pasteur pipette for 30 min. Oxygen gas was then passed over the sample to fill the headspace volume with pure oxygen and the flask was sealed from the surrounding environment with rubber stoppers. Meanwhile, deionized water (25 mL) in a 100-mL 3-necked round bottom flask under magnetic stirring as the release medium was degassed with nitrogen gas until the submerged dissolved oxygen probe (HI5421 Dissolved Oxygen Benchtop meter Hanna Instruments, Woonsocket, Rhode Island, USA) gave a stable reading of zero oxygen in the aqueous release medium. Nitrogen gas was passed over the water to remove any residual oxygen in the headspace, and the flask was sealed from the surroundings with rubber stoppers. Oxygenated sample (1 mL) was removed with syringe and needle and injected into flask containing degassed release medium and dissolved oxygen probe. Dissolved oxygen levels were monitored and recorded over time using the HI5421 probe. From the resulting dissolved oxygen-time profiles, a maximum concentration (C max ) was extracted for each sample and used as a measure of oxygen loading and release. Statistical comparisons were conducted by performing unpaired two-tailed t-tests of dissolved oxygen measurements at each time. All experiments were done in triplicate on samples less than 1 week after preparation.

F Nuclear magnetic resonance characterization. Nuclear magnetic resonance (NMR) spectroscopy
was used qualitatively in all PFC nanoemulsions and to measure the concentration of 19 F nuclei in representative nanoemulsions. 19 F NMR chemical shifts were reported as ppm using a 0.4% (v/v) trifluoroacetic acid (TFA) solution as internal reference with chemical shift set at − 76.0 ppm. PCE nanoemulsions showed a singlet peak at − 92.4 ppm and PFOB nanoemulsions showed a triplet peak at − 83.7 ppm chosen for quantitation. The resulting spectral peaks were integrated, and the resulting areas were used to calculate the mean 19

Data availability
Document named Raw DO compiled and controls_Fileshare contains raw dissolved oxygen measurements from in vitro oxygen release from all design points and controls. Each measurement is average of three independent experiments. Document named FMECA_O2-NE contains the unabridged failure mode, effects, and criticality analysis with risk priority number rankings.