Monitoring drug nanocarriers in human blood by near-infrared fluorescence correlation spectroscopy

Nanocarrier-based drug delivery is a promising therapeutic approach that offers unique possibilities for the treatment of various diseases. However, inside the blood stream, nanocarriers’ properties may change significantly due to interactions with proteins, aggregation, decomposition or premature loss of cargo. Thus, a method for precise, in situ characterization of drug nanocarriers in blood is needed. Here we show how the fluorescence correlation spectroscopy that is a well-established method for measuring the size, loading efficiency and stability of drug nanocarriers in aqueous solutions can be used to directly characterize drug nanocarriers in flowing blood. As the blood is not transparent for visible light and densely crowded with cells, we label the nanocarriers or their cargo with near-infrared fluorescent dyes and fit the experimental autocorrelation functions with an analytical model accounting for the presence of blood cells. The developed methodology contributes towards quantitative understanding of the in vivo behavior of nanocarrier-based therapeutics.


Supplementary Notes Note 1: NIR-FCS measurements in static blood
In order to correlate the recorded FCS curves with the precise position of the observation volume V obs with respect to the blood cells, the cells were stained with the green-fluorescing membrane dye DiI (see Supplementary Method). During the staining procedure, plasma was removed and substituted by phosphate buffered saline (PBS) to which CB1 was added in a nanomolar concentration. In this blood cells suspension (BCS), images recorded with the LSM 880 confocal microscope rendered a first impression of the 3D configuration of the cells. However, in the visible spectrum the main components of red blood cells, oxy-and deoxyhemoglobin, show strong absorption leading to a dramatic loss in signal intensity already at the first micrometers. 1 Strong scattering additionally reduced resolution with the effect that a clear distinction of cellular structures at imaging depths >10 µm was not possible. Within the first 10 µm we observed a close network of packed, sedimented cells ( Supplementary Fig. 4). FCS measurements were performed at penetration depths between 0 and 10 µm (Supplementary Fig. 4a). When V obs was positioned in a cell free spot clear off the glass-sample interface, the corresponding autocorrelation curve could be fitted with a fit function for one type of freely diffusing species (eq.3 with m=1 in the main text) ( Supplementary Fig. 4b). The diffusion coefficient value 21.5 µm² s -1 obtained from the fit differed only marginally from that in pure water, D CB1, water = 20.4 µm² s -1 assuming that the dimensions of V obs were equal in both cases. However, when the measurement was performed in a spot partially occupied by a cell, the detected fluorescence intensity and the amplitude of the autocorrelation curve decreased significantly. Furthermore, the fits to the autocorrelation curve did not yield reliable values for D ( Supplementary Fig. 4c). We concluded that because free Brownian diffusion of the fluorescent tracers can occur only in the liquid fraction of the blood, the full or partial obstruction of the FCS observation volume by cells prevented reliable measurements.
FCS is sensitive to spherical aberrations caused by a mismatch of refractive indices. Here we used a water immersion microscope objective that is optimized for water with n water = 1.33. 2 Red blood cells on the other hand were reported to have a refractive index of 1.399. 3 Thus, passing the laser beam through a cell might lead to an increase in V obs resulting in longer diffusion times and correspondingly a lower diffusion coefficient. We therefore investigated if a measurement at locations above the first layer of cells yielded comparable results to those where no cells crossed the laser beam ( Supplementary Fig. 4d-i). By positioning V obs at such a spot a good autocorrelation curve was measured ( Supplementary Fig. 4h). The fit to this curve yielded a diffusion coefficient of 19.3 µm² s -1 that is only slightly lower than D CB1, water . Thus, we conclude that laser penetration through one layer of cells did not distort the observation volume.
Next, we sought to measure the diffusion in full blood by FCS ( Supplementary Fig. 5). Here, staining of the cell membrane was omitted to preserve the original blood composition. This, however, meant that we could not control if the spot was occupied by a cell or not. Subsequently, we searched for spots that showed sufficient fluorescence signal intensity. In some cases, a slow movement of cells on a second to minutes time scale could be observed by fluctuations of the average count rate.
Furthermore, we assessed that at penetration depths of >10 µm the likelihood of finding good spots decreased. We assumed that this was caused by the increasing distortion of V obs when multiple layers

Note 2: NIR-FCS studies in plasma
The influence of the liquid blood fraction was evaluated by studying CB1 and CB2 diffusion in undiluted human plasma. The diffusion time of CB1 in plasma increased by a factor of 1.43 compared to that measured in water ( Supplementary Fig. 6). This could be the consequence of several effects that require careful consideration.
(1) As plasma contains proteins, ions, and other solutes its refractive index may be slightly higher than that of water. This effect combined with an enhanced scattering in plasma might lead to a slight increase of the FCS observation volume in plasma compared to water especially at higher penetration depths. To address this issue we recorded FCS autocorrelation curves at different penetration depths in water and plasma ( Supplementary Fig. 6b). The results showed no change in the diffusion time of CB1 for penetration depths up to 50 µm in plasma. Thus, we conclude that up to this penetration depth the observation volume V obs remains unaffected and used its radial dimension r 0 calibrated in water to calculate the diffusion coefficient of CB1 in plasma, D CB1, plasma of 14.3 µm² s -1 .
(2) Many nanocarriers are reported to show an immediate adsorption of plasma proteins upon contact with plasma or blood. In 2007, Cedervall et al. coined the term "protein corona" to describe this behavior. 4 The eventual presence of such tightly bound layer of strongly adsorbed proteins would increase the hydrodynamic radius of CB1 and thus its diffusion time in plasma compared to water.
Using Stokes-Einstein relation and assuming the viscosity of water for the plasma, the diffusion coefficient of CB1 in plasma D CB1, plasma =14.3 µm²s -1 translates to a hydrodynamic radius of 16.2 nm.
That is an increase with ΔR H = 4.8 nm as compared to the value in water. In order to get further insight, we also studied the second, larger cylindrical polymer brush, CB2. The diffusion coefficient of CB2 in water was determined by dynamic light scattering (D CB2,water = 10.9 µm² s -1 , R H,CB2 = 21.3 nm). In plasma ( Supplementary Fig. 7), we found again a reduction of the diffusion coefficient compared to water by a factor of 1.43 (D CB2,plasma = 7.6 µm² s -1 ), that translates into a ΔR H of 9.4 nm again assuming water viscosity for plasma in the Stokes -Einstein (eq. 5 in the main text) relation. Considering the analogous synthetic identity of both cylindrical polymer brushes a protein adsorption layer would be expected to consist of the same proteins and therefore give a similar ΔR H . However, as the two differently sized cylindrical polymer brushes showed an identical scaling factor between diffusion in water and plasma and thus different ΔR H we deduced that the difference could not be explained by the formation of a tightly bound protein corona.
(3) Due to its high protein content, plasma has a higher macroscopic viscosity than water. Rolling ball viscometer measurements yielded η plasma = 1.48 mPa*s at 22°C and 1.35 mPa*s at 25°C. Clearly, these values for the macroscopic viscosity of the plasma have to be used with care when the diffusion of small species such as CB1 in a complex environment as the blood plasma is considered and their diffusion coefficient calculated through the Stokes-Einstein relation (eq.5). On one hand, fluorescence recovery after photobleaching studies 5 have shown that the concentration dependence of the longtime self-diffusion coefficient of hard-sphere colloids is very similar to the concentration dependence of the macroscopic viscosity at low volume fractions. On the other hand, FCS studies of tracer diffusion in various crowded enviroments [6][7][8] have shown that the local viscosity or friction that is experienced by the tracers depends on their size as well as on the characteristic length scale of the surrounding matrix and does not necessarily equal the macroscopic viscosity. Nevertheless, the fact that both CB1 and CB2 show the same diffusion slowdown of 1.43 in plasma compared to water strongly suggest that in plasma both cylindrical polymer brushes do not change their size but simply experience an effective viscosity that is 1.43-times higher than that of water. It is worth to note that this slowdown in the CB1 and CB2 tracer diffusion coefficient in plasma compared to water, closely matches the value of 1.4 that was reported 9 for the slowdown in the self-diffusion coefficient of human serum albumin and four other proteins in solutions with high ionic strength and a protein volume fraction of 6%.

Note 3: Combining normal and inverse FCS
A precise analytical derivation of eq. 2 in the main text that combines normal and inverse FCS is outside of the scope of this paper. As discussed below such combination is justified by the order of magnitude difference in the sizes and thus in the diffusion times of the fluorescent species and the blood cells. Indeed, the analytical expression for inverse-FCS (that is used as a part of eq. 2 in the main text) was initially derived by Wennmalm et al. 10  In our case, the unlabeled objects are the red blood cells that are much larger than the nanoparticles used by Wennmalm et al. 10 . Thus, the diffusion time of the blood cells through the observation volume is very slow ~ 500 ms. In the time intervals, in which the observation volume is not occupied by blood cells, it is occupied by the studied fluorescent species (small dye molecules or loaded nanocarriers). In these time intervals high fluorescence intensity is measured. Clearly, this intensity is not constant as in the initial work of Wennmalm et al. 10 , because due to the relatively low concentration of the studied fluorescent species their diffusion in and out of the observation volume creates intensity fluctuations.
However, these fluctuations, caused by small species (~10 nm) with diffusion times in the order of a few ms are orders of magnitude faster than the ones caused by the blood cells. This very large difference in the time scales justifies the use of eq. 2 (in the main text) with two components: one for the small fluorescent species and one for the very large blood cells.

Diffusion coefficient measurements by PFG NMR
The diffusion coefficient of IRDye®800CW-DBCO in aqueous solutions was measured using a 5 mm triple resonance TXI 1 H/ 13 C/ 15 N probe equipped with a z-gradient on the 850 MHz Bruker AVANCE III system. The temperature was kept at 25°C and regulated by a standard 1 H methanol NMR sample using the "topspin 3.1" software (Bruker). The control of the temperature was realized with a VTU For the diffusion measurements a 2D sequence (DOSY, stebpgp1s19) with a stimulated echo was used additionally with water suppression (3-9-19 pulse sequence with gradients). 11 The 2D NMR sequences for measuring diffusion coefficient uses echoes for convection compensation and longitudinal eddy current delays to store the magnetization in the z-axis, and only be dependent on T 1 -relaxation. The calculation of the diffusion value was automatically done with the mono exponential function. 12 Diffusion gradient amplitudes were varied linearly from 1 to 53 G/cm (10 to 470 mT/m) over a total of 32 experiments to achieve a strong diffusion weighting. Diffusion coefficients were then calculated for the integrated peak areas using an exponential decay fit function over the 16 spectra where the diffusion sensitivity factor b was calculated as given in the sequence description: Typical spectra measured for aqueous solution of the IRDye®800CW-DBCO are shown in the Supplementary Figure 3. The fit yielded a value of 265 ± 10 µm² s -1 for the diffusion coefficient of the dye at 25°C that corresponds to a value of 251 ± 10 µm² s -1 at 23°C.

Staining of red blood cell membrane
Human blood (excess from a buffy coat preparation in our institute) was diluted with phosphate