Methods were carried out in accordance with relevant guidelines and regulations and in compliance with ARRIVE guidelines.
Ethics Statement.
Animals were housed under controlled temperature (21 ± 10°C) and humidity (55 ± 10%), with a 12-h light/12-h dark cycle. Food and water were available ad libitum during all experiment. This study was approved by the local committee for animal care (Comité d'éthique en matière d'expérimentation animale Paris Centre No. 59) under the Agreement No. 24933.
Animal Preparation.
In order to map the deep vascular anatomy of the retina,ULM imaging was performed on N=1 male Sprague-Dawley rat weighing 400g. rfUS imaging was performed on two groups of rats of 18 months of age: one group of TgF344-AD rats (TG group, n = 6) and another group of wild type littermates (WT group, n=6). For both types of imaging, animals were anesthetized intraperitoneally with a mixture of Ketamine (Imalgene®, 40mg/kg) and Medetomidine (Domitor®, 0.3mg/kg). After anesthesia induction, the animal was placed in a stereotaxic frame to stabilize its head. The anesthesia level was adjusted, if necessary, throughout the experiment. The body temperature was controlled by a rectal probe and maintained at 37 ± 0.5 °C by a feedback-controlled heating pad. Neomycin‐polymyxin B (Tevemyxine® Collyre; TVM, Lempdes, France) and N-acetylcysteine (N.A.C.® Collyre) ophthalmic solutions were instilled in the imaged eye (2 drops each) to prevent corneal ulceration after the imaging session. For rfUS imaging, one drop of tropicamide (Mydriaticum® 0.5% Collyre) was instilled 30 minutes prior to the acquisition to induce mydriasis and cycloplegia. Tropicamide is an anticholinergic drug that produces short acting mydriasis and cycloplegia and allows better examination of the retina.Anticholinergic agents competitively block the binding of the neurotransmitter acetylcholine at muscarinic receptors located on the ciliary body of the eye, inducing paralysis of the ciliary muscle that controls pupil dilation.
All animals were awakened with a subcutaneous injection of Atipamezole (Antisedan®, 0.1mg/kg) at the end of the experiment.
Retinal Imaging.
Ultrasound Localization Microscopy.ULM imaging was performed non-invasively through the eye. The microbubbles (SonoVueTM, Braco, Italy) were reconstituted in the supplied 5mL of saline solution per vial (according to the manufacturer’s instructions) and administered via intravenous bolus injections of 200 µL in a lateral tail vein during the scan34(fig. 1.a,b). Blocks of 400 ultrasound images obtained at a 1000Hz framerate (each ultrasonic frame is a compound image acquired with transmissions at different compound angles -6°, -3°, 0°, +3°, +6° fired at a 5000Hz Pulse Repetition Frequency, 8V). The functional ultrasound scanner prototype (Iconeus One, Iconeus, Paris, France) is driving a 15-MHz ultrasound probe with 128 elements(fig. 1.c). The scan consists in 23 successive imaging planes acquired every 0.2mm over 4.4mm from an initial position (Xi) over the nasal side of the eye to a final position (Xf) over the temporal side of the eye (fig. 1.d,e). We accumulated images for 180s for each plane (900 compounded images), the entire 3D scan lasted 63 minutes. Density maps (D) were computed by counting all the positions detected in one pixel during the acquisitions. Velocity maps were computed as the mean velocity of every microbubble which passed through the pixel during the acquisition. For more details on the data processing, please refer to the data processing section described in Demene et al35.
Experimental paradigm. All measurements were performed in the right eye according to the following time schedule. After a short resting period to obtain stable hemodynamic conditions, power doppler measurements were performed. An acquisition consists of a baseline recording of 45 seconds followed by 50 consecutive trials and terminated by a recovery period of 75 seconds, during which the RBV is continuously recorded. Each trial comprises 15 seconds without stimulation, 0.8 seconds with white light and another 15 seconds without stimulation. One complete acquisition lasted 1660s.
Light Stimulation. The animal was maintained in a dim light room for 45 minutes before starting the imaging session. For light stimulation, a custom-built device was used, consisting of a white LED delivering light flashes through a microcontroller (Arduino Uno, Ivrea, Italy). The microcontroller was programmed with the Arduino Software (IDE). In order to synchronize the stimulation pattern with the scanner, the LED is triggered by the scanner through the microcontroller to flash for 0.8s every 30s to complete 50 trials (fig. 1.f).
Functional ultrasound sequences. Functional ultrasound imaging acquisitions were performed non-invasively through the eye with the same probe used for ULM imaging. The ultrasonic probe was placed upon the eye with acoustic coupling gel in order to image the retinal plane of interest where the long posterior ciliary arteries (LPCAs) were visible as they represent a reproducible and central landmark of the vasculature of the posterior part of the eye. The transducer was connected to a prototype functional ultrasound scanner (Iconeus One, Iconeus, Paris, France). Data was acquired by emitting continuously a group of 11 plane waves tilted from -10° to +10°, fired at 5.5kHz pulse repetition frequency. The back-scattered echoes of each group were summed to get compound images at 500Hz framerate, that allows correct sampling of the blood signal21. Doppler images were computed from blocks of 200 compound images averaged after filtering with the SVD spatio-temporal clutter filter36 and removing the 60 first singular values to discard noise and tissue motion. Each pixel of the final Power Doppler images is 100x100μm2 in plane. The slice thickness is approximately 300μm.
Activation map. Datapreprocessing consisted in a linear detrending, neither spatial nor temporal smoothing was used. For each animal, activation maps were computed with a single-subject global linear model (GLM) approach as routinely used method for the processing of fMRI data37. The design matrix used to detect retinal activation specific to the light stimulation only included the stimulus pattern as a regressor. The real stimulus duration was 0.8s but a theoretical 5s stimulus was used in the design matrix to account for the hemodynamic response delay. Statistical parametric maps were generated for each session, including a z-score map and a p-value map, a stimulation map and a baseline map. Individual delta RBV maps were obtained by dividing the stimulation matrix by the baseline matrix.
In order to determine which pixels achieved statistical significance considering the comparison of all the pixels of the z-map, we considered p-value < 0.05 statistically significant and performed Bonferroni correction for multiple comparisons. The pixels meeting the defined criteria were considered significantly activated. For each animal, an activation map displaying the activated pixels’ z-score was produced and overlaid onto the corresponding grayscale mean Doppler image of the animal eye in order to accurately locate the neural activity. The mean rCBV value was estimated from a region of interest (1.4 mm diameter) centered on the peak z-score location of the activation map. Data processing and analysis were performed using Matlab (MATLAB Release 2018a, The MathWorks, Natick, Massachusetts, United States).
Hemodynamic response measurement and parameters estimation.
The retinal hemodynamic response was extracted and averaged from the same region of interest. The times-series were averaged between trials for each animal. For a visual comparison, the hemodynamic response was averaged within groups for comparison between the transgenic retinal hemodynamic response and the control response.
In order to extract quantitative parameters, we fitted the hemodynamic response using a simple model. While several models of hemodynamic responses have been proposed in the fMRI literature such as the two-gammas function or inverse logit function38, we choose to use a simpler parametrized model which allows direct interpretation of key physiological parameters39 such as the initial response delay, the response rise time, the response amplitude, the response decay time, the undershoot amplitude and return to baseline time. For this, we have opted to use the 4 cosines model which is the concatenation and scaling of four half-period cosines and is determined by six independent parameters (see supplementary figure 1). Non-linear curve fitting of the RBV response by the 4 cosines model was then performed using the interior point method with constraints (Matlab 2018a, The MathWorks, Inc., Natick, Massachusetts, United States.).
In order to evaluate the number of trials necessary to obtain robust hemodynamic parameters, we performed a bootstrap analysis by randomly sampling with replacement a subset of trials in the experiment and rerunning the fit and parameters estimations. Curves representing the different parameters robustness were constructed by evaluating their standard deviation and mean.
Statistical analysis. To compare the different hemodynamic response parameters of the hemodynamic retinal response of the transgenic (TG) and wild type (WT) groups, a two-tailed Wilcoxon test was performed (because of TG group non-normal distribution) using GraphPad Prism (GraphPad Software, La Jolla California USA). To assess significance, we considered p<0.05 statistically significant.