Experiments were permitted by the local ethics committee on animal research (Landesamt für Gesundheit und Soziales, Berlin, Germany; LAGeSo G0186/16) and in conformity with the German Law for Animal Protection and the National Institute of Health Guidelines for Care and Use of Laboratory Animals. Experiments were reported following the ARRIVE guidelines.

Sixteen male C57BL/6J mice aged 12 weeks and an initial body weight ranging from 27 to 32 g (Charles River WIGA GmbH, Sulzfeld, Germany) were randomly assigned to permanent unilateral internal carotid artery occlusion (ICAO) for stimulation of spontaneous cerebral arteriogenesis (n = 8) or sham surgery (Sham) (n = 8), as previously described (Hecht et al., 2012). Briefly, mice were anesthetized with 70 mg/kg ketamine and 16 mg/kg xylazine and body temperature was maintained at 37°C. Animals were positioned supine and the right-sided ICA was exposed by a midline neck incision. For ICAO, the right ICA was permanently ligated with 8/0 silk. For Sham, the ICA was exposed but no ligation was performed. The skin was sutured with 6/0 nylon. Mice were kept in an enriched environment with free access to food and water.

On Day 21 after surgery, a venous catheter was placed into the femoral vein and CE-MRI was acquired at 7 Tesla with a small animal MR scanner (Bruker BioSpin GmbH, Ettlingen, Germany) using a transmit/receive mouse head cryoprobe (Bruker BioSpin GmbH, Ettlingen, Germany). Mice weight at Day 21 was 34.1 ± 9.7 g for ICAO and 30.2 ± 7.2 g for Sham. Researchers performing the CE-MRI were blinded to the group assignment. The MRI protocol included the following sequences:

T2-weighted sequences and DSC-MRI FOV covered the whole brain, whereas the CE-MRA FOV only covered the dorsal neocortex. A set of 5 saturation slices with identical geometry was used to suppress aliasing artifacts originating from the signal of the ventral part of the brain.

To visualize cerebral blood vessels for histological analysis, mice received an in vivo injection of 100 μl fluorescein-labeled Lectin [DyLight 488 Labeled Lycopersicon Esculentum (Tomato) Lectin, DL-1174, Vector Laboratories, Burlingame, CA] though the femoral vein to stain endothelial cells of perfused blood vessels. Next, the animals were transcardially perfused with 4% paraformaldehyde in phosphate buffered saline and the isolated brains were stored in the same solution at 4°C for 72 h and then kept in 0.1% sodium azide in phosphate buffered saline.

For the purpose of 3R and resource conservation during histological analysis, animals were randomly assigned to two groups: In the first group, n = 4 ICAO and n = 4 Sham animals were used to validate our animal model and confirm neocortical vascular remodeling on Day 21 after unilateral ICAO by confocal laser scanning microscopy (Leica TCS SP8, Leica Camera AG, Wetzlar, Germany) of whole-mount brain sections of 80 μm thickness with 1.5 × 1.5 × 8.9 μm³ [X × Y (plane) × Z (depth)] voxel resolution and laser excitation at 488 nm. The resulting confocal cerebral blood vessel volume (CBV) was defined as confocal-CBV by the fraction of Lectin-positive voxels from all tissue voxels in the right-sided neocortex (Supplementary Figures 1, 2).

After neocortical vessel outgrowth following ICAO was confirmed, n = 2 ICAO and n = 1 Sham animals were used in the second group to validate our iron oxide nanoparticle-based ss-CBV method at whole brain level by using serial two-photon tomography (STP) of perfused brains that was generated with a commercially available platform (TissueVision, Newton, MA, United States) (Ragan et al., 2012). Briefly, block-face 2-photon imaging of brains was performed by slicing 50 μm sections with measurements of two 25 μm optical planes per slice and three channels (RGB) leading to a final resolution of 1.2 μm × 1.2 μm in plane and 25 μm slice thickness. The resulting serial two-photon cerebral blood volume was defined as STP-CBV.

Image co-registration was used to compare MRI and histological data and to annotate brain regions from the Allen brain atlas. MR image pre-processing was performed in ImageJ (v1.52e)¹ according to the following steps: First, all images were transformed to the resolution of the CE-MRI scan using nearest-neighbor interpolation. Second, fixed relations in image geometry planning during acquisition were used to transform images to a final FOV = 18.72 mm³ × 18.72 mm³ × 15.2 mm³, (65 μm)³ isotropic resolution and matching geometry without image registration techniques. Manual correction of shifts due to scanner imperfections were corrected in steps of one voxel using a custom ImageJ plugin and 3 degrees of freedom and nearest neighbor interpolation. T2-weighted images were warped to the Allen brain atlas using the MATLAB (v. R2014a, MathWorks, Natick, MA, United States) toolbox ANTx (Koch et al., 2019). The calculated registration was then applied to all other MRI measurements. Histology images were registered to the Allen brain atlas using custom MATLAB tools (R2016a, The MathWorks, Inc., Natick, MA, United States) (Koch et al., 2019). The following steps were performed to transform STP data to the Allen brain atlas:

Angiograms from CE-MRI (=CE-MRA) were generated by subtraction of image intensities S_pre – S_post. From these, vessel densities were calculated by a vessel tracing algorithm as described previously (Klohs et al., 2012). Based on previous findings in our model and as suggested by the results from our confocal-CBV analysis (Supplementary Figures 1, 2), we hypothesized an increase in vessel densities in the right-sided neocortex of ICAO animals compared to Sham (Hecht et al., 2012, 2015; Marushima et al., 2019).

The change in transverse relaxation rate ΔR₂* = 1/TE × ln (S_pre/S_post) was calculated using ImageJ and ss-CBV was defined using the biophysical model of Yablonsky and Haacke via ss-CBV = 3/(4πγΔχB₀)ΔR₂* with the gyromagnetic ratio γ = 2.67502 × 10⁸ rad/s/T and susceptibility difference Δχ = 0.282 × 10^–8 (CGS units) between blood and tissue at B₀ = 7 T (Lemasson et al., 2016) assuming linear increase with contrast agent dose and saturation of iron magnetization above 4.7 T (Yablonskiy and Haacke, 1994; Lemasson et al., 2015).

Dynamic susceptibility contrast relative cerebral blood volume was calculated using DSCoMAN Version 1.0 (Duke University Medical Center, Durham, NC, United States) (Boxerman et al., 2006). Using ImageJ, raw DSC-MRI images were manually converted into a stack with xyztc order, one channel, 15 slices, and 200 frames. Pixel width and height were set to 0.2925 mm, voxel depth to 0.4 mm. The frame interval was set to 0.4 s, the origin to 0.0 pixels. A timetable was imported. Next, the DSCoMAN GUI was started within ImageJ to calculate raw perfusion images. The first 20 timepoints were excluded for the analysis of rCBV. Last baseline time, last time, slice locations, time points, echo time (msec), whole brain threshold, noise SDs, as well as parameter for last baseline time computation were not modified from the values automatically given by DSCoMAN. The resulting rCBV map was scaled as the raw DSC-MRI images above and saved as a nifti file.

Since no data on CE-MRI was available for our mouse model, we chose an exploratory study design with similar sample size to the original study describing the mouse model with histological methods (Hecht et al., 2012). A total of 16 mice underwent MRI measurements. The researchers performing the MRI measurements were blinded to the group assignment. One animal was excluded because of clots in the brain after iron oxide nanoparticle injection, one animal died during the MRI protocol leading to the final group size of 7 per group. All analysis was performed blinded to group assignment. A volume-of-interest of the right neocortex for vessel tracing was defined manually by a blinded researcher and no iterations were made in the analysis. The right-sided neocortex defined by the Allen Mouse Brain Atlas was chosen as volume-of-interest for ss-CBV and DSC-rCBV analysis prior to the analysis (Supplementary Figure 3; Lein et al., 2007). Analysis of volume-of-interests in processed images was performed using custom code. Plots and statistics were performed using R version 4.0. Individual data points were visualized together with a summarized mean value ± standard deviation. Data were statistically described by the p-value derived from 2-sided Student’s t-tests after passing a Shapiro-Wilk normality test. Spearman’s rank correlation coefficient (r) was used to describe the correlation between ss-CBV and h-CBV.

Blood vessels in the right neocortex 21 days after ICAO or sham surgery were precisely visualized in a three-dimensional space after subtraction of post-contrast from pre-contrast images. As expected from previous findings¹¹ and our confocal modal validation (Supplementary Figures 1, 2), more neocortical vessels were visible on CE-MRA subtraction angiograms 3 weeks after ICAO throughout both hemispheres (Figure 1A). Unexpectedly, however, vessel densities calculated by a previously described, semi-automated vessel tracing algorithm tended to be lower in ICAO than in Sham-treated animals (Sham: 11.8 ± 2.3 mm^–3; ICAO: 9.5 ± 3.5 mm^–3; p = 0.17; n = 7 per group; Figure 1B). Thus, although measurements within the ICAO group displayed large MR vessel volumes that visually corresponded well to the subtraction angiograms, the tracing analysis did not translate these effects into an increased MR vessel density. Additionally, we compared MR vessel densities for both groups and hemispheres with corresponding MR vessel volumes in a post hoc analysis and, interestingly, MR vessel densities were in strong negative correlation to MR vessel volumes (r = -0.76; p = 0.000006; n = 28; Figure 1C).

Next, we explored whether ss-CBV maps could serve as a surrogate marker for parenchymal blood vessel imaging (Figure 2A) and found a significant increase of ss-CBV in the right neocortex following ICAO (Sham: 2.9 ± 0.3%; ICAO: 3.4 ± 0.4%; p = 0.04; n = 7 per group; Figure 2B). Cerebral hemodynamics were also assessed by DSC-MRI, which serves as a commonly applied technique for blood volume mapping by MRI. For between-subject comparisons of relative DSC-CBV (DSC-rCBV) in the parenchyma, the right hemisphere was normalized to the mean value of all measured voxels (Figure 3A). In agreement with our parenchymal ss-CBV analysis but at clearly lower spatial resolution (Figure 3B), a significant increase in DSC-rCBV following ICAO was noted for the right neocortex (Sham: 1.13 ± 0.02 AU; ICAO: 1.24 ± 0.02 AU; p = 0.000002; n = 7 per group; Figure 3C). To confirm the internal validity of our ss-CBV blood volume analysis, neocortical ss-CBV and DSC-rCBV measurements from both hemispheres were normalized and compared with each other using a Bland-Altman plot that yielded a mean difference of 0.000007 AU with limits of agreement of –0.07 to 0.07 AU; n = 14; Figure 3D).

Next, ss-CBV findings were histologically validated at the whole brain level by correlating ss-CBV to corresponding STP-CBV from STP angiographies. For this purpose, 406 regions of the Allen brain atlas were covered with MRI and STP modalities to permit a direct comparison of ss-CBV against STP-CBV across those regions. Importantly, blood vessels according to ss-CBV were accurately traceable by STP (Figure 4) and a correlation analysis across 406 brain regions from the Allen brain atlas demonstrated good correlation between ss-CBV and STP-CBV in the mouse neocortex (Mean r = 0.57; p < 0.000001; Sham: n = 1; ICAO: n = 2; Figure 4).

Clinically, iron oxide nanoparticle-based CE-MRA and ss-CBV maps could serve as a non-invasive alternative to digital subtraction angiography for longitudinal assessment of vascular malformations, brain tumor vascularity, anti-angiogenic treatment effects but also for improved planning of neurosurgical or radiosurgical procedures (Varallyay et al., 2013; Iv et al., 2018; Huang et al., 2019). However, clinical application has been limited to proof-of-concept studies and off-label use due to missing regulatory approval, which is mainly attributed to the concern that the carbohydrate coating of the nanoparticles could be associated with an increased risk of severe allergic response. On the other hand, the safety profile of iron oxide nanoparticles has been reported similar to iodinated contrast agents used for contrast-enhanced CT imaging (Vasanawala et al., 2016). Additionally, there is increasing awareness of permanent gadolinium deposition in the brain and its unknown risk (Gulani et al., 2017) and iron-based contrast agents may be considered as a suitable alternative (Boehm-Sturm et al., 2018).

A main limitation in the present study is the sample size. However, we deliberately decided to only use animals from one sex and a single model of disease because we believe that this optimizes the reliability of our findings by reducing the complexity of the model. Second, the lack of coherence to the previously published vessel tracing algorithm for CE-MRA underlines that such approaches may require individual validation depending on the used model or the diversity of the cohort, including the use of subjects of both sexes and different mouse strains. Third, validation of ss-CBV measurements were naturally limited by the fact that our STP vessel stain only reflected blood vessel walls rather than appreciate the entire volume of the vessel lumen as it is done by ss-CBV. On the other hand, the fact that ss-CBV recognizes the entire vessel lumen also explains why this method seems particularly useful for microvascular vessel volume quantification, because histological CBV may underestimate vessel density in areas of large surface vessels (Di Giovanna et al., 2018). Forth, although ss-CBV offers the possibility to determine vessel density based on perfused blood volume with the benefit of being extremely robust, ss-CBV is unable to distinguish between vessel diameters. For this purpose, vessel size imaging and quantitative blood oxygen level dependent imaging make use of more advanced biophysical models of spin echo MR signal in tissue (Kiselev and Novikov, 2018) but are currently limited by high noise level and low spatial resolution (Boehm-Sturm et al., 2013). MR vascular fingerprinting is a promising first attempt to reduce noise by simulating the MR signal in realistic tissue and finding a best match to the measured signal (Lemasson et al., 2016), which may soon enable vessel size sensitive imaging at similarly high resolution presented here while reducing impact of noise. In the future, automated machine-learning analyses of comprehensive histological and MR angiograms may help overcome the limitations of previous and the present work and yield further insights into neurovascular imaging correlates with the possibility of reliable clinical translation (Di Giovanna et al., 2018; Todorov et al., 2019).