Digital Simulation Assessments of the Sensitivity of Quantitative MRI for Detection of an Iron Oxide Nanoparticle Brain MRI Contrast Agent

There have been substantial efforts to develop targeted exogenous MRI contrast agents to assess specic brain pathologies. In parallel with other efforts, it is important to assess the sensitivity of candidate MRI methods for detection of contrast agents. Here, we propose a digital simulation approach, which includes MR relaxation (R1 and R2) mapping and image co-registration. We simulated the effects of 3 nm iron oxide nanoparticles (IONPs) as a model contrast agent. Two independent relaxation maps acquired from the brain of the same subject were co-registered. The baseline subtraction between the two relaxation maps showed good agreement, demonstrating the high reproducibility of the method. Next, the second relaxation map was digitally altered (“seeded”) to simulate additional MR relaxation values corresponding to several concentrations of 3 nm IONPs in various locations. The maps of absolute differences between the rst relaxation map and the digitally altered second relaxation maps were assessed for conspicuity. Results based on living mouse and human brains scanned at 9.4 T and 3.0 T respectively both indicated reliable conspicuity for signal equivalent to 0.06 mM IONP or higher. Overall, the digital simulation approach is a useful method to improve the development of MRI contrast agents and accompanying MRI methodologies.


Introduction
Magnetic resonance imaging (MRI) has been widely used as a diagnostic tool in both preclinical and clinical studies. Its high spatial resolution allows MRI to examine the central nervous system (CNS) with great anatomical detail [1][2][3][4][5] . The intrinsic biological characteristics of brain tissue can be characterized using the MR properties of longitudinal relaxation (T1) and transverse relaxation (T2) times; these are commonly referred to as rates R1 (1/T1) and R2 (1/T2). Typically, R1 and R2 MRI provide differentiation of brain tissues 1,5−9 and detection of abnormalities [10][11][12][13][14] . However, both R1 and R2 MRI still suffer from the lack of pathophysiological speci city. Thus, multiple research groups are approaching the problem of pathophysiological speci city by developing targeted exogenous contrast agents, such as iron oxide nanoparticles (IONPs) [15][16][17] .
It is well known that IONPs have both R1 and R2 shortening effects 18 , providing positive and negative MR signal enhancement, respectively. IONPs not only produce sensitive MR contrast but also provide a scaffold for conjugation of antibodies or other molecular targeting domains which may be used to provide pathophysiological speci city in brain 19,20 . In general, quantitative R1 and R2 mapping have advantages over T1 and T2 weighted images by providing quanti able MR parameters with reduced variability from MR hardware differences, positioning in the scanner, and other factors. Given that T1 imaging is quite sensitive to the effects of applied contrast agents, there have been efforts to develop MR methodologies to quantitatively detect the R1 relaxation enhancing effects induced by IONPs [21][22][23][24] .
Among these, the Magnetization Prepared -RApid Gradient Echo (MP-RAGE) with 2 inversion times (MP2RAGE) was chosen for R1 mapping in this study for several reasons: MP2RAGE is readily available on both human MRI scanners and preclinical scanners, the method is robust to RF eld inhomogeneity, and it has no requirement for extensive post acquisition image processing 21 .
Ultimately, the effect of IONPs on longitudinal and transverse MR relaxation should be tested in living tissues such as the brain. To do this, the employed contrast agent must be delivered to brain tissue by passing through the blood-brain barrier (BBB). There have been many efforts to develop methodologies to safely carry MR contrast agents across the BBB [25][26][27][28][29] , but at present this remains a challenge. In the meantime, as BBB crossing methods are being developed, it is also important to assess the sensitivity of candidate MRI methods for detection of IONPs. In this way, we will gain a better understanding of the amount of IONPs that will have to be delivered across the BBB to yield a detectible signal in vivo.
Here, we propose a digital simulation method, which includes both R1 and R2 mapping with image coregistration, to evaluate the MR sensitivity of extremely small IONPs for both human and mouse brain. In order to mimic the hypothetical effects of several types of human brain pathology, the digital simulations were focused on brain regions which are well known to be vulnerable to neurological disease or injury.
The cortex of rodent brain shows various pathologies in both transgenic and brain trauma animal models [30][31][32][33] . The hippocampus is well known to be vulnerable in multiple neurological diseases including Alzheimer disease, mesial temporal sclerosis, herpes simplex virus encephalitis, and many more [34][35][36][37] . The depths of cortical sulci are the pathognomonic anatomical locations for the pathology that de nes Chronic Traumatic Encephalopathy (CTE) 38 , likely due to damage from shear deformation induced by rapid rotational acceleration and impact 39 40 . Thus, we focused on cortex in mice, and on hippocampus and cortical sulcal depths in humans.

Digital simulation work ow
The main aim of the study was to establish a digital simulation method to test the sensitivity of proposed R1 enhancing MR contrast agents on the in vivo brain MRI. The digital simulation method is outlined in Fig. 1. The digital simulation required two independent R1 maps ( Fig. 1a and b) of the same subject within a short time interval to reduce the likelihood that there will be any physiology-induced changes.
The second scan R1 map (Fig. 1b) was co-registered to the rst scan R1 map (Fig. 1a) using ANTs (http://stnava.github.io/ANTs/). Then, the absolute ΔR1 (|ΔR1|) map between the 1st R1 map (Fig. 1a) and the co-registered 2nd R1 map (Fig. 1c) was calculated to produce a baseline |ΔR1| map (Fig. 1d). The baseline |ΔR1| map had mostly zero values, 0.009 ± 0.002 (n = 5 mean ± SD), except brain tissue and ventricle border. This demonstrated that the two independent R1 maps were obtained without imaging artifact and that the co-registration was effective at a voxel-by-voxel level. The co-registered 2nd R1 map was then digitally altered ("seeded") to simulate additional R1 signals equivalent to those expected from an R1 enhancing contrast agent (Fig. 1e). The digitally added R1 values in Fig. 1e were 0.0265 (i), 0.053 (ii), 0.0795 (iii), 0.106 (iv), and 0.1325 (v) s -1 , which are equivalent to 5, 10, 15, 20 and 25 % of the in vivo mouse brain cortex R1 value of 0.53 s -1 . Finally, the absolute ΔR1 (|ΔR1|) map between the 1st R1 map ( Fig. 1a) and the digitally seeded 2nd R1 map co-registered to the 1st R1 map (Fig. 1e) was calculated to produce a digitally seeded |ΔR1| map (Fig. 1f). The digitally added R1 enhancements were clearly visible in the |ΔR1| map, though the conspicuity was modest for the less enhanced regions (i and ii). R1 of in vivo mouse brain and 3 nm IONP at both 4.7 T and 9.4 T.
In vivo mouse brain R1 maps were obtained at both 4.7 T and 9.4 T, with ve mice each per scanner ( Supplementary Fig. 1a -j) The ve mice scanned at 9.4 T were the same subjects from our previous report 41 . Region of interest (ROI) analysis was performed on dorsal cortex, and the quanti ed R1 values were 0.62 ± 0.01 s -1 (4.7 T, n = 5, mean ± standard deviation) and 0.53 ± 0.01 s -1 (9.4 T, n = 5, mean ± standard deviation). The relaxivity (r1) of 3 nm IONP were assessed at both magnetic elds resulting in 1.87 mM -1 s -1 at 4.7 T and 0.79 mM -1 s -1 at 9.4 T ( Supplementary Fig. 1k.). Thus, from an intrinsic relaxivity perspective, 4.7 T imaging has substantial advantages over 9.4 T imaging. These longitudinal relaxivities were used to simulate additional R1 signals for both 4.7 T and 9.4 T.
Reproducibility of in vivo mouse brain R1 map at both 4.7 T and 9.4 T.
As described in the digital simulation work ow, two independent in vivo mouse brain R1 maps were obtained from 5 mice within a one-week time interval. The 2nd R1 maps were co-registered to the 1st R1 maps and the absolute R1 difference maps (|ΔR1|) were calculated (Supplementary Fig. 2a-j.) These baseline |ΔR1| maps provided quantitative assessments of the reproducibility of the combined MP2RAGE image formation and co-registration process with ANTs. The reproducibility was clearly higher at 9.4 T than at 4.7 T. The signal to noise ratio (SNR) of the 2nd inversion recovery T1 weighted image of MP2RAGE was substantially higher at 9.4 T compared with 4.7 T as well ( Supplementary Fig. 2aa-jj). In the cortex, the |ΔR1| at 4.7 T was highly variable (0.03 ± 0.02) compared to 9.4 T (0.01 ± 0.01) ( Supplementary Fig. 2k vs. l). Thus, from an imaging reproducibility perspective, 9.4 T imaging has substantial advantages over 4.7 T imaging.
Assessment of sensitivity of MP2RAGE derived R1 maps to simulated 3nm IONP at 4.7 T vs 9.4 T.
The sensitivity of MP2RAGE derived in vivo mouse brain R1 maps to simulated 3 nm IONP was compared between 4.7 T and 9.4 T (Figs. 2-3). The digitally added additional R1 was calculated from the relaxivities of 3 nm IONP at both eld strengths equivalent to the expected R1 enhancement from 0.02, 0.04, 0.06, 0.08, and 0.1 mM [Fe], resulting in 0.034, 0.069, 0.103, 0.138, and 0.172 s -1 at 4.7 T and 0.014, 0.029, 0.043, 0.058, and 0.072 s -1 at 9.4 T. The hypothetical additional R1 was added to random patches in mouse cortex with small 1-8 voxel patches added to right cortex and larger 20-30 voxel patches added to left cortex, mimicking the various pathology sizes. For both low and high eld, the R1 enhancements are much clearer for the larger patches (mouse brain -left side, image left side) than smaller patches (mouse brain-right side, image left side). R1 enhancements equivalent to iron concentration of 0.08 to 0.1 mM were visible but not conspicuous in R1 maps from both 4.7 T and 9.4 T imaging ( Fig. 2a-b). The same R1 enhancements were much more conspicuous in |ΔR1| maps, with conspicuity for R1 enhancements equivalent to iron concentration of 0.06 mM in larger patches ( Fig. 2cd). The digitally added additional R1 values for 4.7 T, r1 = 1.87 mM -1 s -1 , are about 2.4 times of that of 9.4 T, r1 = 0.79 mM -1 s -1 . The intrinsic R1 of mouse brain cortex changed from 0.53 s -1 at 9.4 T to 0.62 s -1 at 4.7 T. Consequently, the R1 maps at 4.7 T showed better conspicuity than those of 9.4 T. However, in the | ΔR1| maps, 9.4 T showed similar conspicuity as 4.7 T to the digital simulated R1 of 3 nm IONP. The higher intrinsic IONP signal at 4.7 T and the improved reproducibility at 9.4 T had approximately equivalent effects, making conspicuity at 4.7 T and 9.4 T in |ΔR1| maps very similar overall. These effects were found across multiple mice at both 4.7 T and 9.4 T (Fig. 3). As expected, outlier mice having low reproducibility (mouse 5 at 4.7 T and mouse 10 at 9.4 T) lost conspicuity to R1 enhancements equivalent to lower iron concentrations (Fig. 3.) Thus, MP2RAGE at both 4.7 T and 9.4 T were found to be equivalently appropriate approaches to assessing 3 nm IONP contrast agents, as long as reproducibility was good.
It is readily apparent that higher relaxivity contrast agents will be more easily detectible. In our previous report 41 we predicted that the lower limit of sensitivity of MP2RAGE derived R1 maps to 3 nm IONP would be 0.03 mM. In the previous report, Kim et al., 2021, 41 , the 3 nm IONPs were intracranially injected into the mouse brain at 2 relatively high concentrations, 0.1 mM and 0.25 mM ( Supplementary Fig. 3, reproduced from Fig. 8 in reference 41) and we extrapolated down to the 95% con dence bound for the reproducibility of the |ΔR1| map at 9.4 T. The 3 nm IONP used in our previous report had a higher relaxivity of 1.25 mM -1 s -1 at 9.4 T, re ecting the moderate batch-to-batch variability in producing these IONP contrast agents. We repeated the digital simulations using relaxivities equivalent to those from the 3 nm IONP with r1 = 1.25 mM -1 s -1 . As predicted, |ΔR1| maps had good conspicuity for larger patches with enhancement equivalent to that expected from as little as 0.03 mM [Fe] of 3 nm IONP with r1 = 1.25 mM -1 s -1 . (Fig. 4). This result underscores the importance of optimizing the r1 of the contrast agents for detection of subtle pathologies or circumstances in which it is challenging to get contrast agents across the blood brain barrier.
Assessment of sensitivity of R2 map on 3 nm IONP at 9.4 T.
While the 3 nm IONPs were selected as contrast agents largely due to their R1 properties, they also may be used as R2 contrast agents. The sensitivity of 3 nm IONP on R2 maps was digitally simulated at 9.4 T (Fig. 5). Two independent in vivo mouse brain R2 maps were obtained from 3D T2 weighted images with repetition time (TR) = 1.0 s. The relatively short TR was chosen to obtain 3D R2 maps of whole mouse brains within a reasonable scan time (10 min total scan time for each R2 mapping); the R2 values of mouse brain were largely preserved compared to measurements made with long TR (8 s) ( Supplementary   Fig. 4). The transverse relaxivity (r2) of the 3 nm IONP at 9.4 T was 20.1 mM -1 s -1 . Using this r2, digital simulations of |ΔR2| were performed analogously to those performed for |ΔR1|. To examine the detection limit of R2 maps for 3 nm IONP contrast agents, digital simulation was also performed based on hypothetical 3 nm IONP having low r2, 15.7 mM -1 s -1 . The digitally added additional R2 values were calculated from the r2s of 3 nm IONPs at 9.4 T equivalent to the expected R2 enhancement from 0.02, 0.04, 0.06, 0.08, and 0.1 mM [Fe]. Image co-registration was done prior to digital seeding of additional R2 as in the R1 digital simulations. The R2 co-registration was accurate e cient in most brain region except brain tissues around ventricle. The visibility of R2 enhancements were better from high r2, 20.1 mM -1 s -1 , than low r2, 15.7 mM -1 s -1 , (Fig. 5a-b). The same R2 enhancements were much more conspicuous in |ΔR2| maps ( Fig. 5c) for simulated high r2 contrast agents, whereas the R2 enhancements from 3 nm IONP with r2 = 15.7 mM -1 s -1 were barely visible (Fig. 5d). This showed that the intrinsically high R2 of in vivo mouse brain at 9.4 T, which was about 25 s -1 compared to R1 of 0.53 s -1 in this study, requires contrast agents having high r2 to visualize R2 enhancement induced by exogenous transverse MR contrast agents. In Fig.   5c, the conspicuity for R2 enhancements from high transverse relaxivity, r2 = 20.1 mM -1 s -1 , were slightly less than those from R1 at 9.4 T. Thus, R2 mapping at 9.4 T provided similar conspicuity to R1 mapping if and only if the contrast agents have relatively high r2. Overall, this raises the intriguing possibility that 3 nm IONPs could be used as both R1 and R2 contrast agents at high magnet eld strengths.
Assessment of sensitivity of quantitative human brain MRI to simulated 3 nm IONP at 3.0 T. Digital simulations were performed using data from in vivo human brain R1 maps at 3.0 T. Two independent R1 maps were obtained from a healthy adult volunteer one-week apart. Image co-registration was done as for the in vivo mouse study. The longitudinal relaxation rate constant, R1, of 3 nm IONP was assessed from 0.02 to 0.  (Fig. 6) and depths of cortical sulci (Fig. 7). The R1 enhancement was not clearly visible from the R1 map with digitally added R1 values equivalent to 0.06 mM of [Fe] or less for hippocampus ( Fig. 6d-i -g-i and Fig. 6j-i -m-i). This low conspicuity was also observed in depths of cortical sulci even with signal equivalent to 0.1 mM [Fe] (Fig. 7e-i). However, the absolute R1 difference (|ΔR1|) maps had high conspicuity of R1 enhancement even with much lower simulated iron concentrations, down to 0.04 mM for hippocampus (Fig. 6f-ii and 6l-ii), and down to 0.06 mM cortical sulcal depth (Fig. 7c-ii). The | ΔR1| maps following image co-registration also effectively suppressed the tissue regions having intrinsically high R1, providing clear conspicuity of digitally added R1 enhancement ( Supplementary   Fig. 6.) As for the in vivo human brain R1 maps, digital simulations were also performed on in vivo human brain R2 maps. Two independent R2 maps were obtained within 2 hrs from the second healthy volunteer ( Supplementary Fig. 7), who is different from the volunteer for the R1 map. The subject was taken out from the MR scanner after acquiring the 1st R2 map and re-positioned into MR scanner for the 2nd R2 map. The hypothetical R2 enhancement was calculated using transverse relaxivity of 3 nm IONP at 3.0 T, 12.9 mM -1 s -1 ( Supplementary Fig. 5 simulations were performed on both hippocampus and depths of cortical sulci (Fig. 8). Unlike the digital simulation results from R1 map, the R2 enhancement was not apparent for the absolute R2 difference (| ΔR2|) maps for the iron concentration below 0.1 mM. Higher simulated iron concentrations, between 0.1 to 0.3 mM [Fe], were needed to visualize the R2 enhancement. Overall, this shows that 3 nm IONPs would provide more sensitive detection of pathology when used as R1 contrast agents at 3.0 T.

Discussion
In this study, we described an approach to assessing the enhancing effect of proposed longitudinal (R1) and transverse (R2) MR contrast agents in both human and mouse brain. We used digital simulation in this study, which avoids physical damage and can be done even without the availability of effective BBB crossing methods. To simulate the hypothetical in vivo condition, the proposed digital simulation method involved two independent R1 or R2 maps from the same subject within a short time interval, one week or less in this study, to minimize physiology-induced changes. After co-registration of the two independent R1 or R2 maps and digital seeding of simulated changes on the second map, the absolute difference map between them clearly showed the hypothetical enhancement that would be expected from the 3 nm IONP contrast agent at various concentrations. Overall, the digital simulations provided quantitative analyses for prediction of the sensitivity of candidate MRI contrast agents in brain tissues.
Importantly, the use of difference maps substantially improved the conspicuity of the simulated effects of contrast agent for both in vivo mouse and human brain. It is also likely that routine T1 or T2 weighted imaging, for which baseline subtraction is not routinely employed, will be less sensitive than R1 or R2 mapping-based approaches where the baseline subtraction can be readily performed. Thus, R1 or R2 mapping at pre-and post-injection combined with image co-registration could be a preferred approach to quantitatively detect MR contrast agents like 3 nm IONPs. Interestingly, 3 nm IONPs showed both R1 and R2 enhancing effects, indicating its potential role as a dual contrast agent. This suggests that acquiring both R1 and R2 maps would provide improved reliability in assessing tissue localization of the enhancement by 3 nm IONP MR contrast agents.
Besides factors relating to the MR contrast agent itself, the proposed digital simulation methodology heavily depends on the reproducibility of the R1 and R2 maps. In this study, the reproducibility was quantitatively presented using absolute difference maps at a voxel-by-voxel level. Alternatively, the testretest signal to noise ratio (TrTSNR) is also applicable to provide quantitative reproducibility of these imaging methodologies as in our previous report 41 . Thus, reporting the voxel-by-voxel reproducibility of the MR methodology for all MR based biomarker studies is recommended. Another critical imaging related factor is the image co-registration. The quality of co-registration will determine the overall sensitivity of the employed MR contrast agents, especially when the pathology is patchy or sparse.
Importantly, the co-registration quality depends on both raw data reproducibility and spatial resolution. There have been extensive efforts towards the development of improved MR hardware including multichannel RF coil, high power imaging gradients, and high magnet eld for human scanner. The improvements of MR imaging hardware will continue to improve the spatial resolution and ultimately image co-registration quality. One advantage of the R1 mapping over the R2 mapping in this study was the improved co-registration accuracy for the R1 maps; R2 map co-registration suffered from artifacts in the peri-ventricular region, which was not the case for R1 maps.
Our current approach has certain limitations regarding measuring the true R1 or R2 enhancement of 3 nm IONPs in brain tissue. First, our approach is based on assumption that all brain tissues have the same amount of R1 or R2 enhancement from the 3 nm IONPs, which might not be true in vivo. Importantly, the delivery of the contrast agent to the target in vivo has not been addressed. The amount of 3 nm IONPs delivered to speci c brain regions will vary depending on the approaches used to confer BBB crossing and molecular speci city. Furthermore, the half-life of 3 nm IONPs in brain may also affect the R1 or R2 enhancement if its chemical properties change over time. Additionally, it is unknown how long IONPs would remain in the brain. Future studies are planned to address these questions.
In conclusion, the proposed digital simulation provides a useful approach to quantitatively assess the MR signal conspicuity for candidate MRI relaxation contrast agents in humans and mice. The same approach should be applicable to any quantitative MRI like T2 star or T1 rho. In addition, the proposed approach can provide the reliability of proposed alternative MRI acquisition methodologies. Thus, the digital simulation approach will be bene cial for the development of both MRI contrast agents and MRI methodologies. In vivo subjects Human brain MRI images were performed under the National Institutes of Health Institutional Review Board (NIH IRB) approved protocol (NCT00001711). Informed consent was obtained prior to a health subject R1 mapping MR scans, which were conducted in the Clinical Center at the National Institutes of Health (NIH). Human brain MRI image sets for R2 mapping were acquired from a health subject following informed consent at the National Intrepid Center of Excellence. All human related procedures and methods were performed in accordance with the relevant guidelines and regulations.

Relaxivity of 3 nm
All animal experiments were conducted under protocols approved by the National Institute of Neurological Disorders and Stroke (NINDS)/ National Institute on Deafness and Other Communication Disorders (NIDCD) Animal Care and Use Committee in the NIH Clinical Center. All in vivo animal related procedures and methods were performed in accordance with the protocols approved by the Institutional Animal Care and Use Committee (IACUC) at the NIH and were in accordance with ARRIVE guidelines (https://arriveguidelines.org). C57BL6 female mice at 10 weeks of age were purchased from Jackson labs and used at 12-weeks of age. A total 11 mice were used in this study, ve mice at 4.7 T and six mice at 9.4 T. Among the 6 mice scanned at 9.4 T, ve mice, (mouse ID 06-10), were from our previous report 41 . In vivo MRI of mouse brain at 4.7 T and 9.4 T In vivo mouse brain MR scans were performed on Bruker 4.7 T and 9.4 T scanner (Bruker, Ettlingen, Germany) with imaging gradient 180 mT/m for 4.7 T and 260 mT/m for 9.4 T on Paravision 6.01 platform. All employed naïve mice underwent two independent MR scans, MP2RAGE or multi echo T2 scan, within one-week. At 4.7 T, a single channel RF coil was used to acquire MP2RAGE data at 160 µm ×160 µm × 480 µm voxel size, zero-lled to 80 µm ×80 µm ×80 µm, covering the entire brain with 8 averages and 60 minutes total scan time. The thick image slice was selected to increase signal to noise ratio at magnetic eld, 4.7 T. The MP2RAGE imaging parameters were TR/TE/TI1/TI2 (ms) = 8000/3/1100/2600, 9-degree ip angle, and 640 ms segment duration with 2 inversion for a k-space plane. In vivo mouse brain R2 map at 4.7 T was not pursued due to the limitation of gradient maximum rise time (or slew rate). At 9.4 T, 4-channel RF coil used to acquire MP2RAGE data at 160 µm ×160 µm × 160 µm voxel size, zerolled to 80 µm ×80 µm ×80 µm, covering the entire brain with 6 averages and 95 minutes total scan time. The MP2RAGE imaging parameters were TR/TE/TI1/TI2 (ms) = 8000/3/1300/3600, 9-degree ip angle, and 640 ms segment duration with 2 inversion for a k-space plane. The MR protocol of MP2RAGE at 9.4 T is the same as our previous report 41 . The T1 maps of in vivo mouse brain was estimated from raw MP2RAGE and converted into R1 maps. In addition, multi echo T2 weighted images were acquired at 160 µm ×160 µm × 240 µm voxel size, zero-lled to 80 µm ×80 µm ×80 µm, covering the entire brain with TR/ echo spacing/ echo train number/ average/ total scan time = 8.0 s/ 8.0 ms/ 20/ 1/ 10 minutes. The R2 map was estimated using the Bayesian analysis toolbox (http://bayesiananalysis.wustl.edu/). The mouse brain T1 (1/R1) was 1.88 s in the current study. Thus, the short TR might cause T1 contamination on mouse brain R2 map. The R2 value of in vivo mouse brain at 9.4 T dependence on TR was examined by comparing 2D MR sequence having long TR (8.0 s) with 3D having short TR (1.0 s), see Supplementary Fig. 4.
In vivo MRI of human brain at 3.0 T Two different 3.0 T human scanners were used for in vivo human brain MR scan with two normal adult subjects. In vivo human brain MP2RAGE MR scans were performed on a Siemens Prisma 3.0 T magnet (Erlangen, Germany) with 80 mT/m imaging gradient and 32 channel radio frequency (RF) coil. The rst normal adult subject underwent two independent 3-dimensional MP2RAGE scans within one-week, at the same time of the day on each session. The MP2RAGE images were acquired with the following parameters: echo time (TE) = 2.88 ms, repetition time (TR) = 5000 ms, inversion delay times TI1/TI2 = 700/2500 ms, eld of view (FOV) = 256×256×192 mm, matrix = 256×256×192, ip angles = 4-degree for TI1 and 5-degree for TI2, band width = 240 hz/pixel, GRAPPA acceleration factor = 3, total scan time = 8 minutes 56 second. The obtained MP2RAGE data had 1.0 ×1.0 ×1.0 mm isotropic voxel size. The T1 maps of in vivo human brain were estimated from raw MP2RAGE data using the vendor providing imaging handling software, Syngo MR D13D. The obtained T1 maps were converted into R1 maps (= 1/T1).
In vivo human brain T2 mapping scan using T2 weighted imaging (T2WI) sequence were performed on a GE SIGNA 3.0 T magnet (Waukesah, Milwaukee, USA) with 80 mT/m imaging gradient and 32 channel radio frequency (RF) coil. The second normal adult subject underwent two independent T2 mapping MR scans within 2 hours. The second subject took a short break between the 1st and 2nd MR scan.

Data and image processing
For both 4.7 T and 9.4 T, two MP2RAGE data were acquired from the same mouse within one-week time frame producing two independent R1 maps. Using the advanced normalization tools (ANTs, http://stnava.github.io/ANTs/) 47 , the 2nd R1 map was co-registered to the 1st R1 map. After coregistration the absolute ΔR1 (|ΔR1|) map between 1st R1 map and the 2nd R1 map co-registered to 1st R1 map was calculated, producing a baseline. The hypothesized R1 enhancing effects on in vivo mouse brain induced by 3 nm IONPs were simulated by digitally adding R1 values to the 2nd R1 map coregistered to 1st R1 map. This approach simulates the result that could hypothetically be obtained if a subject were scanned at baseline, administered an IONP contrast agent, and then scanned again at an appropriate time. The hypothetical R1 enhancement of the 3 nm IONPs in mouse brain was calculated from equation [1].
The digital simulated additional R1 values were randomly placed in both small (from 1 to 8 voxels) and larger (from 20 to 30 voxels) zones in the cortex of the 2nd R1 map co-registered to 1st R1 map. It was assumed that all brain tissues have the same degree of R1 enhancing effect from the same concentrations of 3 nm IONP. This procedure was performed for ve hypothetical IONP iron concentrations: 0.02, 0.04, 0.06, 0.08, and 0.1 mM. Similar to the baseline |ΔR1| map, the |ΔR1| maps between the 1st R1 map and the co-registered 2nd R1 map having the hypothesized effects of 3 nm IONP were calculated to assess the sensitivity of MP2RAGE derived R1 maps to 3 nm IONP contrast agents. In vivo mouse brain R2 maps underwent the same procedures where the hypothetical R2 enhancement of the 3 nm IONPs in mouse brain was calculated from equation [2]. ΔR2 = r2 (mM − 1 s − 1 ) × [Fe] (mM) [2] The same procedures were performed on in vivo human brain R1 maps obtained at 3.0 T. Following image co-registration, the baseline |ΔR1| was obtained. The digital simulated additional R1 values, which were calculated from the r1 of the 3 nm IONPs at 3.0 T, were randomly placed in both small (from 1 to 2 voxels) and larger (from 3 to 9 voxels) zones in representative human brain regions. The |ΔR1| map between the 1st R1 map and co-registered 2nd R1 map after digitally added R1 equivalent to several concentrations of 3 nm IONP were calculated. Analogous procedures were conducted using R2 maps. Figure 1 The digital simulation work ow. a, b. two independent R1 maps (1st and 2nd R1 map) from the same  In vivo mouse brain R1 maps after digital simulation of additional R1 and the absolute ΔR1 (|ΔR1|) maps for low (4.7 T) and high (    In vivo mouse brain R2 maps with digital simulated R2 effects of IONPs and absolute ΔR2 (|ΔR2|) maps at 9.4 T.  IONPs. The 2nd R1 map was co-registered to the 1st R1 map prior to digital simulation. d-ii -o-ii. The absolute ΔR1 (|ΔR1|) maps. Single voxels in hippocampus are conspicuous at R1 equivalent to 0.04 mM [Fe] or higher in the subtraction maps, but much less prominent without baseline scan subtraction.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. DigitalSimulationSupplement20211stSubmission.docx