Structural correlates of active-staining following magnetic resonance microscopy in the mouse brain

Extensive worldwide efforts are underway to produce knockout mice for each of the ~ 25,000 mouse genes, which may give new insights into the underlying pathophysiology of neurological disease. Microscopic magnetic resonance imaging (μMRI) is a key method for non-invasive morphological phenotyping, capable of producing high-resolution 3D images of ex-vivo brains, after fixation with an MR contrast agent. These agents have been suggested to act as active-stains, enhancing structures not normally visible on MRI. In this study, we investigated the structural correlates of the MRI agent Gd-DTPA, together with the optimal preparation and scan parameters for contrast-enhanced gradient-echo imaging of the mouse brain. We observed that in-situ preparation was preferential to ex-situ due to the degree of extraction damage. In-situ brains scanned with optimised parameters, enabled images with a high signal-to-noise-ratio (SNR ~ 30) and comprehensive anatomical delineation. Direct correlation of the MR brain structures to histology, detailed fine histoarchitecture in the cortex, cerebellum, olfactory bulb and hippocampus. Neurofilament staining demonstrated that regions of negative MR contrast strongly correlated to myelinated white-matter structures, whilst structures of more positive MR contrast corresponded to areas with high grey matter content. We were able to identify many sub-regions, particularly within the hippocampus, such as the unmyelinated mossy fibres (stratum lucidum) and their region of synapse in the stratum pyramidale, together with the granular layer of the dentate gyrus, an area of densely packed cell bodies, which was clearly visible as a region of hyperintensity. This suggests that cellular structure influences the site-specific distribution of the MR contrast agent, resulting in local variations in T2*, which leads to enhanced tissue discrimination. Our findings provide insights not only into the cellular distribution and mechanism of MR active-staining, but also allow for three dimensional analysis, which enables interpretation of magnetic resonance microscopy brain data and highlights cellular structure for investigation of disease processes in development and disease.


Introduction
An ever-increasing number of mice are being created through genetic manipulation to advance the goal of translational neuroscience to produce and identify mouse models of human conditions. In order to investigate and validate these mouse models, detailed studies must be undertaken of the mouse brain. Phenotyping the brain conventionally relies on examination of cellular and anatomical structures by histology. While this method is ideal to visualise physiological or pathological changes on a cellular and architectural level, tissue processing artefacts and shrinkage affect the reliability of 3D volumetric assessments. Effective methods for rapid, non-invasive phenotyping are thus needed for the analysis of mouse mutants.
In recent years these efforts have been advanced by the increasing use of microscopic MRI (μMRI) to characterise brain phenotypes of novel mouse mutants. MRI offers a number of advantages over traditional histological sectioning, namely its ability to produce non-invasive datasets, which allow the accurate calculation of volumes without distortion. Morphometric techniques may then be used to allow the identification of novel phenotypes. Work in this field has lead to the creation of non-invasive mouse brain atlases (Dorr et al., 2008;Ma et al., 2005) and has generated new insights into models such as Huntingdon's disease and age-related neurodegeneration Sawiak et al., 2009b;Wetzel et al., 2008).
μMRI produces high resolution datasets of ex-vivo tissue with isotropic resolutions of less than 100 μm (typically between 21.5-70 μm in most brain studies (Badea et al., 2007b;Sawiak et al., 2009a). This can be technically challenging, with each voxel representing a tissue volume of b1 nL, conventionally requiring long scan times to obtain sufficient signal-to-noise (SNR). The innovation of tissue active staining, where gadolinium-chelate MR contrast agents are used during the fixation process to shorten T 1 relaxation, has enabled the acquisition of high SNR 3D datasets with greatly reduced scan times (Johnson et al., 2002a).
The appearance of μMR images acquired in this manner is sensitive to preparation and scan parameters including the contrast agent, fixation concentration and pulse-sequence used (Cleary et al., 2009). It has been recently shown that the use of higher contrast agent concentrations (N5 mM Gd-DTPA) as part of the fixation of mouse brains excised from the skull (ex-situ), in combination with a 3D gradient-echo sequence, can produce images with high contrast, delineating a number of brain structures seen on conventional histology (Huang et al., 2009;Kim et al., 2009).
The combination of a high contrast agent concentration with in-situ preparation may produce enhanced neuroanatomical contrast and allow more detailed and sensitive morphometric atlases, with clearer delineation of regions of phenotypic change. Resulting volume measurements and morphometric analysis may also be less affected by preparation than ex-situ brains (Badea et al., 2007a). However, there is little data in the literature comparing ex-and in-situ brains by MRI. Furthermore, the timecourse for optimal fixation of in-situ brains with an MR contrast agent is unknown and a direct histological correlation of brain structures enhanced through gradient-echo imaging together with active-staining has not been performed.
The specific aims of our study were to investigate enhanced tissue contrast using an active-stain μMRI preparation and optimised scanning protocol (Cleary et al., 2009) for imaging adult mouse brains. Specifically we have assessed the effect of active-staining on perfuse-and simple immersion-fixed ex-situ brains and monitored the in-situ MR tissue characteristics over a period of 9 weeks. Furthermore, we sought to identify the optimal imaging parameters to enable rapid imaging of insitu brains in 3 h with a standard imaging coil. In particular we have assessed contrast-enhanced brain structures via a direct comparison with tissue histology, which may give insights into the cellular distribution and mechanism of MR active-staining.

Ex-situ brain study
In order to investigate the preparation of excised brains we compared the effects of perfusion-fixation to simple immersionfixation. Two of 4 mice were given an initial saline flush (15-20 ml of normal saline at a flow rate of 3 ml/min introduced via a needle in the left ventricle) and then perfusion-fixed with 50 ml of 4% buffered formol-saline (Pioneer Research Chemicals, Colchester, UK) with 8 mM Gd-DTPA (Magnevist, Bayer-Schering Pharma, Newbury, UK) at a flow rate of 3 ml/min; the remaining 2 mice were simply culled. All 4 brains were then carefully extracted from the skull and post-fixed in 4% buffered formol-saline (with 8 mM Gd-DTPA) at 4°C.

In-situ study
For in-situ brain MRI, 4 mice were perfusion-fixed as previously described with 2 fixed using formol-saline perfusate alone and 2 with formal-saline with 8 mM Gd-DTPA. After this, mice were decapitated and skin, muscle, lower jaw, tongue, nasal bones and zygomatic arches removed, with the remaining intact skulls post-fixed in 4% buffered formol-saline (with 8 mM Gd-DTPA) at 4°C.

Imaging
All imaging was performed on a Varian 9.4 T DirectDrive VNMRS system (Varian Inc., Palo Alto CA, USA) with a 26 mm quadrature birdcage coil (RAPID Biomedical GmbH, Würzburg, Germany). Samples were removed from fixative and excess solution carefully blotted with a paper towel. They were then immersed in Fomblin perfluoropolyether (type PFS-1, Solvay Solexis S.p.A., Bollate, Italy) in 20 ml plastic syringes and immobilised with surgical gauze. Samples were allowed to equilibrate for at least 2 h at room temperature prior to imaging. Air bubbles in samples were minimised by the equilibration time and gentle agitation of the syringe.
The effect of echo time on brain contrast was also investigated experimentally by repeated imaging of one brain at 5 weeks fixation with 5 different echo-times (3.8-7 ms).
Image processing T 1 and T 2 * maps were created by a pixel-by-pixel, least-squares fit of the resulting data using in-house C-programs. The inverse of these were also calculated, using ImageJ (National Institutes of Health, Bethesda MD, USA), producing corresponding maps of R 1 and R 2 *. Statistical analysis of map values was performed in Prism v5.00 (GraphPad Software, San Diego CA, USA). 3D volume images were reconstructed and converted to Analyze format using custom Matlab code and reviewed in ImageJ. SNR and CNR were measured in a number of brain regions by taking the ratio of average and signal differences in wholestructure volumes and the standard deviation in a region of background noise Eq. (2): Visualisation of brain structures was performed in Amira software (v5.2.2, Visage Imaging Inc., Andover MA, USA).

Results
Ex-situ study: comparison of immersion and perfusion-fixed brains at two fixation durations.
After 7 days immersion in contrast-fixative, there was no significant difference in T 1 and T 2 * values between the immersion and perfusion-fixed brains (p N 0.40 in all comparisons) (Table 1). T 1 and T 2 * measurements were repeated at 2 weeks ( Fig. 1). We observed a similar change in whole-brain T 1 (perfusion: 33.2 to 31.4 and immersion: 35.1 to 31.7) and T 2 * (perfusion: 3.7 to 3.5 and immersion: 3.5 to 3.2) between 7 days and 2 weeks (Fig. 1). We noted that the immersion-fixed brains had sustained some additional cortical damage during the extraction process when compared to the perfusion-fixed brain (Fig. 2). Furthermore, cortical vessels, although present in both immersion and fixed brains, were more apparent in the perfusion-fixed brains, otherwise the brains from the two methods appeared similar. This comparison indicates that the use of a simpler immersion-fixation method, where the brain is removed directly from the skull without perfusion-fixation, would facilitate magnetic resonance microscopy techniques, if damage from extraction is not a confounder.

In-situ study
Given that varying degrees of cerebellar and cortical damage was apparent in all brains imaged, we explored the feasibility of contrastenhanced imaging with the brain in-situ. Initially we investigated whether the addition of 8 mM Gd-DTPA in the initial perfusate would improve the rate of T 1 reduction over perfuse fixation and immersion fixation alone. After measurement of T 1 and T 2 * values in each group after 1 week fixation, it was apparent that there was no significant difference (pN 0.65) in T 1 data: therefore all brains were pooled in subsequent analysis. We measured MR parameters over 1, 2, 3, and 5 weeks. Table 2 demonstrates that T 2 * values were found to reach a minimum at 3 weeks immersion (mean thalamus-midbrain = 2.9 ms, cortex = 3.8 ms). In contrast, T 1 values continued to reduce after 3 weeks, approaching uniformity across the brain by the 5 week timepoint (mean thalamus-midbrain region = 48± 3 ms, cortex = 43 ± 2 ms, Table 2). Fig. 3 illustrates the corresponding regional R 1 changes occurring over 5 weeks. We observed that a greater immersion time lead to a lower T 1 (e.g. whole-brain T 1 : 1 week 109 ± 53 vs. 5 weeks 44 ±2 ms, pb 0.05, Table 2) and also reduced the magnitude of T 1 difference between cortical and midbrain-thalamus structures deep in the centre of the brain (e.g. mean difference at 1 week= 15 ms vs. 5 weeks=5 ms, Table 2). An increase in the conspicuity of white matter structures (such as fibres of the internal capsule) was observed on 3D images with increasing immersion time (Fig. 3).

Scan parameter optimisation
From calculations using mean T 1 across the whole-brain at 5 weeks immersion, we determined that parameters of TR = 17 ms, NSA = 6, at the whole-brain Ernst angle (47°) would produce the highest SNR (assuming constant noise), in a 3 h scan time. From the T 2 * data at 1 week we estimated the ratio of M 0 in the corpus callosum relative to cortex in the 4 brains to be 0.96 ± 0.01. Using these estimates and T 2 * values from our maps at 5 weeks, we calculated that a TE of approximately~3.4 ms would give optimal cortex/corpus callosum contrast. To ascertain the optimal TE at 5 weeks experimentally, we performed repeated high-resolution scans of a single brain from our dataset with the calculated TR, NSA and flip angle above. Fig. 4 shows the effect of varying TE from 3.8 ms (the minimum permitted by the system at 100 kHz bandwidth) to 7 ms. Although SNR in all structures was highest at TE = 3.8 ms (whole-brain SNR = 31.5), cortex-corpus callosum CNR appeared to be maximal at TE = 4 ms (CNR = 7.5, whole-brain SNR = 30.3). Interestingly white-grey matter CNR in the cerebellum peaked slightly later at a TE = 5 ms (CNR = 22.4). As indicated in Fig. 5, it was noted that increasing echo time appeared to enhance the delineation of more subtle structures, such as cortical layers and the layer of Purkinje cells in the cerebellum, indicating a T 2 * dependency to this contrast. Opting for a balance of high SNR and high grey-white matter CNR, we chose final parameters of TE = 4 ms, FA = 47°(the Ernst Angle), TR = 17 ms and NSA = 6 averages for future studies. Table 1 Table showing T 1 and T 2 * values in brains after 7 days immersion only or with initial perfusion-fixation (mean values ± SD).

Brain MR parameters at 9 weeks
After identifying the need for a long immersion duration to enable full penetration of contrast agent, we applied the above optimised parameters (FA = 52°, appropriate for these T 1 s) to a further 4 in-situ brains after 9 weeks immersion, acquiring T 1 and T 2 * maps and highresolution 3D volumes. Compared to 5 week data, while R 2 * values remained similar in both cortical and thalamus-midbrain regions to our previous brains (Fig. 6B), R 1 showed a more marked increase (Fig. 6A), with the corresponding whole-brain T 1 measured as 35 ± 2 ms compared to 44 ± 2 ms of previous brains at 5 weeks. As seen in a corresponding R 1 map of a sagittal slice through an example brain (bottom right, Fig. 6), values have a narrow distribution across the brain (SD of T 1 values over whole-brain ROI = 3.8 ± 0.1 ms, n = 4) indicating that 9 weeks immersion may be more suitable. Wholebrain SNR was comparable to previous 5 week data and measured to be~30.

Structural correlates of active-staining in the brain
The resulting gradient echo images acquired at the selected parameters (TE/TR/FA/NSA = 4 ms/17 ms/52°Ernst angle/6 averages), after 9 weeks in contrast-fixative, demonstrated excellent contrast, enabling the delineation of a number of structures in the cortex,  hippocampus, olfactory bulb and cerebellum (Fig. 7, sagittal image). In the hippocampus, individual layers were clearly visible such as a bright granular cell layer of the dentate gyrus, as well as bright region of the stratum lucidum and dark layer in the region of CA3 (Fig. 7, coronal view A, blue panel). Individual nuclei were also visible, including medial and lateral habenular nuclei and a number of thalamic nuclei (Fig. 7, coronal view A, green panel). White-matter tracts were easily visualised, appearing with strong negative contrast on gradient-echo images (Fig. 7, coronal view B).
Sagittal MR images were compared to histology where a good correlation was observed between a number of white matter structures defined on a neurofilament stained section and equivalent dark areas on a similar MR image (Fig. 8A). Structures such as the anterior commissure, superior colliculus commissure and fasciculus retroflexus were easily visualised.
In the cerebellum, (Fig. 8B) structures on MR correlated well to those seen on histology, such as the bright outer molecular layer, consisting of mostly Purkinje cell dendrites (seen on calbindin section, Fig. 8B), a bright granular cell layer and dark Purkinje cell layer (evenly-spaced, large Purkinje cells visible in all sections, Fig. 8B) and axonal fibre tracts (Fig. 8B, neurofilament section). While MR signal intensity was high in both molecular and granular layers, the granular layer appeared to be slightly hypointense.
In the olfactory bulb (Fig. 8C) the main features of mitral, glomerular and external plexiform layers were easily identified in MR images, with these regions correlating well to all stains. Sublayers in the granular cell layer, seen on H&E and calbindin sections (Fig. 8C), were not readily apparent on MR with images best correlating to the more uniform appearance of the granular cell layer observed in the neurofilament section.
As hippocampal anatomy was especially well defined on MRI, we sought to identify the source of the regional MR contrast by histological comparison. In addition to standard histological stains, we also compared our data to a standard Timm stained section-an established method of visualising the presence of intracellular Zn 2+ that characterises hippocampal mossy fibres (Haug et al., 1971). Fig. 9A shows a comparison of hippocampal structures with a similar axial Timm section from a C57BL/6 mouse (Crusio et al., 1986). We observed a conspicuous area of hypointensity in the stratum pyramidale region of CA3 (orange arrows, Figs. 9A and B), which correlated with unmyelinated mossy fibres synapsing with pyramidal cells (Figs. 9A(iv) and B, Timm and calbindin sections respectively). Previously, we had observed that hypointensity was associated with myelinated white matter structures (Fig. 8A), which is not the case in the stratum pyramidale. An area of myelinated fibres in the dentate gyrus molecular layer (Fig. 9B, blue arrows) did not result in a hypointensity on MRI. Closely associated and lateral to the stratum pyramidale (concave aspect; red arrows) are the suprapyramidal mossy fibres (running in the stratum lucidum) a region of unmyelinated axonal fibres (Figs. 9A and B, Timm and calbindin sections) that appeared as an area of hyperintensity within CA3. The granular layer of the dentate gyrus (green arrows, Fig. 9A) was clearly visible as a region of hyperintensity that correlated to a dense area of closely packed cell bodies (Fig. 9B, H&E and calbindin sections).

Discussion
After an initial investigation of ex-situ brains which showed a high degree of extraction damage. We have demonstrated that in-situ brains prepared in a solution of 8 mM Gd-DTPA and imaged with gradient-echo sequence produced high-resolution (40 μm isotropic), high SNR images, with a 3-hour scan time. We have also shown that structures delineated on MR directly correlate to those defined on histology, such as major white matter tracts and layers of the olfactory bulb and hippocampus. This comparison has then enabled us to assess the effects of tissue microstructure on MR contrast of actively-stained tissue.

Ex-situ brain imaging
Ex-situ brain imaging with a gradient-echo sequence has been reported previously (Huang et al., 2009;Kim et al., 2009) and our ex-situ brains show T 1 and T 2 * values are in line with those studies (T 1 ≈ 30 ms, T 2 ≈ 7 ms, 10 mM Gd-DTPA, after 4 days immersion in contrast-fixative ) displaying similar image contrast. Additionally we  . 3. Comparison of representative R 1 maps and similar volume slices from the same in-situ brain over 5 weeks fixation. As seen in R 1 maps (top) a greater amount of time is required for values in the centre of the brain to equilibrate with more cortical regions. Although a modest SNR enhancement may be seen in 3D gradient-echo images after successive weeks (bottom), there is a noticeable improvement in the quality of structural delineation, particularly in the hippocampus (blue-arrows) and white matter structures (e.g. internal capsule, red arrows; mammilothalamic tract, yellow arrows). (Images identically scaled).
have shown that these values remain relatively unchanged even after extending the immersion time to two weeks (e.g. mean T 1 perfusionfixed went from 33 to 31 ms) to leave more time for Gd-DTPA penetration. We have demonstrated that brains need not be perfusion-fixed and that simple immersion-fixation produced equivalent image quality, thus enabling magnetic resonance histology in laboratories where perfusion-fixation is not readily available. Imaging of extracted, ex-situ brain tissue does have some advantages, in that smaller RF coils can be used, improving SNR and a smaller field of view may improve imaging-time ). However the tissue damage observed in our ex-situ brains may compromise their use for automated and quantitative morphometric analyses where accurate anatomical volumes and morphology are essential for sensitivity.

In-situ brain optimisation
Given the effect of damage in ex-situ imaging we have focussed on in-situ brain preparation and MR methods optimisation to produce detailed images of the mouse brain. We demonstrated that full penetration of Gd-DTPA into brain tissue, as assessed by R 1 values in the cortex and basal ganglia, had equilibrated across the brain by 9 weeks. Indicating that a much greater immersion time is required than in the ex-situ brain in order to produce the maximal increase in SNR, and thus image quality, across the brain.
Currently there is no standard protocol for in-situ brain preparation and scan parameters, with a number of approaches in current use. A T 2 -weighted 3D fast spin-echo sequence (FSE) with brains fixed in Fig. 4. Representative sagittal slice (A) and measured SNR (B) and CNR (C) from 3D volume data of a brain imaged after 5 weeks fixation at five different echo times. 2 mM gadoteridol (a non-ionic macrocyclic Gd-chelate) for at least 7 days (Tyszka et al., 2006) is able to produce T 2 -weighted images of multiple in-situ brains with a 32 μm resolution in a 11.3 h scan, and has been successfully used for morphometric phenotyping (Ellegood et al., 2010) and atlasing (Dorr et al., 2008) studies. While this can offer good T 2 -weighted images with an improved imaging time, the use of an FSE sequence can lead to image blurring and a loss of structural definition (Mulkern et al., 1990).
Partial-Fourier spin-echo has also been successfully used for in-situ phenotyping studies giving T 1 -weighted images with 21.5 μm resolution in a~2 h scan time (Badea et al., 2007b). Sharief and Johnson (2006) developed an innovative multi-echo T 2 sequence where reconstructed images combine data from a number of echo times. This has the effect of improving contrast, especially in T 2 -dependent structures such as in the cortex and brain nuclei. In both these imaging protocols, brains are initially perfusion-fixed with a high concentration (50 mM-assuming a standard 0.5 M clinical solution) of gadoteridol and fixed overnight 10% formalin prior to imaging.
Our results indicate that, in the in-situ brain, such short immersion periods may not be sufficient to give maximal T 1 reduction (and thus SNR gain) over the whole brain. If we assume that a T 1 of~31 ms (2week ex-situ data) is a likely minima for values after Gd immersion, we nearly achieve this level after 9 weeks (whole-brain T 1~3 5 ms). The need for such an increase in immersion time compared to ex-situ brains for Gd penetration is most likely due to the presence of intact meningeal layers and limited entry points for fixative into the skull (e.g. through the foramen magnum and orbits) resulting in a greatly reduced exposed brain tissue surface-area.
Although a virtue of gradient-echo imaging is a short TR and thus reduced acquisition time, this makes it sensitive to T 1 differences. The heterogeneity seen in R 1 maps, corresponding to higher T 1 values in more central structures compared to cortical and cerebellar regions, will thus limit T 1 recovery, causing signal reduction in these portions of the brain. As the majority of these regions, such as basal ganglia and midbrain contain white-matter fibres, there is likely to be an intrinsically lower signal due to the lower T 2 of myelinated structures (Counsell et al., 2003). Thus it is important to ensure there is adequate agent penetration to ensure full T 1 recovery in these regions. Although Fig. 6. Graph showing the timecourse of R 1 (A) and R 2 * change (B) with immersion time in fixative + Gd-DTPA solution, incorporating values from the previous 5 week study and 9 week brains indicating a minima has been reached. Representative R 1 maps from 5 and 9 week brains shown for illustration (C). TR could be increased to compensate, we have previously shown that increased signal averaging is generally preferable to small TR increases to boost SNR (Cleary et al., 2009) and is thus not ideal when imaging with a fixed scan time. The benefits of allowing a greater degree of T 1 reduction are readily seen in our gradient-echo images taken over 5 weeks (Fig. 5), in which structures of the hippocampus   9. (A) Hippocampal view (ii) from an axial MRI slice (i) compared with a representative section from a C57BL/6 mouse hippocampus stained with Timm silver sulphide stain (iv), showing regions of high ionic zinc concentration, a feature of hippocampal mossy fibres. Red arrows indicate correspondence with mossy-fibre bundles, orange with pyramidal cells and green with the granular cell layer of the dentate gyrus (Timm section adapted from (Crusio et al., 1986). (B) Sagittal view of hippocampal anatomy from a brain fixed for 9 weeks compared with H&E, calbindin and myelin basic protein (MBP) stained sections. Dark regions on MR (orange arrows) appear to correlate with heterogeneous regions containing mossy fibres and pyramidal cells. Corpus callosum (white arrows) indicated for comparison. Blue arrows in the MBP section indicate a diffuse region of myelinated fibres in the molecular layer of the dentate gyrus. and white matter tracts are more conspicuous by the last 5 week timepoint.
Although active-staining primarily acts to reduce tissue T 1 due to administration of contrast agent (Johnson et al., 2002b;Petiet et al., 2007) there is also a corresponding reduction in T 2 (Cleary et al., 2010;Petiet et al., 2007). This may also be exploited to produce image contrast using much shorter TEs than in normal tissue.
In our method, the use of a relatively high contrast agent concentration, leads to a large mean reduction and compression of the whole range of normal tissue T 2 values. For example, as T 2 * is now 3.8 ms as measured in the cortex (in-vivo T 2 ≈ 45 ms at 9.4 T (Uematsu et al., 2007)), our chosen echo-time of 4 ms results in T 2 *-weighting in that region, allowing cortical layers to be visualised. Similarly, increased T 2 *-weighting, improves visualisation of the Purkinje layer of the cerebellum. Thus an 8 mM Gd-DTPA concentration with a gradientecho sequence with a relatively short TE and TR combines the potential of a short scan time and a high number of signal averages and an optimal T 2 * weighting to reveal tissue contrast.
The contrast observed in our gradient echo images may be explained by a combination of intrinsic tissue MR relaxation and effect of contrast agent concentration. For example, myelinated white matter is known to have a lower T 2 than grey matter, due to higher bound water content (Counsell et al., 2003), giving a characteristically darker signal on MR images in-vivo, which we also observed in our data. This is probably best explained by contrast agent reducing T 2 values but still preserving T 2 differences between tissues, which can be exploited by the correct MR parameters. Yet it has also been suggested that distribution of the agent itself may be enhancing contrast, particularly in non-myelinated structures (Huang et al., 2009). Although this is unlikely to account for our observed white matter contrast: using R 1 maps as an indication of contrast agent concentration, the lack of regional contrast in maps at 9 weeks does not suggest a selective regional distribution of contrast agent, where higher concentration of contrast agent would be expected to result in area of increased R 1 .
Interestingly we also observed negative contrast in classically unmyelinated regions of the hippocampus that did not appear to depend on the presence of even small traces of myelin (as seen in our MBP-stained section, Fig. 9B, blue arrows). Previous work in the ex-situ mouse brain has suggested that~23% of brain volume is inaccessible to the Gd-DTPA indicating that the agent is not distributed uniformly at a cellular level (Huang et al., 2009). Our study reinforces this view, as the dark appearance of these unmyelinated regions are likely to be due to increased local Gd-DTPA distribution and MR susceptibility at a local level. Thus the choice of a gradient echo sequence is ideally suited to detection of such regions, especially those with subtle T 2 * differences.
Both intrinsic contrast and susceptibility could explain the subtly different appearance of molecular and granular layers of the cerebellum on MR. Although the granular layer contains a great number of densely packed cells, as seen on the neurofilament stain, there are still a large number of axonal fibres passing through this region. This could result in the observed hypointensity either through intrinsic relaxation mechanisms where the presence of such myelinated tracts, with a lower T 2 *, leads to a partial volume effect with a slight loss in signal, or the regional cytoarchitecture, comprising a mixture of cell bodies and axons, could affect local Gd-DTPA distribution, causing a susceptibility effect.
Similarly we observed an enhancement of the layer of Purkinje cells with increasing echo time. This layer also contains a mixture of large Purkinje cells surrounded by basket cell axons. However we have observed a large T 2 * dependence on contrast in this region, indicating that susceptibility effects through structural heterogeneity play a greater role, probably due to distribution of contrast agent.
In the olfactory bulb, we observed contrast that delineated the mitral and glomerular but not granular layers. In previous in-vivo high-resolution T 2 -weighted images (Boretius et al., 2009) all of these regions were observed including granular layers. It is conceivable that the more complex cytoarchitecture of both glomerular and mitral layers may explain this observation. Glomeruli are relatively large structures that consist of a number of nerve synapses, surrounded by glial cells (Kosaka et al., 1998); also mitral cells have axonal fibres that contribute to the olfactory tract (Shepherd et al., 2007). In contrast, granule cells possess only dendritic processes and are without axons (Price and Powell, 1970). This more complex structure may lead to greater susceptibility effects in the presence of the contrast agent in these regions.
Comparisons with previous diffusion-weighted (Shepherd et al., 2006) and in-vivo work (Boretius et al., 2009) may offer some explanation for the enhanced contrast seen in the hippocampus. The area of the stratum lucidum has been seen as hyperintense on diffusion images in the rat (Shepherd et al., 2006) indicating a region of restricted diffusion that may affect the access of Gd-DTPA to the region and thus maintain a longer T 2 * than surrounding tissue. Similarly, the dense packing of granular cells in the dentate gyrus may prevent infiltration of contrast into the region and also maintain a long T 2 * resulting in its bright appearance.
Previous work (Benveniste et al., 2000) has shown contrast enhancement in the hippocampus on subsequent ex-vivo images monitored over a period of 2 to 50 h after initial in-vivo perfusionfixation with Gd-DTPA. Images acquired at 2 h post fixation showed a general increase in signal over the whole hippocampus, but the granular layer of the dentate gyrus appeared particularly hyperintense. The stratum pyramidale was visible as a dark band. Over subsequent timepoints, while signal enhancement over the whole hippocampus appeared to be maintained, the conspicuity of both granular layer and stratum pyramidale was greatly reduced. This evolution of contrast was attributed to the vascular nature of this region, with intravascular retention and extravasation of contrast agent leading to the appearance of reduced and increased signal respectively. The initial enhancement seen in the granular layer may indicate that the hippocampus is compartmentalised, possessing regions accessible to the contrast agent to a greater or lesser degree. High-resolution in-vivo imaging, without contrast, has also demonstrated delineation of hippocampal structures in T 2 -weighted images, including the stratum pyramidale, and granular and polymorphic layers of the dentate gyrus (Boretius et al., 2009). In images with strong weighting (TE = 82 ms), contrast in these regions was markedly different to that of ours as both dentate gyrus and stratum pyramidale appeared hypointense, with the stratum pyramidale visible as a continuous line from CA3 to CA1. Also the granular layer of the dentate gyrus was not easily distinguished from the polymorphic layer. These findings suggest that in certain structures local accumulation of the contrast agent can contribute to MR appearance more than intrinsic tissue T 2 * contrast. This suggests the exciting possibility that targeting such regions could provide the basis for the investigation of novel MR microscopy stains.
While the presence of contrast agent is likely to contribute most to changes in tissue parameters, an additional consideration is tissue fixation. Aldehyde fixatives cause the cross-linking of proteins (Fraenkel-Conrat and Olcott, 1948), reducing water mobility, and resulting in a reduction of tissue MR relaxation times (Dawe et al., 2009). A study in the rat cortical slice model (Shepherd et al., 2009) found T 1 reduced 21% after N10 day fixation, although an 81% T 2 reduction could be reversed after washing out the fixative by immersion in PBS. Timecourse studies in fixed human brain have also shown progressive decreases in T 1 and T 2 , with the T 1 change occurring fastest over the initial 1-2 weeks and a slower decline after 5 weeks (Tovi and Ericsson, 1992). Grey and white matter T 1 (Pfefferbaum et al., 2004) and T 2 values (Blamire et al., 1999) also tend to converge with a longer fixation time. There are, however, few studies in the literature assessing the effect of fixation on the use of Gdchelates. Work in Xenopus oocytes (Purea and Webb, 2006) indicates that fixation may alter cell membrane structure, enabling the entry of contrast into the cytoplasm. However, further experiments are needed to examine the role of fixation on contrast in active-stained μMRI of the mouse brain.
The time needed for the penetration of agent in our methodology may be a disadvantage, as histological techniques, especially in light of developments in automatic sectioning systems, can provide more detailed information in a similar amount of time. However, MRI does offer the benefits of non-destructive 3D image acquisition.

Conclusion
We have shown that active-stained μMRI can provide detailed anatomical images of the in-situ mouse brain with high resolution and signal-to-noise. Additionally, we have used a variety of stains to identify and compare the MRI appearance of brain structures to those defined by histology. Combining this data, we have highlighted possible mechanisms for the enhanced contrast in these regions, surmising that contrast depends on both preserved, intrinsic T 2 * differences as well as regions delineated by distribution of active-stain due to tissue microstructure. This methodology could enable greater sensitivity for the phenotypic characterisation in mutant mouse models and enhance regions of the mouse brain that may be targeted in future transgenic studies.