A time-course study of actively stained mouse brains: DTI parameter and connectomic stability over one year

Abstract

While the application of diffusion tensor imaging (DTI), tractography, and connectomics to fixed tissue is a common practice today, there have been limited studies examining the effects of fixation on brain microstructure over extended periods. This mouse model time-course study reports the changes of regional brain volumes and diffusion scalar parameters, such as fractional anisotropy across twelve representative brain regions as measures of brain structural stability. The scalar DTI parameters and regional volumes were highly variable over the first two weeks after fixation. The same parameters were consistent over a two to eight-week window after fixation which means confounds from tissue stability over that scanning window are minimal. Quantitative connectomes were analyzed over the same time with extension out to one year. While there is some change in the scalar metrics at one year after fixation, these changes are sufficiently small, particularly in white matter to support reproducible connectomes over a period ranging from two weeks to one year post fixation. These findings delineate a scanning period during which brain volumes, diffusion scalar metrics and connectomes are remarkably consistent.

GRAPHICAL ABSTRACT

This test/retest diffusion weighted (DW) MRI connectomics study determines the time period post-fixation that an actively perfused mouse brain (C57BL/6J) can be remeasured and still maintain consistency in result. We determined from a mix of analysis of connectomics and DW metrics that within the three distinct time periods, early (prior to 2 weeks), middle (2 weeks to 8 weeks), and late (post 8 weeks), the early time period is the most unstable of the three time periods.

INTRODUCTION

The use of diffusion tensor imaging (DTI) has grown over the past few years, spawning technical advances¹ and applications in connectomics^2–4. In multiple human populations studies (eg. the human connectome project, the ENGIMA consortium and the ADNI initiative), DTI has proven particularly useful in understanding human cognition, teasing out genetic differences and understanding pathology ^5–8. The application of DTI has expanded into animal studies, where the scalar parameters have proven useful in understanding structural changes of the brain during maturation, changes that are genetically driven, and changes due to disease or injury^{9 10}.

The development of disease models and the genetic control that mouse studies offer provide a controlled approach to the understanding of disease mechanisms not possible in a clinical setting. Postmortem studies allow a level of imaging precision otherwise unachievable in human or live animal studies, since one can employ specialized staining methods and long scan times^11–13. However, the study of postmortem tissue brings its own unique challenges. Diffusion in unfixed postmortem tissues changes over time due to autolysis¹⁴. Fixation, which affects water diffusivity in tissue and promotes cross-linking of proteins¹⁵, can be used to preserve and stabilize tissue architecture. Consequently, the errors associated with a postmortem study, like tissue decomposition, can be minimized.

Active staining introduces fixative and contrast agent through direct cardiac infusion of the live animal minimizing tissue destruction from autolysis^12,16,17. Active staining reduces the spin lattice relaxation time (T1) using a gadolinium contrast agent, leading to improved signal to noise, spatial resolution, and image contrast. However, until the fixative diffuses uniformly throughout the tissue, the brain remains susceptible to autolysis^15,18. Crosslinking induced by fixation takes place on a time scale of minutes to years during which the relaxation and diffusion properties can undergo profound changes^18,19. Furthermore, the effects of fixation on spin-spin relaxation (T2) over time post fixation^19–22 can contribute to variability in DTI measures in several ways. Fixation shortens T2 which results in reduced signal to noise. The change in T2 over the course of the study is our major concern. Fluctuating T2 during a 12- hour imaging session would be a major confound by contributing to variation in the DTI measurements at different angles. Young-Hing et al ²² studied this in some detail. They found T2 shortening in white matter (at 1.5T) from 80 ms to ~ 40 ms over ~ 770 hours. Their work was done with immersion fixation of whole human brains so diffusion of the fixative into the tissue probably contributes to the relatively slow change. With a faster fixation method, like the active staining used in this study, changes in T2 between scans performed at 1 day versus 1 week (168 hours) pose a potentially more important confound than that identified by Young-Hing.

Previous studies have examined diffusion parameters in live versus ex vivo fixed tissue. Most studies have concluded that while the absolute values may change, relative values of MR parameters in fixed tissue (eg. ratio fractional anisotropy of gray matter to white matter) are similar to those in vivo^23–26. However, few studies have looked at the variation of DTI parameters at extended time post fixation. Most reports suggest scalar DTI parameters (fractional anisotropy, FA; axial diffusivity, AD; and radial diffusivity, RD) remain unchanged over time post fixation^27–30. The impact of post mortem interval (PMI), i.e. the time between death and fixation during which the tissue undergoes autolysis has been studied for the mouse¹⁴. But scan interval (SI), the time post fixation until scanning, has been seen as something of minimal impact³¹. Other studies have investigated the effects of SI on T1 and T2. Most of these studies found that the T2 decreases sharply within the immediate period following fixation – a period found to last a week - before reaching a plateau up to 1-year postmortem^{20–22,30,32}. It could thus be hypothesized that the DTI parameters, which are affected by T1 and T2, should stabilize following an initial period after fixation, once the brain has fully fixed. As emphasized through these dichotomous hypotheses, the previous literature observations with regards to DTI stability over time have been mixed.

While studies have separately addressed the changes in tissue variables such as volume or diffusion metrics over time post fixation, no study has yet to cohesively analyzed all these metrics across a large and consistent set of time points to allow group comparative analysis. Furthermore, no study has investigated the effects of active staining perfusion fixation on parameters that contribute to connectome tracking nor analyzed statistical differences in connectomes across time.

Our current workflow for quantitative connectomics involves active staining of a batch (n=6-24) of animals at a time. Our scan protocol acquires 51 three-dimensional (3D) volumes requiring a total scan time per specimen of ~12 hours. The practical consequence is that there exists a spread in the interval between perfusion and scanning. The goal of this study was to understand the sources of variability in actively stained mouse brain over time to establish a scan interval during which changes from fixation would have limited effect on group comparisons of scalar DTI measures, volumes and connectomes.

METHODS

All animal experimentation was performed in accordance with the National Institutes of Health Guidelines for Animal Care and Use of Laboratory Animals and the protocols of the Duke University Animal Care and Use Committee.

SPECIMEN PREPARATION

Eight male C57BL/6J adult (90 day) mice were acquired (The Jackson Laboratory, Bar Harbor, ME, Cat# 000664) for MR imaging. On day 0 of fixation, the mice were prepared using an active staining technique¹⁷ in which they were perfused through the left ventricle with a solution of 10% ProHance (Bracco Diagnostics, Princeton NJ), and 10% buffered formalin. Following fixation, the heads were removed and placed in 10% buffered formalin for 24 hours to allow fixation to continue. At the end of 24 hours (on post fixation day 1), the brains, still in the cranial vault were removed from the formalin, rinsed and immersed in a 0.5% ProHance/phosphate buffered saline (PBS) solution to restore the T2. The specimen were stored in the ProHance/PBS solution until the time of scan and between each scanning time.

DATA ACQUISITION

All data were acquired on a 9.4-Tesla Agilent Direct Drive MRI system utilizing a Stesjkal-Tanner spin echo sequence³³. Acquisition time was reduced using a compressed sensing algorithm with a compression factor of 8 ³⁴. A 420 x 256 x 256 image matrix was acquired over an 18.9 x 11.52 x 11.52 mm³ field of view with a native isotropic spatial resolution of 45 μm³ using MRI scan parameters (repetition time (TR) = 100 ms; echo time (TE) = 12.7 ms, b-value = 4000 s/mm²). Forty-six diffusion-weighted 3D volumes were acquired with the gradient direction of each of volume uniformly distributed on the unit sphere. Five base line (b0) volumes were acquired. A post processing pipeline registered the five b0 volumes together to generate the average baseline. The diffusion weighted volumes were registered to this baseline to correct for eddy currents¹³.

MR images were gathered at seven time points post-fixation (2-3 days, 1 week, 2 weeks, 3 weeks, 4 weeks, 8 weeks, and nominally 1 year). The scan intervals for each specimen are shown in Figure 1. Two specimens were scanned for the 2-3 day, 2-week, 3-week, 4-week, and 8-week time period. Four specimens were scanned at all time points. Two additional specimens were scanned only at the 1-year time point.

Figure 1:

Scan time after perfusion for the 8 specimens used in this study: (a) all the time points (2-3 days to 1 year post fixation), and (b) detail of the period from 2-3 days to 8 weeks post fixation.

The time periods of comparison can be separated into categories of early (2-3 Day, 1 week), middle (2-8 weeks), and late (~1 year). Late is the last repeated scan which was >6 months since prior testing, and nominally 1 year since fixation. That scan represents an extreme case of retesting within our protocol. The separation of early and middle time points was determined by what was suspected to coincides with enough time to likely provide sufficient T2 stabilization (2 weeks).

Diffusion measurements are sensitive to temperature changes. Since the scans are acquired over extended periods, we have added a dedicated temperature monitoring system that tracks the temperature variation over the entire course of a 12-hour run. The system monitors two PTS 100 thermistors on the inner and outer z gradient coil. A fiber optic probe attached to the inner gradient monitors the bore surface temperature at the sample.

In previous work, we have tracked the temperature over ten days (Figure S1 of Calabrese et al ³⁵). The mean temperature for the diffusion weighted scans is typically 24.7 ± 0.5°C. The gradients are cooled with chilled water so the temperature drops to 21°C for the baseline images. The 3.7°C difference between baseline and gradient weighted images can produce a slight bias on the absolute eigenvalues but has limited impact on the angular measurements.

ATLAS MAPPING TO MR IMAGES

The SAMBA pipeline ³⁶ was used to register the Waxhom Space (WHS) atlas³⁷ with labels onto each specimen. The pipeline uses two steps to create the label mapping: an affine step using the diffusion weighted image (DWI) and diffeomorphic registration utilizing FA image. The WHS atlas includes 166 (symmetric) regions on each hemisphere in an adult C57BL/6J mouse which are mapped onto all the diffusion data in this study³⁷. This provides ready extraction of the mean and standard deviation of the volume and diffusion scalar metrics from all 332 regions of interest in all the scans via the use of diffusion toolkit (the command line form of TrackVis³⁸).TrackVis derives three eigenvalues providing strength of diffusion within a 3D measurement framework on a voxel-by-voxel basis. From the three eigenvalues (${\lambda }_1,{\lambda }_2,{\lambda }_3$), complementary scalar images can be generated: mean diffusivity (MD), the arithmetic average of all eigenvalues ; axial diffusivity (AD) the principal eigenvalue(${\lambda }_1$ ); radial diffusivity (RD), the arithmetic average of eigenvalues along minor axes (${\lambda }_2,{\lambda }_3$); fractional anisotropy (FA) from all three eigenvalues, $\sqrt{\frac{3}{2}(\frac{{\left({\lambda }_1-MD\right)}^2+{\left({\lambda }_2-MD\right)}^2+{\left({\lambda }_3-MD\right)}^2}{{\lambda }_1^2+{\lambda }_2^2{+\lambda }_3^2}}$. SAMBA mapping also provides the nodes used in defining the connectome for each specimen.

SELECTION OF CHARACTERISTIC REGIONS

Twelve regions of interest (ROI) were followed across the entire time span; nine white matter and three gray matter regions. These regions are shown in Figure 2. The white matter regions: the anterior commissure (AC), corpus callosum (CC), cingulum (CG), fimbria (FI), internal capsule (IC), raphe nucleus (RN), cerebral peduncle (CP), middle cerebellar peduncle (MCP), and inferior cerebellar peduncle (ICP), were selected because of their relatively large volumes, to minimize partial volume errors, while still being representative for the whole brain. Some of these regions represent challenges with mapping: proximity to the ventricles (FI) and smaller volume (cerebellar peduncles). The grey matter regions: somatosensory cortex (S1), cingulate cortex (C), and primary visual cortex (V), were selected because these represent bulk cortical regions which also represent a challenge with automated mapping^39,40. The contrast between cortical regions (eg. motor and sensory) is not particularly high in the MR images. In addition, layers within some cortical sections are not well delineated in the MR images. So, we have combined some of these regions to simplify the segmentation.

Figure 2:

The regions analyzed in this study visualized on an example C57BL/6J specimen with (a and c) showing the white matter regions – AC (red), CC (orange), CG (yellow), FI (teal), IC (blue), RN (purple), CP (green), MCP (violet), and ICP (dark blue) – viewed from superior and sagittal axis respectively and (b and d) showing the grey matter regions – S1 (pink), C (dark pink), V (white) – viewed from superior and sagittal axis respectively.

CHARACTERIZATION OF DTI SCALARS AND VOLUME MEASUREMENTS

To minimize partial volume effects at the periphery of the ROIs, ImageJ was used to erode each ROI prior to statistical analysis of the scalar DTI parameter. Figure 3 demonstrates the impact of the erosion on the histogram of FA over three ROI, exemplifying the narrowing and normalization of voxel histograms, and a gradual increase of peak FA with increasing erosion. We chose to apply the erosion command once (erosion level 1 in Figure 3) to the binary masks describing our regions, to balance the removal of edge effects while maintaining an adequate sample of the ROI. This removes one pixel at the boundary of each region.

Figure 3:

Effects of erosion on the histogram of FA for the (a) anterior commissure, (b) corpus callosum, and (c) internal capsule at the 3 weeks after fixation.

The peak values of each histogram were extracted at each time point as a main study metric. The means and standard deviations of these peaks were the basis of statistical comparison. Data were compared using an ANOVA test to determine if there was significant variation over time with the diffusion scalar or volume -- with posthoc pairwise comparison via Tukey’s test to see notable time point variations.

The volume variation of each region was calculated in a similar way as the DTI metrics (using an ANOVA test then posthoc pairwise testing). No erosion was applied to the region sizes and the mean percent volume differences, compared to three-week time point after fixation – the middle of the middle time points, is presented as the key study metric within the accompanying graphs.

TRACTOGRAPHY

DSI studio ⁴¹ version 20181115 was used to generate tractography for the connectomes. We used the GQI (Generalized Q-Sampling Imaging) model which generates a spin distribution function using a model free reconstruction⁴². Tractography was generated using a tracking plan which terminated after generating 2 million fiber tracts with lengths between 0.5 – 200 mm utilizing all fiber directions at a turning angle of 45 degrees with smoothing and step size of 0.01. The GQI algorithm generates a quantitative anisotropy (QA) analogous to FA. The QA can be normalized by its maximum value, generating the normalized QA (NQA) on a range [0 1]. The threshold for tracking is found for each specimen from rank ordering the NQA histogram and finding the NQA value corresponding to 10% of the total number of points in the histogram. The mean NQA threshold value for all the specimens over all timepoints was 0.105 ± 0.011 (mean ± standard deviation). The connectome values were calculated by counting the number of tracts passing through each region. All connections were included, i.e. there was no thresholding.

Similar methods were used in a recent study of connectome differences between four different strains of mice⁹ to determine the across strain brain size differences within the tract generation. In that analysis, approximately five million tracts were generated for each specimen in each strain. The number of tracts were normalized to the brain volume. In that study, there was little difference in connectomes generated with two million tracts and five million tracts for the same specimen. Thus, in this current study, we standardized on two million tracts normalized to a reference C57BL/6J brain volume (435 mm³) to account for potential changes in brain size across specimens. This scalar factor is applied to each connectome matrix as a volume accounting method of brain volume differences with an average value across this study of 2.60 +/− 0.07. The average brain volume of this data set is 453 +/− 12.0 mm³.

The degree of connection - a measurement of the number of connections from one brain region to another - was extracted for all the regions in the atlas ⁴³. Several cortical areas were combined leaving a total number of possible connections for each ROI at 296. The degree measurement was normalized to this value so that degree of every region ranged between [0 1]. The variation in degree for the 12 example regions is plotted as a function of time after perfusion in the same manner as was used for the scalar DTI metrics.

OMNI-MANOVA

Connectome differences were compared using a graph theory process called OMNI-MANOVA⁴⁴. The connectome is a matrix that expresses the strength of connectivity between each node (i.e. ROI) of the brain. A profile of a specific node, eg. CC, can be constructed based on all the tracts passing through that node. Constructing these profiles allows all the nodes to be expressed as a smaller vector in a common space reducing the complex connectome (332x332 for each specimen) to a more manageable 332x ~3-5 for each specimen. To first create the omni-embedding, a block matrix is created from each corresponding connectome and average responses between connectomes. That block matrix is then decomposed using singular value decomposition and reconstructed into a 3D matrix of dimensions (brain regions by number of specimen by vector elements) describing the node within the newly defined whole study space. The total number of elements in the reduced vector describing the connectivity is based on selecting the average number of principle eigenvalues via the elbow of all scree plots generated from subjects⁴⁵. Those elbows are selected by utilizing Zhu and Ghodsi automatic detection method⁴⁶.

This term reduction allows the use of multi-variate statistics like MANOVA. MANOVA helps determine, in a dataset where the primary mechanism is time since fixation and the tractography differences if completely preserved and properly rested would hypothetically be null, variation in connectivity amongst the time points. We wish to see which nodes the connectome differs significantly from that case. The MANOVA test is a multi-variate form of the ANOVA (a one-way analysis of variance) applied to each node profile described above. We execute the MANOVA test to determine the discrepancies of nodes across all times points and then, by looking at the regions that are significantly different, investigating if that region is known to be problematic, i.e could produce varying tractography due to its structure and features to attempt to separate normal automatic mapping variation from actual structural differences uncovered by the study. The p-values corresponding to each region of interest are corrected for multiple comparisons using false discovery (FDR), Benjamini-Hochberg correction⁴⁷. While this approach can provide more false positive results than bonferroni correction, this concern is mitigated by our desire for any positive result in our test/retest criteria rather than the strictest interpretation of a positive connection result.

A classical multidimensional scaling was formed in 2-D based on the distances of the Omni-MANOVA embedding tensor (the reconstructed 3D matrix) by a frobenius norm^48,49 This visualizes how each specimen and group relate to the entire set of study animals. The classical multidimensional scaling is similar to principle component analysis⁵⁰. Rather than focusing on illustrating maximum variation in the set (principal component analysis), the multidimensional scaling illustrates the similarity of the scans. This embedding generates a simplified view, removing region-based analysis of the embedding utilized in the OMNI-MANOVA, so we can make assumptions of the general relationship of our set and compare the spread of groups scanned at different times within the same embedding space.

DIFFERENTIAL TRACTOGRAPHY

After the initial investigation using the DTI scalar metrics and graph theory connectomic analysis, we investigated tract differences across the middle time points (2 weeks versus 8 weeks) and the ending time point (8 weeks versus 1 year time point) of normal and extreme retesting respectively using differential tractography in DSI Studio⁵¹. The method was developed for comparing scans of the same patient at different time points by mapping the anisotropy into a common space and tracking the difference in anisotropy along all the tracks. By looking at the difference in anisotropy, one can determine the stability of the tracking across those two periods. Differential tractography was performed on 4 different specimens – creating a temporal comparison of these individual specimen for two analyzes: 2 versus 8 week, for testing how much tracking varies across the self-consistent period, and 8 week versus 1 year, indicating how much tracking varies from the self-consistent period to the late time point.

The initial tracts for differential tractography were generated using DSI Studio version 20200627 to create a series of 50,000 seeds with the same parameters as the initial tracking. We used QA instead NQA, because the percent change criteria in differential tractography is based on QA. The QA thresholding was determined at 10% of the actual QA values. For comparisons between time points, we looked at the changes in QA with four different thresholds, ±20% and ±50%, which had been used previously in a clinical study of demyelination and axonal loss⁵².

RESULTS

A representative dataset for this work is available on VoxPort (https://civmvoxport.vm.duke.edu/voxbase/studyhome.php?studyid=746). To access the database, a user will need to register an account. Additional data used in this study can be obtained via contact with corresponding author (gjohnson@duke.edu).

FRACTIONAL ANISOTROPY

The mean peak FA and standard deviation across the specimens at each time point are shown in Figure 4. The mean peak FA of the gray matter regions, Figure 4j–4l with range 0.04-0.24 (−), was substantially lower than the white matter regions, Figure 4a–4i with range 0.23-0.83 (−). The mean peak FA of all ROIs broadly follows an exponential decay, with the most dramatic change from 2-3 days to 2 weeks, decreasing to a plateau from 2 to 8 weeks. For most regions, the FA is relatively unchanged up to a year after fixation.

Figure 4:

Mean and standard deviation of peak FA across all time points for white matter regions: (a) AC, (b) CC, (c) CG, (d) FI, (e) IC, (f) RN, (g) CP, (h) MCP, and (i) ICP; and gray matter regions: (j) S1, (k) C, and (l) V. Red asterisks denote the time points for a region that differs significantly (p<0.05) in peak FA from at least 2 other time points in the self-consistent middle period (2 to 8 week post fixation).

An ANOVA test revealed all ROIs changed significantly (p < 0.05) over all time points. Posthoc pairwise Tukey testing (with p < 0.05) determined the intervals of lowest FA variability. Supplemental Table 1 shows the p-value results of all pairwise comparisons. All twelve regions showed no significant change in the peak FA from 2 to 8 weeks after fixation. All of the pairwise comparisons in the middle time (e.g. 2-3 week, 2-4 week, 2-8 week, 3-4 week etc) are consistent with no significant differences. To compare the consistency between the early and late time periods to the middle time period, we investigate similar pairwise comparisons. For simplicity, we indicate the regions in the early and late time period for which FA varied significantly with at least two time point comparison in the middle time set (50+% of the middle time points, the majority of time points in the middle time period). The time points which had significant differences in the pairwise comparison to at least two points in the middle time zone (2 to 8 weeks) are marked in Figure 4 by red asterisks for each region. Nine regions (S1, RN, V, IC, MCP, FI, CC, C, AC) at the 2-3 day time point varied in this way, while the number of variable regions is much reduced at the 1 week time point (1 region: CG). At 1 year, five regions (S1, RN, MCP, CG, C) differed from at least two time points in the middle time region.

To understand how these regions differ further across the early and late time periods, we compared the FA at the early and late time points to the average values of the 2-8 week period for each of the twelve regions of interest, expressed as a percent difference. At the 2-3 day point, the maximum change is 40% . Regions which have at least 2 pairwise time comparisons that were significant were S1: 34%, RN: 18%, V: 40%, IC: 12%, MCP: 9%, FI: 15%, CC: 28%, C: 38%, and AC: 11%. In the regions in which the changes were not significant (3 of 12 regions), the average change in FA was 11.5%. Across all twelve regions, the average change in 2-3 day FA data relative to the average 2-8 week response is 20%.

For the 1 week time point there was a 14% change in FA for CG which was the only region which differed in at least 2 of the pairwise comparisons to the 2 to 8 week measurements. The remaining 11 regions did not show significant changes (average change across all 11 was 5.4%).

In pairwise comparisons of the FA at 1 year to the 2-8 week period there were five regions in which there were significant differences in at least two time points: S1: −18%, RN: −14%, MCP: −5%, CG: −17%, and C: −20%. The remaining 7 regions did not show significant changes (average change across all 7 was −4.9%). One can get a sense of the stability of the tissue over time by comparing the changes across the 2-8 week period to the changes that take place at a year. In a comparison across the 2-8 week period, the average change in FA from the mean 2-8 week response across all twelve regions ranges ranged from 3.4% (2 week versus average 2-8 week response) to −4.9% (8 week versus average 2-8 week response). The average FA change at 1 year across all twelve regions was −8.9% which is approximately equal in magnitude to the range of FA in the 2-8 week period.

VOLUME

We investigated the volume changes across the 12 selected regions by normalizing to 3 weeks after fixation, the median time point in our sampling set. Figure 5 shows the average and standard deviation of the volume change across all time points. An ANOVA test determined that eight of the twelve regions had statistically significant change in volume across time (p < 0.05; S1, RN, V, IC, MCP, ICP, C, CP). The four regions without significant volume differences over time (FI, CC, CG, AC) were white matter regions.

Figure 5:

Mean and standard deviations of volume change compared to the 3-week time point for white matter regions: (a) AC, (b) CC, (c) CG, (d) FI, (e) IC, (f) RN, (g) CP, (h) MCP, (i) ICP; and gray matter regions: (j) S1, (k) C, (l) V. Asterisks denote the time points where a region differs in volume (mm³) from at least 2 other time points in the consistent middle period 2 to 8 weeks post fixation.

Post-hoc testing was performed on the eight regions that changed significantly (p < 0.05) in volume. Supplemental Table 2 shows the p-value results of all pairwise comparisons. Similar to the results of FA, there were no significant changes in volume between 2 and 8 weeks after perfusion, the middle versus middle time point comparisons.

We see from the pairwise comparisons that volumes at the 1 year time point varied significantly from more than two points in the self-consistent middle time period for five of the eight regions (S1, RN, V, IC, CP). While the number of regions with volumes that were different at 1 year is large, the percent difference of volume is small, with a mean change amongst those five regions of 4.4% (specific regional values -- S1: 3.1%, RN: 5.2%, V: 4.8%, IC: 5.0%, CP: 3.7%) and a mean volume change across all measured regions of 2.9%. Unlike the FA, only the volumes of two of the eight regions (MCP, ICP) differed significantly at the 2-3 day time point in comparison to at least 2 of the middle time points. For those two identified regions at the 2-3 day time point mean volume change is 7.0% (specific regional values -- MCP: 8.1%, ICP: 5.8%). No volume of any region at 1 week after fixation differed significantly using that same metric. The mean change for all twelve regions at 2-3 day it is 2.5%, while for 1 week is 2.0%. Like FA, as the one moves from the early time period towards the middle time points, the average change decreases. In Figure 5, time points with significant pairwise changes from at least two points in the middle time zone are marked by red asterisks for each region.

SUPPLEMENTAL SCALAR METRIC ANALYSIS

Four additional scalar metrics (mean diffusivity (MD), radial diffusivity (RD), axial diffusivity (AD), and normalized degree) were analyzed for this work in the same way as FA and volume. The resulting figures (Supplemental Figures 1–4), pairwise tables (Supplemental Tables 3–6), and corresponding analysis are included within the supplement. The results for those additional metrics follow that of the FA and volume.

OMNI-MANOVA

The Omni-MANOVA analysis in Figure 6 provides an expanded analysis of the stability of the connectome by showing the reduced vector comparisons at multiple time points. In Figure 6a, in which all the time points are compared, there is a marked separation of the 2-3 day and 1 week data (red and green) from the rest of the groups (blue, cyan, magenta, yellow and black). This indicates the 2, 3, 4, 8 week, and 1 year measurements are more similar to one another than the early time points. The rank ordered p-value results of the MANOVA test, identified 160 of the 332 regions with p_BH<0.05 thus indicating that vectors describing connections through nearly half of our nodes have changed significantly amongst all the time points. In Figure 6b, the 2-3 day and 1 week time points have been removed. The 1-year time point does not overlap the middle time points as closely as in Figure 6a but the difference is less than one standard deviation of the constituent groups and the total reconstruction covers less area showing that the data is more similar than all the time groups together. The MANOVA test indicates that profiles (vectors) of 4 out of the 332 regions of interest are significantly changed amongst the time points (p_BH<0.05), Table 1. All are cortical and none are bilateral. Boundaries of these regions are poorly defined in our atlas so the use of FA as an automated mapping criterion results in higher variability in mapping. Suspecting this variation is due to the boundaries within the embedding, 2 weeks and 1 year, we continue by trying two different embeddings each removing one of these boundary time points. In Figure 6c, the 2-week data has been removed, the plot shows 3 weeks to 1 year. There is little change from Figure 6b and no significant changes in connectivity. Figure 6d plots the data from 2-8 weeks. Note the spread (standard deviation) in this plot is small and not that different than Figure 6c. In addition, there were no profiles which had changed significantly. This suggests that the 1-year data is not drastically different from that gathered during the middle time points.

Figure 6:

Reconstructions of the adjacency spectral embedding created using Omni-MANOVA framework resulting from (a) all time points, (b) 2, 3, 4, 8 week, and 1 year time points, (c) 3, 4, 8 week, and 1 year time points, and (d) 2, 3, 4, and 8 week time points. The dotted lines are centered on the mean of individual points in the specific time periods with the spread corresponding to the standard deviation in embedding dimension 1 and 2. Individual points in the plots represent each specimen in this study.

Table 1:

The regions associated with 2, 3, 4, 8 week and 1 year embedding that varied significantly amongst the selected time points according to BH correction, sorted by raw p-value.

Region Abbreviation	Region Name	Hemisphere
V1B	primary visual cortex, Binocular Area	Right
DLEnt	dorsolateral entorhinal cortex	Right
A24aPrime	cingulate cortex, area 24a prime	Right
S1FL	primary somatosensory cortex, forelimb region	Left

DIFFERENTIAL TRACTOGRAPHY

Differential tractography was employed as a final test of tissue stability. We compared four specimens at 2, 8 week, and 1 year time points, Figure 7. The change in QA at 2 weeks and 1 year utilized the 8 week scans as the base point. Using 50,000 seeds with whole brain tracking, the average track number and length for the groups were nominally the same, 21482 +/− 240 and 5.03 +/− 0.111 mm respectively, at the three time points, indicating that there is not a drastic change in the diffusivity on which the tractography is based. The average numbers associated with each time point are shown in Supplemental Table 7.

Figure 7:

A comparison of (a) differential tractography measuring the change of QA at 2 weeks and 1 year using 8 weeks as a basis and tracking parsed to only those that pass through the CC at (b) 2 weeks (c) 8 week, and (d) 1 year post fixation. All images are produced with specimen 180423-2:1 which had the largest difference in track length and number at ±20% and ±50% in the differential QA data.

When QA change is used as the thresholding condition, the mean tract length was approximately a fifth of the tract length in the QA images. The mean tract number and mean tract length at ±20% and ±50% thresholds are shown in Supplemental Table 8. The differential tracts are small, scattered, and likely are related to edge effects. Differences between the 2 and 8 week scans are negligible. The differences between the 8 week and 1 year scan are most notable at −20% suggesting some systematic though small reduction in FA as the tissue rests in PBS for a year. One can relate the differential tracts to the global length and density of tracts comparison to all tracts passing through the left portion of the CC for the same specimen (180423-2:1) at the three time points from the 50,000 seeded whole brain track data. We picked 180423-2:1 because in the whole brain tracking, the response of number of tracts and track length was most close to the average result. Example images of this tractography at all three time points are shown in Figure 7b–7d. Both Figure 7b and Figure 7d used Q-Space Diffeomorphic Reconstruction⁵³ (QSDR) to place the 2 week and 1 year data into the same orientation alignment as the 8 week data (Figure 7c), and thus needed to use the 8 week label for segmentation of CC tracts. We see more grouping of key tracts, such as the arching structure of the CC and projections into the olfactory bulb, than illustrated by the differential tractography. As a comparison, the average number of tracks shown in Figure 7b–7d is 2520 +/− 60 which is approximately the same number of average tracks found for all specimen at −20% threshold comparison of 1 year and 8 weeks. Interestingly, while the number of tracts was the same, the average tract length of Figure 7b–7d is 11.5 +/− 0.796 mm which is an order of magnitude longer than the multi-specimen averaged −20% threshold comparison of 1 year and 8 weeks. Thus, this confirms that our remaining differences in tractography shown in the differential tractography are negligible overall.

DISCUSSION

Through our analysis of fractional anisotropy, axial diffusivity, radial diffusivity, and mean diffusivity, we have sought to understand the changes in diffusion tensor scalars due to fixation over time. Similar to the literature results for T1 and T2 changes post-fixation ^{21 22,54 20}, we found that some of the DTI scalar values varied significantly during the first two weeks following perfusion fixation, but eventually stabilized to a nominally constant value. Changes in the diffusion metrics were statistically insignificant between 2 and 8 weeks. Specimens scanned during this window can be compared without confound from fixation. As noted in the introduction of this work, the changes of T2 between 1 day and 1 week (168 hours) are most likely to induce the variability seen in the first 1-2 weeks in our data. Allowing tissue fixation to stabilize for 2 weeks avoids reduced SNR and reduces T2 changes across study time which likely contributes to the stability we see in our analysis post 2 weeks from fixation. As the scalar DTI metrics in multiple gray and white matter regions changed significantly over the first two weeks of fixation we suggest that standard operating procedure should include a stabilization period of at least 2 weeks prior to scanning.

For additional confirmation of variation within the 2-8 week self-consistency period, the standard deviation within group to across group (Supplemental Table 9) was calculated, where a group in this sense is the data belonging to one specimen itself across the 2-8 time period. For the majority of ROI in the scalar contrasts, the variation found within group was lower than that across group. Only degree and RD had select regions with more variation across group, indicated by bolded entries in Supplemental Table 9 (degree: Anterior Commissure, Corpus Callosum, and Fimbria; RD: Fimbria, Inferior Cerebellar Peduncle, and Primary Visual Cortex). This indicates that, although the variation naturally for this specimen type is low due to the inbred nature of the mouse strain, we still can determine that more variation in the 2-8 week time period is attributable to across group differences than within group differences.

There are statistically detectable differences in the scalar metrics between the 2 to 8 week period and 1 year. But the DTI parameters at 1-year post fixation have changed only marginally from the measurements during the middle self-consistent time points. The scalar metric with the largest change across all structures measured at 1 year is FA at −8.9%. This metric is primarily influenced by regions with average FA below 0.35, which includes cortical regions. Excluding these lower value mean FA cortical regions (CG,RN,C,V,S1), the average change drops to −4.7%. This suggests that specimens may be reexamined for a longer period without serious confound to the study.

Volume of the 12 structures was statistically unchanged during the 2 to 8 week time period. Volume change out to 1 year post fixation was ~ 3% over all the sampled regions. De Guzman et al have performed a nicely detailed study of volume changes using protocols similar to ours (i.e. perfusion fixation, 1-5 day formalin immersion, long term storage in saline or water) in a mouse model⁵⁵. The period of observation (to 150 days) is comparable to ours. They found a rapid change in volume (3.5%/day) when tissue is stored in formalin and ~3%/month when tissue is stored in PBS. Our data is consistent with theirs, though the time points of comparison are not identical. De Guzman showed volume changes at 5 months of white matter ranging between ±4% and grey matter between −4% and +5%. For our data, the maximum volume change in any volume at 1 year is 5.2% and an average change of 2.9% over the twelve regions at 1 year. While small, these regional changes are still larger than the whole brain volume change in our data, which increases by 1.5% from 2 weeks to 1 year (average, masked via labels, brain volume of 471 ± 8.06 mm³ at 2 week as compared to 478 ± 12.9 mm³ at 1 year). But it bears repeating that during the 2-8 week period there were no statistically discernable changes in volume in the 12 structures surveyed. In a recent publication, we measured the coefficient of variation between left and right hemispheric structures in four different strains of mice as a measure of technical error ⁹. That coefficient of variation was <5% for structures with volumes > 1 mm³. The variation seen in that study is probably an indication of the reproducibility of the registration pipeline. Storage in buffered saline resulted in minimal tissue swelling, even after 1 year.

The Omni-MANOVA results are consistent with changes in the scalar metrics during the first two weeks and self-consistency of the middle time points. The points associated with the early (2/3 day and 1 week) time period stand out as particularly different than the other groups. Figure 6a maps all the data into a common space. In this mapping, there are 160 of 322 regions with connectivity that differs. In Figure 6b, the 2/3 day and 1 week data have been removed and there are only 4 regions (all cortical) that differ with p values that are significant (p_BH<0.05). The connectomes are remarkably stable from 2 weeks out to one year. Similar to the conclusion derived from the parametric analysis of the scalar DTI metrics, the 1-year connectome data is similar enough to the stable period that the 1 year time point can be considered in practice invariant from the middle time points.

The differential tractography results also confirms the stability of the tissue over the 2-8 week period with minor changes between 8 weeks and 1 year. There is a suggestion of increased anisotropy (+20%) particularly in the cerebellum in comparison to the 2 and 8 week data, but the mean track length and number (0.91 mm, 570 tracks) in the difference image vs mean track length and number (5 mm, 21516 tracks) in the base images suggest this is probably an edge effect. The mean track length in the 8 week and 1 year difference image is also small (0.97mm @ +20%) suggesting an edge effect. But there is an increase in number and mean length at −20% from 0.82 mm with 111 tracks @ −20% to 1.18 mm with 2490 tracks when comparing 8 weeks and 1 year. This is consistent with the bulk scalar metrics.

This work focuses on ex vivo measures. Comparison of in vivo and ex vivo measurements of brain volumes have been performed previously by Ma et al⁵⁶. Subsequent work by Lerch et al explored the statistical reproducibility of brain volume measures in in vivo and ex vivo measurements⁵⁷. Lerch concludes with regards to the hippocampus, “there is a clear advantage to imaging in-vivo where population variance predominates, because each mouse can be fit with an independent intercept,” although “this advantage is largely offset if relative volumes can be used instead of absolute volumes, as this significantly reduces the population variance.” Since connectomic studies performed in vivo could not be achieved with the precision we obtain in ex vivo studies, we focused this work exclusively on sources of variation in these ex vivo measurements.

Numerous researchers have studied the effects of autolysis, fixation, length and type of fixative and storage (in fixative, water, saline) on the diffusion properties of tissue ^{14,19,24,25,27,30,58,59}. For high resolution MR histology, good scan quality is vital. Fixation is essential to limit autolysis. Perfusion fixation is considerably more effective than immersion. Extended fixation (>24 hours) causes continuing reduction in T2 (and therefore reduced signal to noise). Storage in buffered saline is desirable to minimize tissue swelling/changing. And active staining ^16,60, i.e. perfusion with a contrast agent is essential for microscopic imaging. However, few studies have looked at the variation of DTI parameters at extended time post fixation.

Most reports (cited studies are of human, cat, and pig) suggest scalar DTI parameters (fractional anisotropy, FA; axial diffusivity, AD; and radial diffusivity, RD) remain unchanged over time post fixation^27–30. These studies have several marked difference from our study protocol. We are scanning the mouse brain^{9,10,12–14,16,17,24–26}, rather than a human^{1,3–6,8,20,22,26,28,30–32}, cow¹⁵, rat ^19,21, marmoset²³, cat²⁹, or pig²⁷. Human DTI differs greatly from that of the mouse. Neither DTI human study^28,30 could utilize perfusion fixation which can easily be performed on a mouse. The pig²⁷ study used a brain segmentation atlas with much fewer segmentations and did not utilize a gadolinium contrast agent to enhance signal contrast. The cat study²⁹ was performed on spinal cords and used ADC to determine the diffusion coefficients of the white matter tissue. This study provides analysis across several diffusion metrics. The resolution of this study is much higher (45 um³ i.e 91 pl) than the human study^28,30 and pig²⁷ prior DTI studies which used clinical resolution. The spatial resolution of DTI in clinical MRI is generally >1 mm³, i.e. a difference of nearly 11,000 X in voxel volume. The studies performed here provide an increased confidence in our methods.

CONCLUSION

In this work, we study the repeatability of mouse specimen diffusion connectomic data by measuring a set of actively perfused specimen at 2-3 day, 1 week, 2 week, 3 week, 4 week, 8 week, and 1 year time points post fixation. We determine that the early time points, prior to 2 weeks, have an undesirable amount of variability in the resulting connectome parameters, both scalar metrics and tractography results. After two weeks and leading to 1 year with minimal changes indicating there is high self-consistency of the data produced across various DTI metrics and tractography analysis.

Connectomics have been put into clinical and basic science pipelines; generating remarkable insight into both structure and health of the brain. However, there has been caution towards the results of connectomic analyses. Maier-Hein et al raised significant concerns over false positive connections even with optimized clinical protocols and pipelines ⁶¹. Comparison of clinical protocols with methods for the mouse brain is fraught. Retroviral tracers continue to be the gold standard for quantitative differentiation of afferent and efferent connections in the mouse brain ⁶².Yet diffusion derived connectomes at microscopic resolution provide intriguing applications that cannot be done with tracers. ^9,13,63 The protocols must be optimized and the limits understood. We have taken the protocol that we have been refining via comparison with retroviral tracers ³⁵ and generation for routine use ³⁴, now to create a much larger scale comparison for determining self-consistency of the utilization across time, applying confidence to the connectome answers for large scale experiments.

ABBREVIATIONS AND UNITS

SI

scan interval

PMI

post mortem interval

DWI

diffusion weighted imaging with a metric of diffusion weighted intensity [unitless]

WHS

Waxholm Space, label mapping system utilized in this work³⁷

PBS

phosphate buffered saline

ROI

Selected Regions of Interest

AC

anterior commissure

CC

corpus callosum

CG

cingulum

CP

cerebral peduncle

FI

fimbria

IC

internal capsule

ICP

inferior cerebral peduncle

MCP

middle cerebral peduncle

RN

raphe nucleus

C

cingulate cortex

S1

primary somatosensory cortex

V

primary visual cortex

DTI

(Diffusion Tensor Imaging) Measure Abbreviations

FA

fractional anisotropy [unitless]

AD

axial diffusivity [mm³/s]

RD

radial diffusivity [mm³/s]

MD

mean diffusivity [mm³/s]

GQI

(Generalized Q-Sampling Imaging) Measure Abbreviations

QA

quantitative anisotropy [arbitrary, scales with spin density function]

NQA

normalized quantitative anisotropy [unitless]