Neurochemical and cognitive changes precede structural abnormalities in the TgF344-AD rat model

Abstract Alzheimer’s disease is a progressive neurodegenerative disorder with a decades-long pre-symptomatic phase, substantiating the need for prodromal biomarker development and early intervention. To deconstruct the processes underlying disease progression and identify potential biomarkers, we used neuroimaging techniques with high translational potential to human clinical studies in the TgF344-AD rat model which recapitulates the full spectrum of Alzheimer’s neuropathology (progressive amyloid deposition, tauopathy, frank neuronal loss, gliosis, and cognitive dysfunction). We employed longitudinal MRI and magnetic resonance spectroscopy in conjunction with behavioural testing to characterize multiple facets of disease pathology in male and female TgF344-AD rats (n = 26, 14M/12F) relative to wildtype littermates (n = 24, 12M/12F). Testing was performed at 4, 10, 16, and 18 months, covering much of the adult rat lifespan and multiple stages of disease progression. The TgF344-AD model demonstrated impaired spatial reference memory in the Barnes Maze by 4 months of age, followed by neurochemical abnormalities in the hippocampus by 10 months and major structural changes by 16 months. Specifically, TgF344-AD rats displayed increased total choline and lactate, and decreased total creatine, taurine, and N-acetylaspartate to myo-inositol ratio, dentate gyrus hypertrophy, and atrophy in the hippocampus, hypothalamus, and nucleus accumbens. Overall, these findings support the use of MRI and magnetic resonance spectroscopy for the development of non-invasive biomarkers of disease progression, clarify the timing of pathological feature presentation in this model, and contribute to the validation of the TgF344-AD rat as a highly relevant model for pre-clinical Alzheimer’s disease research.


Graphical Abstract Introduction
Alzheimer's disease is a progressive neurodegenerative disorder that accounts for 60-80% of the 50 million dementia cases worldwide. 1 Aspects of Alzheimer's pathology can occur decades before clinical onset, 2-4 substantiating the need for prodromal biomarker development and early intervention.
Early-stage disease characterization in transgenic animal models represents one promising avenue towards the development of new biomarkers and intervention approaches at a clinical level. The most common transgenic models are rodents expressing human genes harbouring mutations known to drive amyloid-β accumulation and cause familial or early-onset Alzheimer's disease in humans, such as presenilin-1 (PS1), presenilin-2 (PS2), and amyloid precursor protein (APP). [5][6][7] Amyloid-β-overproducing rodents are considered 'gold standard' models but most do not display robust tauopathy or neuronal loss-two major hallmarks of Alzheimer's disease-unless additional human transgenes are expressed that are not associated with familial Alzheimer's disease (typically microtubule-associated protein tau, MAPT 7,8 ), as is the case with the widely used 3xTg mouse. 9 To date, one of the only rodent models to recapitulate the full spectrum of Alzheimer's disease neuropathology without insertion of a human tau transgene is the TgF344-AD rat which displays progressive amyloid-β deposition, tauopathy, gliosis, neuronal loss, and cognitive impairment, despite only expressing mutant human APP (APPswe, KM670/671NL) and PS1 (ΔE9) genes. 10 Not only does the TgF344-AD model represents a major advancement for Alzheimer's disease research, but the numerous advantages of studying rats over mice-they are physiologically and genetically closer to humans, display a richer behavioural phenotype, and have larger brains 11 -make the TgF344-AD rat a particularly salient option for pre-clinical biomarker development.
Magnetic resonance (MR) techniques such as MRI and MR spectroscopy (MRS) enable non-invasive, longitudinal assays of brain structure and tissue chemistry at the preclinical and clinical level. [12][13][14] MRI studies have identified reduced cortical thickness and atrophy of the medial temporal lobe as prominent features of Alzheimer's disease in human subjects 13,15 that may precede clinical diagnosis. 16,17 Similar studies in transgenic models have replicated some of these findings. [18][19][20][21][22] Proton MRS studies in rodent models of Alzheimer's disease have identified altered metabolic profiles in the hippocampus, cingulate cortex, and frontal cortex, including reduced N-acetylaspartate (NAA) and glutamate (Glu), and increased myo-inositol (Ins), glutamine (Gln), and total choline (tCho). [23][24][25][26] These changes parallel those observed in similarly affected brain regions in human Alzheimer's patients such as the posterior cingulate and occipital cortices, 27,28 while also informing on physiological processes involved in disease pathogenesis, including neuronal viability, cell membrane turnover, antioxidant capacity, neuroinflammation, energy metabolism, and neurotransmission. 29,30 Despite the relevance of MRI and MRS in Alzheimer's disease research, few studies have simultaneously examined the longitudinal progression of anatomical, neurochemical, and cognitive changes in either rodent models or humans. As such, a gap in knowledge exists regarding the relative timing of the appearance of these pathological features, limiting the understanding of disease stages and subsequent design of therapeutic approaches.
The aim of this study was to characterize the manifestation and time course of pathological change in neuroimaging biomarkers and cognition in the TgF344-AD rat in vivo to determine whether it recapitulates major features of human-AD and to contribute to better disease staging of this model. We employed MRI, MRS, and Barnes Maze testing at 4, 10, 16, and 18 months of age to distinguish longitudinal changes in neuroanatomy, neurochemistry, and cognitive function in male and female TgF344-AD rats relative to wildtype (WT) littermates.

Animal care and study design
The TgF344-AD model [transgenic (Tg)] is a double Tg line created on a Fischer 344 background that expresses the 'Swedish' mutant human APP (APP swe : APP KM670/ 671NL) and deletion of exon 9 mutant of human PS1 (PS1ΔE9). Male hemizygous TgF344-AD rats (Terrence Town Laboratory, University Southern California, CA, USA) and female homozygous Fischer 344/NHsd WT rats (Envigo, Madison, WI, USA) were bred in-house. Offsprings were a mixture of hemizygous Tg and homozygous WT rats. Tail snips were obtained from each rat to identify the presence of the APP swe and PS1ΔE9 transgenes (genotyping by Transnetyx, Memphis, TN, USA). Rats were weaned on post-natal Day 21 and housed in same-sex pairs on a 12 h light-dark cycle with ad libitum access to food (Envigo, Teklad Global 18% Protein Rodent Diet) and water. All animal procedures and experiments were performed in accordance with the guidelines of the local institutional Animal Care Committee.
In total, 24 WT rats (12M/12F) and 26 Tg rats (14M/12F) were studied longitudinally, with Barnes Maze testing and neuroimaging performed at 4, 10, 16, and 18 months of age. Behavioural testing was performed prior to neuroimaging to avoid confounds of anaesthesia on behaviour. Sample size calculations for comparison between genotypes while controlling for sex were performed using a population simulation-based power analysis tool 31 and can be found in Supplementary Methods. We also present exploratory analyses of the intersecting effect of genotype and sex on all neuroimaging and behavioural markers. Group sizes at each time point are included in Supplementary Table 1. MRI data acquisition and regional volume estimation MRI data were acquired using a 7 Tesla Bruker Biospec 70/ 30 scanner (Bruker, Billerica, MA, USA) with an 86 mm (diameter) volumetric birdcage coil for transmission and a four-channel surface array coil for signal reception (Bruker). The level of anaesthesia (1-4% isoflurane in oxygen gas) was adjusted to maintain a breathing rate between 50 and 75 breaths/min throughout the procedure and warm air (37°C) was blown into the bore of the scanner to maintain a constant body temperature (SA Instruments Inc., Monitoring System, Stony Brook, NY, USA).
High-resolution 3D anatomical MR images were acquired using Rapid Acquisition with Relaxation Enhancement using scan parameters identical to those described previously. 32,33 Scan resolution was 114 µm isotropic and images were T1-weighted. All pre-processing methodology is described in detail elsewhere 32

and in Supplementary
Methods. After pre-processing, images were examined for motion artefacts, Gibbs ringing artefacts, and other image anomalies, following which 15 of a total 179 scans were excluded from further analysis. Seven rats [four Tg (1M/3F) and three WT (2M/1F)] were excluded at 4 months, one at 10 months [1 Tg (M)], and five at 16 months [three Tg (1M/2F), two WT (1M/1F)]. The remaining 164 scans were co-registered using the two-level deformation-based morphometry pipeline in Pydpiper, as described by Friedel et al. 34 and in Supplementary Methods. This process creates deformation fields for each subject at each time point, reflecting the amount of expansion or compression required to deform each individual anatomical image to the subject average. 35 Deformation fields are then resampled into the common study space allowing comparison between subjects. The Fischer 344 rat atlas was used to estimate the volume of 120 regions. 36 Proton MRS data acquisition and quantification Immediately following MRI data acquisition, MRS data acquisition was performed using the same methodology as described previously. 37 Automated localized shimming was performed using the FASTMAP method 38 (ParaVision 5.1, Bruker). Proton MRS scans were acquired from a 2.5 mm × 3.5 mm × 3.5 mm voxel in the dorsal hippocampus using a Point RESsolved Spectroscopy sequence (acquisition time = 13m0s0ms, TR = 3000 ms, TE = 11.12 ms, 2048 acquisition data points, spectral width = 4006 Hz) in combination with outer volume suppression. In total, 256 averages were acquired with VAPOR water suppression 39 and 8 averages were acquired without water suppression for eddy current correction and as a reference for absolute metabolite quantification.
Spectral pre-processing was performed in the FID-A toolbox (github.com/CIC-methods/FID-A, version 1.0 40 ) in MATLAB (R2012a, The MathWorks Inc., Natick, MA, USA), and consisted of removal of motion-corrupted scans and spectral registration to correct frequency and phase drift errors. Processed spectra were analysed using LCModel (version 6.3, Stephen Provencher Inc., Oakville, Ontario, Canada) 41 , with a neurochemical basis set consisting of 18 simulated metabolite resonances and 9 macromolecule basis functions. Methods detailing the acquisition of macromolecule spectra for parameterization and inclusion into the quantification basis set are described elsewhere. 37 Absolute quantification was performed using the unsuppressed water signal as a reference. A correction was applied to account for T1 and T2 relaxation constants of water and measured neurochemicals, and an assumed NMR-visible water concentration of 4300 mM given that our voxel contained mostly grey matter (GM). 42 For details on the correction formula, see the supplementary material in our previous publication. 37 Neurochemical concentrations are reported in mmol/L (mM). Details regarding the basis set and quality control methods are included in Supplementary Methods.

Behavioural phenotyping via the Barnes Maze test
We assessed hippocampus-dependent spatial reference memory using a shortened variation 43 of the popular Barnes Maze protocol. 44 Detailed methodology is described in Supplementary Methods. Briefly, a circular maze with 20 holes was used, and rats were trained to locate a single escape hole that led to a box underneath. Rats were given three 3-min trials on Day 1 and two 3-min trials on Day 2, for a total of five training trials. A probe trial was used to test longterm spatial reference memory. The probe trial was conducted 48 h after the last training trial and involved blocking the escape hole so that no escape was possible. All sessions were recorded using a Logitech QuickCam Pro 9000. The following metrics were measured during the probe trial using EthoVision XT Software (Noldus Information Technology, Wageningen, The Netherlands): % time in target quadrant, % time in target holes, success or failure to locate the escape hole, average speed (cm/s), and number of holes searched.

Statistical analysis
Statistical analyses and visualizations were performed in R (version 3.6.3 45 ). Brain volume and metabolite concentration data were modelled using linear mixed-effects models as they appropriately model the covariance structure resulting from repeated measurements in the same subjects and handle data with missing values. 46 Brain volumes were predicted by a quadratic age-by-genotype interaction (Model 1), and metabolite concentrations were predicted by a linear age-by-genotype interaction (Model 2), with sex covaried and a random intercept for each subject. Genotype effects were evaluated as a group effect of Tg rats relative to WT rats at each time point using four age-centred models, with age centred at the average cohort age (129.6, 310.7, 494.3, and 572.9 days). All continuous variables were z-scored. For MRS data, the fixed effect of water linewidth was included to control for the effect of linewidth on metabolite concentration estimates. 47 A weighting factor of the inverse absolute Cramer-Rao Lower Bound (CRLB) for each metabolite accounted for differences in fitting reliability between samples. We also examined a three-way interaction of age-by-genotype-by-sex with the same covariates as mentioned above for brain volumes (Model 3) and metabolite concentrations (Model 4).
For all linear models, the false discovery rate (FDR) method 48 was used to control the family-wise Type I error at a level of 5% for each predictor of interest. Details on attached base packages in R, Akaike information criterion comparisons, and Linear Models 1 through 4 are included in Supplementary Methods.
Barnes Maze data were analysed cross-sectionally. % success, number of holes searched, and speed were analysed using a linear model with genotype as a fixed effect and sex covaried, or genotype and sex interacting (secondary analysis). % time in the target quadrant and % time in target holes were assessed using a one-sample t-test or a Wilcoxon signed rank test (if test residuals were non-normal) against a mean of 25% (chance amount of time) within WT and Tg rats, as well as for genotypes split by sex (WT males, WT females, Tg males, and Tg females). Bonferroni correction was applied at each time point for the primary and secondary analyses separately, whereby the P-value threshold was set at 0.05/7 tests (P , 0.00714) or 0.05/11 tests (P , 0.00455), respectively.

Data availability
Data are reported within the text, figures, and supplementary material. Raw data are published to the publicly available repository, Zenodo, at 10.5281/zenodo.6338797.

Results
TgF344-AD rats display altered local brain volume, primarily in GM structures Volume changes for the age-by-genotype interaction term of Model 1 are illustrated as t-statistic maps in Fig. 1A for voxelwise (left) and regional (right) analyses. Significant effects were generally consistent between the two methods and the majority were bilateral. Brain structures demonstrating significant age-by-genotype interactions are summarized in Table 1. Of the significant interactions, 19 occurred in GM regions, 7 in white matter (WM) regions, and one in the ventricular system. Most interactions (16 of 19 for GM, 5 of 7 for WM) were negative, indicating decreased volume with age in Tgs relative to WTs. The strongest interactions were atrophy in the basal forebrain, caudoputamen, fimbria, hippocampus, and nucleus accumbens, unilateral atrophy in the right fornix, and hypertrophy in the dentate gyrus. Weaker effects were present as increases in cerebellar WM and aqueduct volume, and decreased ventral pallidum, lateral septum, and hypothalamus volume. The basal forebrain was the only structure to demonstrate a significant quadratic interaction, indicating different curvilinearity in the volume trajectory of Tg rats relative to WTs. Whole-brain GM, WM, and CSF volumes were also quantified but did not differ by genotype.
Four age-centred models examining the main effect of genotype at each time point were used to provide group difference snapshots of the timeline along which structural changes occur in the TgF344-AD model. As shown in Table 1, volume differences between WTs and Tgs were present at 4 months for the aqueduct, caudoputamen, dentate gyrus, nucleus accumbens, and fimbria. Hippocampal atrophy was not significant until 16 and 18 months of age and was preceded by a period of marginally increased volume relative to WTs. Volume trajectories for selected structures are shown in Fig. 1B. Figure 1C shows trajectories for two structures split by sex and depicts results from the analysis using a three-way interaction between quadratic age, genotype, and sex (Model 2) to predict regional volume. The linear age-by-genotype-by-sex interaction term for the left hypothalamus was negative and the left dentate gyrus demonstrated a positive quadratic age-by-genotype-by-sex effect. However, neither these effects nor any other structures evaluated with Model 2 survived FDR correction.
A summary of linear model results for brain regions analysed using Models 1 and 2 is shown in Supplementary Table 2. Brain volumes in mm 3 at each time point, both collapsed across and split by sex, are summarized in Supplementary Table 3. Trajectories of brain structures showing a significant age-by-genotype interaction via Model 1 can be found in Supplementary Figs 1 and 2. Those demonstrating significant three-way interactions via Model 2 (prior to FDR correction) are shown in Supplementary Figs 3 and 4.
The TgF344-AD model recapitulates neurochemical features of Alzheimer's disease A total of 27 hippocampal neurochemicals were quantified longitudinally in Tg rats relative to WT littermates. The neurochemical profile contained nine macromolecule resonances which have yet to be quantified in this model. High-quality spectra were consistently obtained, as shown by the representative spectrum obtained from a 10-month-old WT female (Fig. 2), and by the low % CRLB values shown in Supplementary Table 4. The average signal-to-noise ratio of the NAA peak at 2.02 ppm was 61.77 [+13.96 (+ standard deviation), range: 24.51-107.35], and the average linewidth of water was 9.21 Hz (+0.73, range: 7.74-12.79; measured as the full width at half max of the unsuppressed water peak in the reference scan).
The primary analysis explored the interaction between linear age and genotype while controlling for sex (Model 3). As shown in Table 2 and Fig. 3A, tCho and Ins demonstrated significant positive age-by-genotype interactions, whereby metabolite concentration increased more steeply with age in Tg rats than in WTs, but did not survive FDR correction. Upon examining the main effect of genotype at each time point using age-centred models, several metabolites differed between WT and Tg rats, with the earliest differences detected at 10 months (Table 2). Total creatine (tCr), taurine, and the ratio of NAA to Ins (NAA/Ins) were decreased in Tg rats at 10 months of age and remained lower at 16 and 18 months, whereas the ratio of aspartate (Asp) to Glu (Asp/Glu) was significantly lower at 10 months only. NAA was significantly lower at 10, 16, and 18 months but not after FDR correction. Higher concentrations of lactate (Lac) and tCho were evident at 10, 16, and 18 months, while Ins was significantly higher in Tg rats only at 16 and 18 months of age. None of the macromolecule peaks differed between Tg and WT rats.
A secondary analysis explored a three-way interaction between age, genotype, and sex (Model 4). Both glucose (Glc) and Ins demonstrated three-way interactions but were not significant after FDR correction. Neurochemical trajectories for these metabolites, along with tCho and Gln, which showed sub-threshold (P , 0.15) three-way interactions prior to FDR correction, are shown in Fig. 3B. A full summary of linear model results is shown in Supplementary  Table 4, with the concentration of each neurochemical (mM) included in Supplementary Table 5. Trajectories of select metabolites are shown in Supplementary Fig. 5.
The TgF344-AD model displays cognitive impairment by 4 months of age Long-term spatial reference memory in Tg and WT rats was evaluated at each time point via the probe trial of the Barnes Maze test, conducted 48 h after the last training trial. Figure 1 Genotype-dependent differences in local brain volume with age. (A) Voxel-wise (left) and regional (right) statistical maps for the linear age-by-genotype interaction term are shown. The plot range for each set of t-values displays an interaction significant between 5 and 1% FDR, with t-values corresponding to FDR , 1% displayed at the 1% value. Positive t-values (warm colours) represent increasing volume in TgF344-AD rats over time relative to WT littermates (positive age-by-genotype interaction), whereas negative numbers (cool colours) represent decreasing volume in TgF344-AD rats (negative age-by-genotype interaction). Select regional volume trajectories in WT and TgF344-AD (Tg) rats are shown in (B). The mixed-effects model used to fit the data is represented by a line of best fit and 95% prediction interval (shaded). Significance symbols are shown for the linear age-by-genotype interaction term ( ‡) and quadratic age-by-genotype interaction term ( ‡ 2 ). The main effect of genotype as determined by the four age-centred models is shown by ( †), with the subscript denoting at which age the main effect was significant. (C) A three-way interaction between age, genotype, and sex was also explored, with a 5% FDR correction applied. Significance symbols for the linear and second order age-by-genotype-by-sex interaction terms are denoted by X and X 2 , respectively. • denotes an effect significant at the P-value level but not after FDR correction.
Cognitive impairment in Tg rats was evident as early as 4 months of age, as determined by testing the percentage of time WT and Tg rats spent in the target quadrant against the chance amount of time a rat would spend in each quadrant. A mean significantly above 25% is suggestive of intact spatial memory recall, which WT rats demonstrated throughout the study, while Tg rats did not meet the significance threshold at any time points (Fig. 4A (left)). A similar effect was seen when testing the percentage of time spent exploring holes within the target quadrant ( Fig. 4B (left)). WT rats consistently spent more than a chance amount of time exploring holes in the target quadrant, whereas Tg rats did not. Additionally, as shown in Fig. 4C (left) the rate of success versus failure in locating the escape hole was lower among Tg rats throughout the study, though this difference was not statistically significant. The number of holes searched and average speed were also measured during the probe trial to characterize level of exploration and mobility, respectively. As shown in Fig. 4D, E (left), both metrics differed between WT and Tg rats, with Tgs searching fewer holes throughout the study and moving more slowly than their WT littermates. However, after Bonferroni correction, the differences in holes searched was only significant at 10 and 16 months, while the difference in speed was only significant at 10 months.
A secondary analysis examining the interaction between genotype and sex was also performed. As shown in Fig. 4A (right), only WT males consistently spent significantly more than 25% of the time in the target quadrant, although the 4-month time point did not reach the significance threshold after the Bonferroni correction. WT males generally also spent a higher percentage of time in the target holes than WT females, Tg males, or Tg females, but only reached the significance threshold at 16 months (Fig. 4 (right)). Interestingly, and as shown in Fig. 4C, D (right), Tg females demonstrated the lowest success rate across all time points, and the lowest rates of exploration (fewest holes searched) at three of four time points, though no significant genotype-by-sex interactions were found for either metric. Finally, as seen in Fig. 4E (right), no genotype-by-sex interactions were found for speed at any time point. All Barnes Maze statistics can be found in Supplementary

Discussion
The TgF344-AD rat is unique in its manifestation of amyloid and tau pathology despite only expressing mutant APP and PS1, and therefore closely replicates human Alzheimer's disease. Thorough characterization of the manifestation and progression of physiological abnormalities-particularly those that can be measured non-invasively-comprising each disease stage in this rat model is required for designing effective therapeutic approaches. While other authors have characterized pathological features in this model, the majority of our neuroimaging findings are being reported for the first time and ours is the earliest assessment of cognitive function. This study also explores the intersecting influence of sex Beta + + + + + SE P-value Beta + + + + + SE P-value Beta + + + + + SE P-value Beta + + + + + SE P-value Beta + + + + + SE P-value Aqueduct 6.0 + 1. 8  and genotype on neuroimaging and behavioural markers, which is crucial for improving diagnostic methods and interventions given that Alzheimer's disease prevalence and manifestation can differ between men and women. 49 As determined via post-mortem histology studies, decreased brain volume detected via MRI is an accurate marker of Alzheimer's disease-related neurodegeneration that can be used to support a clinical diagnosis in humans. 13,50,51 Models of human disease progression indicate early tissue pathology and atrophy in the entorhinal cortex 52 and other regions that comprise the limbic system, particularly the hippocampus. 50,53,54 The TgF344-AD model does not demonstrate significant cortical atrophy by 18 months, but does recapitulate volume reductions in limbic structures, including the hippocampus, basal forebrain, fimbria, fornix, hypothalamus, and lateral septum. In opposition to dentate gyrus atrophy reported in human Alzheimer's disease, 15,55 we observed hypertrophy in the TgF344-AD model, with significantly larger volume at 16 and 18 months in Tgs relative to WTs. This is not entirely unexpected given that Fischer 344 rats display increased dentate gyrus volume during normal aging, 56,57 and therefore further hypertrophy during Alzheimer's disease may represent a pathological feature, similar to how normal hippocampal atrophy with age is exacerbated in Alzheimer's disease. 58 Additional structures with significant atrophy were the nucleus accumbens, caudoputamen, and ventral pallidum, all of which display amyloid and tau pathology and structural changes in human Alzheimer's disease. 15,54,59 The only other structural MRI-based study in the TgF344-AD rat reported a lack of major structural differences, but this study used only female rats and therefore a direct comparison between their results and ours is not particularly meaningful. 60 Age-centred analyses revealed that while many structures atrophied faster in Tgs, this was occasionally preceded by hypertrophy. For example, Tg rats demonstrate larger caudoputamen, fimbria, and nucleus accumbens volumes until 10 months, and then smaller volumes at 16 and 18 months relative to WTs. Supporting these findings, a neuroimaging study in PS1 mutation carriers reported increased caudate volume in asymptomatic individuals but decreased volume in symptomatic individuals, 61 suggesting different processes underlie morphometric change at different stages of disease progression. Reactive neuronal hypertrophy in the hippocampal CA1 region has been shown in Alzheimer's disease subjects prior to symptom onset 62 supporting early regional volume increases, either as a cellular response to amyloid and tau deposition, or a compensatory process prior to degeneration of neurons and synapses. 63 Future work combining MRI-based volumetric analysis and design-based stereology, similar to studies in transgenic mice, 64,65 would help fill critical gaps in knowledge regarding mechanisms underlying pathological morphometric change in the TgF344-AD model.
MRS allows for quantification of brain tissue metabolites, providing insight into the biochemical underpinnings of altered brain structure and function. 12 Similar changes in NAA, Ins, and NAA/Ins to what we report have been shown in the TgF344-AD rat, 66 the McGill-R-Thy1-APP rat model, 26 mouse models, 25,67,68 and in human studies. 27,28,69 Decreased NAA reflects reduced neuronal viability-specifically mitochondrial dysfunction-as opposed to purely neuronal density. 29,30 The possible mechanisms behind increased Ins are more varied and may reflect increased glial cell activation and/or inflammation, increased phagocytic activity, or cellular membrane disruption, as Ins is a precursor for  Linear age-by-genotype interactions and main effects of genotype are shown under age:genotypeTg and genotypeTg (age) columns, respectively. Betas are standardized and SE denotes the standard error of the beta estimate. Unadjusted P-values are shown, with bold font denoting P-values that were still significant after applying FDR correction at 5%. The number following each macromolecule denotes the frequency at which the peak is located. Asp/Glu, aspartate/glutamate; Cr + PCr, creatine + phosphocreatine; Glu + Gln, glutamate + glutamine; GPC + PCh, glycerophosphocholine + phosphocholine; GSH, glutathione; MM, macromolecule; NAA/Ins, N-acetylaspartate/myo-inositol; NAA, N-acetylaspartate; NAAG, N-acetylaspartylglutamate; PE, phosphoethanolamine. The mixed-effects model used to fit the data is represented by a line of best fit and 95% prediction interval (shaded). Significance symbols are shown for the linear age-by-genotype interaction term ( ‡), and the main effect of genotype ( †) at each time point as determined by age-centred models. The subscript denotes the age at which the genotype effect was significant. (B) A three-way interaction between age, genotype, and sex was also explored, with a 5% false discovery rate correction applied. A significant three-way interaction term is denoted by X. • denotes an effect significant at the P-value level but not after FDR correction. inositol lipid synthesis, a constituent of membrane lipids, and an osmolyte. 30,70,71 Decreased taurine in the TgF344-AD rat is in agreement with the literature supporting the role of taurine in neurite outgrowth, synaptogenesis, and synaptic transmission, 72 all of which are dysfunctional in Alzheimer's disease. 73 Other differences between Tg rats and WT controls included a lower Asp/Glu ratio, lower tCr, and higher Lac at 10 months of age. These differences suggest the TgF344-AD model replicates the well-documented phenomenon of altered bioenergetics in human Alzheimer's disease, 74,75 specifically, disrupted excitatory neurotransmission and a shift towards non-oxidative energy metabolism. 29 These findings also indicate tCr should not be used as an internal reference in this model. Finally, our report of increased tCho is in agreement with studies in human Alzheimer's disease patients 27,76,77 and likely reflects increased cell membrane turnover (a feature characteristic of neuronal degeneration 78 ) and/or inflammation and astrocytosis. 30,79 Overall, the neurochemical profile of the TgF344-AD rat closely replicates that of human patients and provides insight into numerous pathological processes, substantiating its application in Alzheimer's disease research. In particular, our findings build upon the study in TgF344-AD rats by Chaney et al., 66 by studying both male and female rats, characterizing additional metabolites, and through the inclusion of individual macromolecule peaks, which, while they did not differ between WT and TgF344-AD rats, inclusion of basis functions for these peaks as performed here has been shown to improve metabolite quantification. 80,81 Deficits in hippocampus-dependent spatial learning and memory are among the earliest complaints in Alzheimer's disease subjects. [82][83][84] Previous studies indicate that 5-monthold TgF344-AD rats require more trials to learn a delayed non-match-to-sample task 85 and impaired reversal learning on the Morris Water Maze 86 and Barnes Maze 10 by 6 months. To test spatial navigation in our TgF344-AD rats, we used a shortened version of the Barnes Maze test 44 which detected impairment in 3xTg mice earlier than traditional protocols. 43 Fittingly, ours is the earliest report of cognitive disturbance in this model, with impairments in long-term spatial reference memory present by 4 months of age. We also noted genotypedependent differences in speed and number of holes searched, substantiating the choice to use speed-and motivationindependent measures rather than the frequently chosen escape latency or number of errors. 87,88 Regarding the interaction between sex and genotype, Tg females demonstrated stronger pathological effects in several brain volumes and more cognitive decline than Tg males, whereas sex effects on metabolite concentration were also present but did not consistently impact Tg females. To our knowledge, we are the first group to examine the interaction between age, genotype, and sex on neuroimaging markers in this model. However, sex differences in the TgF344-AD rat exist in open field and buried food tasks 89 and the Morris Water Maze test, 90 and sex differences in neuroanatomy and hippocampal tissue chemistry during normal aging in the Fischer 344 rat have been reported. 32,37 These findings generally recapitulate human Alzheimer's disease data. Sex-specific patterns of neurodegeneration exist in human patients 91,92 and men frequently present with later and less severe cognitive deficits than women. 93,94 There is also an established role for oestrogen in regulation of metabolic pathways affected by Alzheimer's disease such as Glc transport, aerobic glycolysis, and mitochondrial function. 73,94,95 While additional work is required to corroborate our findings, the TgF344-AD model appears to recapitulate known sex differences in several aspects of disease presentation.
Regarding corroboration of the timing of pathological changes that we report in the TgF344-AD rat, previous studies show by 5-6 months of age, TgF344-AD rats display disruption in hippocampal-dependent synaptic circuits, 85,96 dysfunction of the noradrenergic system, 86 and loss of functional connectivity prior to the appearance of microstructural alterations. 60 Reduced maximum synaptic transmission in the hippocampus occurs between 9 and 12 months, in the absence of reduced dendritic spine density, 97 continuing to support a timeline of functional change prior to significant morphometric change. Profound cerebral microvascular and neuronal network dysfunction is present at 9 months, 98,99 along with reduced antioxidant capacity, and increased reactive oxygen species and pro-inflammatory cytokines at 10 months. 100 Our report of neurochemical changes by 10 months is consistent with these previously described molecular events. At 13 months, TgF344-AD rats demonstrate deficits in hippocampal neuronal differentiation, migration, and survival, 101 and display significant tau pathology, neuronal damage, and cognitive impairment between 16 and 26 months. 10,66,102 Given that reduced synaptic density and neuronal loss are associated with MRI-detectable volume changes, 51,103 these reports of altered neurogenesis and neuronal damage may reflect some of the processes underlying the volumetric changes we report.
This timeline of biochemical changes preceding substantive structural abnormalities is corroborated by models of disease progression 3 and studies exploring upstream and downstream processes of amyloid and tau deposition. 63,73,104 Additionally, the timing of biochemical and structural changes around midlife to the beginning of senescence suggests altered neurochemistry and neuroanatomy may be in response to amyloid and tau pathological load, 10,86 and are not evident before the appearance of gross tissue pathology. In contrast, early cognitive impairment differs from disease progression in humans where cognitive complaints are among the last pathological features to manifest. Given numerous studies have validated the consistency with which this model mimics the spread of components of human Alzheimer's pathology, this difference in timing of cognitive dysfunction may reflect inconsistencies in how cognition is tested or presents in rodent models versus humans. Behavioural testing is also extremely variable and subjective. Neuroimaging is considerably more objective and less variable, thus providing a better powered, sensitive, accurate, and efficient means to characterize disease progression. To our knowledge, we are the first group to simultaneously characterize the timing of the manifestation of neuroanatomical, neurochemical, and cognitive abnormalities in the TgF344-AD rat, thus providing new information that can be used to define disease stages in this model.
There are limitations to consider when interpreting the results of the present study. First, this study does not include any histological data, limiting our ability to understand the cellular underpinnings of the neurochemical, neuroanatomical, and cognitive changes that we report. Thorough characterization of tissue pathological load of disease hallmarks in the TgF344-AD model has been performed by other authors, confirming the presence of amyloid-β deposition, gliosis, neuronal loss, and tauopathy 10,66,100 in regions and along a time course similar to that of human Alzheimer's disease. 79 A future study combining in vivo neuroimaging with ex vivo histological analyses, particularly examining both number and morphology of microglia, astrocytes, and neurons would greatly contribute to our knowledge regarding the cellular basis for the changes in neuroimaging markers that we report here.
Second, the use of a polynomial age term in the volumetric analyses, which was necessary given the non-linear change with age that we and others report, 19,56,105,106 likely reduced our power to detetct age-by-genotype-by-sex interactions. Given that most structures demonstrated volume change towards 16 months, a paradigm where brain volumes are quantified from midlife onwards may permit the use of a linear age term and provide more power to detect three-way interactions. Second, restrictions on facility access due to the COVID-19 pandemic resulted in fewer animals being tested and increased variation in testing dates at the final time point, as well as inconsistency in the time between staining and imaging during immunofluorescence experiments. While these inconsistencies were accounted for in the statistical modelling, this is likely to have increased the overall variation in the data, possibly masking or muting some of the effects at the final time point. Finally, the lack of histological analyses at early time points meant we were unable to determine if pathological changes in neuroimaging markers precede those at the cellular level in the TgF344-AD rat. This gap in knowledge limits our ability to interpret the origin of the altered neuroimaging and cognitive markers that we report.
Altogether, our results provide a comprehensive review of multiple phenotypic components of pathology in the TgF344-AD model, characterized from early to late stages of disease progression. This longitudinal multimodal study demonstrates that the TgF344-AD rat recapitulates major neurochemical, neuroanatomical, and cognitive features of human Alzheimer's disease, and furthers our understanding of the many processes comprising disease progression. These findings support the use of MRI and MRS to monitor disease progression in rodent models of Alzheimer's disease in vivo and contribute to the growing body of work validating the TgF344-AD rat as a highly relevant model of Alzheimer's disease.