Abstract
The overexpression of calcineurin leads to astrocyte hyperactivation, neuronal death, and inflammation, which are characteristics often associated with pathologic aging and Alzheimer's disease. In this study, we tested the hypothesis that tacrolimus, a calcineurin inhibitor, prevents age-associated microstructural atrophy, which we measured using higher-order diffusion MRI, in the middle-aged beagle brain (n = 30, male and female). We find that tacrolimus reduces hippocampal (p = 0.001) and parahippocampal (p = 0.002) neurite density index, as well as protects against an age-associated increase in the parahippocampal (p = 0.007) orientation dispersion index. Tacrolimus also protects against an age-related decrease in fractional anisotropy in the prefrontal cortex (p < 0.0001). We also show that these microstructural alterations precede cognitive decline and gross atrophy. These results support the idea that calcineurin inhibitors may have the potential to prevent aging-related pathology if administered at middle age.
SIGNIFICANCE STATEMENT Hyperactive calcineurin signaling causes neuroinflammation and other neurobiological changes often associated with pathologic aging and Alzheimer's disease (AD). Controlling the expression of calcineurin before gross cognitive deficits are observable might serve as a promising avenue for preventing AD pathology. In this study, we show that the administration of the calcineurin inhibitor, tacrolimus, over 1 year prevents age- and AD-associated microstructural changes in the hippocampus, parahippocampal cortex, and prefrontal cortex of the middle-aged beagle brain, with no noticeable adverse effects. Tacrolimus is already approved by the Food and Drug Administration for use in humans to prevent solid organ transplant rejection, and our results bolster the promise of this drug to prevent AD and aging-related pathology.
Introduction
Alzheimer's disease (AD) is the most prevalent neurodegenerative disorder in the world, affecting >45 million people (Dos Santos Picanco et al., 2018). It is primarily characterized by dementia, a decline in memory, and other cognitive skills beyond what is typically observed in healthy aging (Reitz and Mayeux, 2014). The greatest risk factor for AD remains age, and most people in whom the disease develops are older than 65 years (Inouye et al., 2010).
The two major neuropathological features of AD are abnormally folded β-amyloid (Aβ) peptides and the accumulation of hyperphosphorylated tau proteins in amyloid plaques and neurofibrillary tangles (Perl, 2010; Holtzman et al., 2011; Stancu et al., 2014; Forestier et al., 2015; Hendrie et al., 2015; Liu et al., 2015). Because of the overwhelming evidence that Aβ plaques play a role in AD and the amyloid cascade hypothesis (Hardy and Higgins, 1992), most therapeutic strategies have focused on reducing or at least controlling the formation of these plaques. However, clinical trials that use Aβ-reducing approaches have shown limited clinical efficacy, prompting the exploration of treatments that target other factors or pathways driving this disease, perhaps upstream of Aβ accumulation (Pahnke et al., 2009).
Mounting evidence suggests that the Ca2+/calmodulin-dependent protein phosphatase, calcineurin, and downstream signaling pathways are an attractive target for ameliorating cognitive decline in AD and related disorders (Reese and Taglialatela, 2011; Sompol and Norris, 2018). Calcineurin is found at high levels in neurons and reactive glial cells where it modulates synaptic plasticity, neuroinflammation, glutamate regulation, and memory formation (Mansuy, 2003). Elevated levels of calcineurin expression and signaling are found in the hippocampus and other cortical areas at the outset of cognitive decline in humans (Abdul et al., 2009; Mohmmad Abdul et al., 2011) and are highly correlated with pathologic features in later disease stages (Liu et al., 2005; Abdul et al., 2009). Overexpression or hyperactivation of calcineurin in experimental models recapitulates key features of AD, including glial reactivity (Norris et al., 2005), synaptic degeneration (Wu et al., 2010), and cognitive dysfunction (Malleret et al., 2001). Conversely, the inhibition of calcineurin signaling via genetic or pharmacologic means reverses many of these AD-related biomarkers in animal models (Reese et al., 2008; Taglialatela et al., 2009; Rozkalne et al., 2011; Furman et al., 2012; Hudry et al., 2012; Rojanathammanee et al., 2015; Kumar and Singh, 2017; Sompol et al., 2017). In the clinic, calcineurin inhibitors, like tacrolimus, are used primarily as immunosuppressants to combat organ transplant rejection and other autoimmune disorders. However, an epidemiological study in 2015 showed that kidney transplant patients treated with tacrolimus had a significantly lower incidence of dementia relative to age-matched individuals in the general population (Taglialatela et al., 2015). Collectively, this work suggests that tacrolimus and other Food and Drug Administration (FDA)-approved calcineurin inhibitors could be repurposed for the prevention of AD and dementia.
The FDA-approved status, and the well known safety profiles and contraindications of calcineurin inhibitors would certainly make the path to AD clinical trials easier. But, to ensure that calcineurin inhibitors have the best chance of succeeding as anti-AD therapeutics requires further optimization in a preclinical model that better approximates human metabolism, neural function, treatment course, and biomarker milestones.
Dogs have a metabolism that is very similar to that of humans and are excellent preclinical models for testing pharmacological agents (Dalgaard, 2015). More importantly for investigating anti-AD treatments, dogs naturally show age-related amyloid plaque pathology, neuroinflammation, and neurodegeneration (Sarasa and Pesini, 2009; Prpar Mihevc and Majdič, 2019). Human-like deficits in cognition also arise with aging and correlate well with pathologic features. Because of their longer life span, larger brain size, and complexity, and ease of training, dogs are amenable to the longitudinal assessment of neurologic function using complex cognitive/behavioral batteries and brain imaging, which are common to most modern human clinical trials. (Patronek et al., 1997; Hoffman et al., 2018). Given these clear benefits, we explored the microstructural consequences of tacrolimus on the brain of a preclinical aging beagle model (age, 4–8 years) using diffusion-weighted imaging (DWI). Though traditional MRI procedures (like T1- and T2-weighted imaging) are noninvasive, they only provide a mesoscopic view as even their highest resolutions are well too coarse to resolve changes at the expected microscopic level, at least directly. In contrast, DWI provides measures that are sensitive to the underlying microstructure and its changes in disorders such as AD (Chua et al., 2008). Here, we use the following two types of diffusion analysis techniques to survey the potential cytoarchitectural changes (or lack thereof) that tacrolimus could induce in the beagle brain: (1) traditional diffusion tensor fitting (Basser et al., 1994); and (2) neurite orientation dispersion and density imaging (NODDI) analysis (Zhang et al., 2012).
Materials and Methods
Animals and drug delivery
Forty-five (7 males and 38 females) purpose-bred beagles ranging in age from 5 to 8 years were assessed for general health status and cognition, as described previously (Head et al., 1998; Milgram et al., 1999, 2002; Tapp et al., 2003; Christie et al., 2005; Studzinski et al., 2006). Dogs ranged in weight from 8.6 to 14.5 kg. Since tacrolimus has previously been associated with nephrotoxicity in renal transplant patients (Randhawa et al., 1997), blood samples were taken every 6 months to monitor the overall health and to assess blood urea nitrogen (BUN), creatine, and phosphorous levels of the dogs. All institutional and national guidelines for the care and use of laboratory animals were followed.
Cognitive testing
Cognitive testing used a modified Wisconsin General Test Apparatus described previously (Head et al., 2008). Dogs were given 10–12 trials/d, 5 d/week, depending on the cognitive task. All tasks were reward motivated and based on visual cues. Dogs were given baseline tests of visual discrimination learning and reversal learning to assess learning and executive function. Subsequently, a spatial delayed nonmatch to sample task was used to assess spatial learning and memory. At the end of baseline testing, dogs were ranked according to cognitive test scores and balanced into three groups consisting of 15 animals/group. Groups were also balanced for age.
Drug administration
Oral tacrolimus at a concentration of 0.075 mg/kg twice a day (n = 15; two males) or an oral placebo control (n = 15; two males) was administered for 1 year. The remaining 15 animals were assigned to another intervention study not relevant to this article; only their baseline data are included here to improve the statistical power of age associations. The concentration of the drugs was designed to provide minimal immunosuppression to reduce adverse effects (Margarit et al., 1998).
MRI image acquisition
Dogs were placed under general anesthesia using propofol (4–8 mg/kg, i.v., by slow injection to effect). After orotracheal intubation and maintenance on isoflurane 1–4%, delivered in 100% O2, dogs were scanned using a 3 T MRI scanner (Prisma, Siemens) both at baseline before treatment and after 1 year of treatment.
T1 weighted.
A high-resolution T1-weighted (T1w) MPRAGE image was collected [repetition time (TR) = 2530 ms; echo time (TE) = 2.49 ms; flip angle = 7°; matrix size = 0.4 × 0.4 × 0.7 mm; averages = 1; average acquisition time = 10 min, 30 s] for structural analysis and image registration.
DWI.
Diffusion imaging (TR = 5700 ms; TE = 62 ms; 48 coronal slices in the animal reference frame; phase encoding, superior–inferior; average acquisition time = 12 min, 30 s) was performed using a double refocused echoplanar sequence with an isotropic 1.6 mm voxel for three gradient values: b = 500, 1000, and 2000 s/mm2. Gradients were applied in a total of 114 directions, along with 13 images with no diffusion weighting (b = 0).
Diffusion data preprocessing
All preprocessing steps used MRtrix3 (Tournier et al., 2012; https://www.mrtrix.org/) commands or MRtrix3 scripts that linked external software packages. Physiologic noise arising from thermal motion of water molecules in the brain was removed first (Veraart et al., 2016), followed by the removal of Gibbs ringing artifacts (Kellner et al., 2016). The image intensity was then normalized across subjects in the log-domain (Raffelt et al., 2012; Andersson and Sotiropoulos, 2016).
Structural data processing
The T1w images were corrected for intensity inhomogeneities using advanced normalizations tools (ANTs) N4 bias field correction (Tustison et al., 2010). The structural image of each dog was then nonlinearly coregistered to their respective preprocessed b0 image, so that the structural and diffusion images were in the same space for the rest of the analyses. To help standardize our results, we used the Aguirre high-resolution ex vivo template (Datta et al., 2012). In this space, we generated a high-resolution central tendency template from the structural scans using ANTs with an initial cohort of 10 animals and a bootstrapping approach. We then used this template to generate initial priors for both brain extraction and tissue type segmentation (gray matter, white matter, CSF, deep gray matter, and cerebellum) again using ANTs and a bootstrapping approach with more refined priors. The result was a set of priors that can be used in a modified ANTs cortical thickness pipeline to generate final tissue segments for each subject. Though the Aguirre template provides an excellent standard template space complete with high-resolution ex vivo scans, it does not provide adequate labels for ROI-specific analyses. To solve this problem, we coregistered two different atlases, hereby referred to as the Nitzsche atlas (Nitzsche et al., 2019) and the Czeibert atlas (Czeibert et al., 2019) to the same template space using the affine+SyN nonlinear registration in ANTs. The Nitzsche atlas was used for large-area ROIs like entire lobes, while the Czeibert atlas was used for more specific subregions like the hippocampus. All resulting images were visually inspected for quality and rerun with new command line parameters when necessary (Fig. 1). Region-wise volumes were determined by warping the annotated atlas back to each individual's subject space and quantifying the number of voxels that made up each region of interest.
Summary of the analysis pipeline.
Deriving diffusion metrics
We calculated traditional tensor metrics [fractional anisotropy (FA) and mean diffusivity (MD)] using FSL (version 6.0.1; Jenkinson et al., 2012) and higher-order, multicompartment metrics [neurite density index (NDI), orientation dispersion index (ODI), and fractional isotropy (FISO)] using the NODDI (Zhang et al., 2012) model with the Microstructure Diffusion Toolbox (Harms et al., 2017). The traditional tensor metrics are widely used, but typically are applied only to white matter. NODDI metrics are tissue type agnostic and can readily be used in gray matter as it characterizes diffusion within each voxel as a combination of intracellular, extracellular, and CSF-based components. The intracellular compartment ostensibly captures neurite membranes and myelin sheaths and is modeled as a set of sticks with restricted diffusion perpendicular to the orientation of the axonal bundles and unhindered diffusion along them. The extracellular compartment is designed to model the space around the neurites, composed of glia and somas, as hindered Gaussian anisotropic diffusion. The CSF is modeled as isotropic diffusion. A summary of all the diffusion metrics used is provided in Table 1.
A description of the diffusion metrics used
The hippocampus, parahippocampal gyrus, and prefrontal cortex (PFC) were selected as a priori regions because aging and AD present early changes in these regions in dogs (Shimada et al., 1992; Thal et al., 2002; Ezekiel et al., 2004; Su et al., 2005; Tapp et al., 2004, 2006; Hwang et al., 2008; Head, 2011). Region-specific averages were obtained by aligning the Czeibert atlas (Czeibert et al., 2019) to each subject's parametric maps. Diffusion metrics were then averaged across each region of interest using AFNI (Analysis of Functional NeuroImages; Cox, 1996). All statistical analyses were performed in Python Scipy (Jones et al., 2001; https://www.scipy.org) or GraphPad Prism 8.3.0. All regression analyses were simple linear regressions. The effects of interventions were assessed with ANOVA, and multiple comparisons were corrected using Holm–Sidak statistical hypothesis testing (Holm, 1979).
Whole-brain exploratory analysis was conducted to measure global longitudinal changes in each group separately using a paired t test in a voxelwise manner using 3dttest++ in AFNI. The AFNI clusterwise simulations (Forman et al., 1995) were used to correct for multiple comparisons. Parametric maps of each subject were passed in after they were registered to a common space, and a brain mask was passed in to improve power. To assess the interaction between intervention and time, a difference image was created (T0-T1) for each metric and each subject, and the difference images of each group were compared through an unpaired t test. The -clustsim option was used to determine the minimum cluster threshold for each individual test to maintain a final α of 0.05.
Data availability
Code for data processing and analysis is available at https://github.com/StarkLabUCI/Woofusion.
Results
The NDI of the beagle hippocampus and parahippocampal gyrus increases with age
Our first question was whether diffusion within hippocampal and parahippocampal gray matter was affected by age. Previous work in our laboratory has shown that the NDI of the hippocampus as a whole (Venkatesh et al., 2020), and specifically the DG/CA3 subfields (Radhakrishnan et al., 2020), is higher in older humans (59–84 years of age) than young adults (20–38 years of age) and that this increase is negatively correlated with memory performance (Radhakrishnan et al., 2020). Here, we found a similar relationship between age and hippocampal NDI at baseline, before treatment, in the canine model across all groups (simple linear regression: R2 = 0.111, p = 0.031; Fig. 2). Moreover, we observed a similar relationship between age and parahippocampal NDI (R2 = 0.131, p = 0.018; Fig. 2). This relationship between age and NDI was insignificant overall when averaging over the entire temporal lobe (R2 = 0.008, p = 0.567), suggesting a focused change in these regions with age.
At baseline, before treatment, the NDI of the hippocampus and the parahippocampal gyrus are positively correlated with age. Individual dots represent individual subjects. The line of best fit is in black, and the teal lines represent 95% confidence intervals.
To determine whether these results were driven by gray matter or white matter voxels, we classified individual voxels in these regions into gray or white matter by FA thresholding. Those voxels with FA > 0.4 were classified as likely white matter voxels, while those with FA < 0.4 were classified as likely gray matter voxels (Kumar et al., 2016). We found that the ratio of gray matter to white matter voxels was, on average, 7.48:1 in the hippocampus and 9.07:1 in the parahippocampal gyrus, suggesting that a clear majority of the signal we were detecting in these ROIs was driven by gray matter. Moreover, removal of the white matter voxels from the regions of interest when averaging across the parametric maps did not significantly change the results.
None of the other studied metrics showed a reliable relationship with age in the hippocampus or the parahippocampal gyrus, further bolstering our claim that the NDI might be capturing unique aging-associated microstructural properties in hippocampal and parahippocampal gray matter not typically detected by simple tensor metrics. We found no significant differences in NDI between hemispheres in both regions. Our male/female distribution did not permit us to test for sex differences.
One year treatment with tacrolimus results in a decrease in hippocampal and parahippocampal NDI and an increase in parahippocampal ODI
Dogs treated with tacrolimus for a year had significantly lowered hippocampal NDI (repeated-measures ANOVA, Sidak multiple-comparisons test: t = 3.976, p = 0.001, df = 25) and parahippocampal NDI (t = 3.711, p = 0.002, df = 25) NDI compared with baseline, suggesting that the drug might be rescuing some level of age-associated change (Fig. 3). Such a change was not observed between the time points for the control dogs in either of the regions. Though the dogs are not old enough to be exhibiting significant cognitive deficits (Milgram, 2003), previous studies in humans using structural equation modeling show that increased hippocampal NDI mediates age-related cognitive decline (Radhakrishnan et al., 2020), indicating that the drug may have the potential to protect against cognitive deficits if administered for a longer period of time.
One year treatment with tacrolimus significantly reduces the NDI in both the hippocampus (t = 3.976, p = 0.001) and the parahippocampal gyrus (t = 3.711, p = 0.002). There was a significant interaction between intervention and time in the hippocampus (F = 4.482, p = 0.044, ANOVA), but not in the parahippocampal gyrus (F = 2.579, p = 0.120). Error bars show the SEM.
The parahippocampal ODI significantly increased after a year in the control dogs (t = 3.197, p = 0.007), but not in the dogs treated with tacrolimus (t = 0.082, p = 0.995). We also observed a critical interaction between drug and time on ODI (F = 4.660, p = 0.040, ANOVA). We did not notice any correlations between parahippocampal ODI and age at baseline (R2 = 0.039, p = 0.805; Fig. 4), possibly because the dogs are middle aged, and we have a relatively restricted range. However, age-related increases in ODI have been reported in human studies with negative consequences (Nazeri et al., 2015; Mole et al., 2020; Venkatesh et al., 2020).
Though there was no correlation between age and parahippocampal ODI at baseline, the ODI of the parahippocampal gyrus increased in the control dogs after 1 year (t = 3.197, p = 0.007), but not in the dogs treated with tacrolimus (t = 0.082, p = 0.995), with a significant interaction between intervention and time (F = 4.660, p = 0.040, ANOVA). Error bars show the SEM.
As with the previous analysis, removal of the white matter voxels from the regions of interest when averaging across the parametric maps did not significantly change the results. No other studied metric showed an effect of time or intervention in these regions. We found no significant difference in the diffusion metrics between hemispheres for all regions studied.
Tacrolimus protects against structural changes in the prefrontal cortex
We next turned to changes outside of the hippocampal region. One of the first regions to be affected in the aging canine brain is the PFC. MRI studies have shown that the PFC starts reducing in volume at an earlier age (8–11 years) compared with the hippocampus (Tapp et al., 2004). Cognitively, aging also leads to poorer performance on tasks associated with the PFC, like reversal learning and visuospatial memory (Head et al., 1998; Tapp et al., 2003; Studzinski et al., 2006). While it is unclear whether the prefrontal cortex is an early region affected by age-related neuroinflammation, it is one of the first areas in the canine brain to develop plaques (Wieshmann et al., 1999; Bosch et al., 2012). Formation of these plaques has consistently been reflected in diffusion MRI studies as a reduction of fractional anisotropy (Tievsky et al., 1999; Wieshmann et al., 1999; Kealey et al., 2004). It is not very surprising that we found no significant relationship between age and prefrontal NDI or ODI at baseline (NDI: R2 = 0.001, p = 0.806; ODI: R2 = 0.006, p = 0.608), and these metrics did not significantly change in either group over the year. However, despite the lack of a significant relationship between age and prefrontal FA at baseline (R2 = 0.015, p = 0.440), it decreased in the control dogs after a year (t = 5.042, p < 0.001). This observation is directly analogous to the negative correlation between age and FA consistently observed in humans (Bennett et al., 2010; Kantarci et al., 2013). Interestingly, prefrontal FA did not decrease in the dogs treated with tacrolimus for a year (t = 1.890, p = 0.135), suggesting that the drug may be preventing age-associated structural deterioration in the prefrontal cortex (Fig. 5). The lack of a cross-sectional relationship with age at baseline might be attributed to individual differences and the dogs not being old enough to exhibit clear differences.
After 1 year, FA significantly reduced in the prefrontal cortex of the control dogs (t = 5.042, p < 0.0001), but not in the dogs treated with tacrolimus (t = 1.890, p = 0.135). The interaction between intervention and time was also significant (F = 4.568, p = 0.042, ANOVA). Error bars show the SEM.
We also segmented the PFC into white matter and gray matter regions, as described in the The NDI of the beagle hippocampus and parahippocampal gyrus increases with age subsection. We found that the ratio of gray matter to white matter voxels was, on average, 60.55:1, showing that an overwhelming majority of the signal was driven by gray matter. Removal of the white matter voxels from the regions of interest when averaging across the parametric maps did not significantly change the results. No other studied metric showed an effect of time or intervention in these regions. All effects reported were bilateral. Our male/female distribution did not permit us to test for sex differences.
Whole-brain exploratory analysis revealed disorganized decreases in white matter of the control dogs, but not of the dogs treated with tacrolimus
Following these a priori regional analyses, we conducted a whole-brain exploratory analysis to determine whether these changes were unique to these areas or whether they were found previously as well. Voxelwise comparisons were performed in a pairwise manner for each dog in both groups using the AFNI 3dttest++. We used the -clustsim option to determine the minimum cluster threshold to ensure an false discovery rate-corrected p value of at least 0.05, with an α of 0.05. We observed a disorganized, but large-scale, decrease in FA in many white matter regions (Fig. 6; 59,207 voxels survived thresholding) only in the control dogs. This was not unexpected, as the loss of white matter integrity is a classic hallmark of aging (Vernooij et al., 2008; Bennett et al., 2010; Madden et al., 2012). The dogs treated with tacrolimus did not show this same decrease (no voxels survived thresholding), further suggesting that the drug may be protecting against even sporadic neurodegeneration. However, these results should be interpreted cautiously as we found no significant interaction between intervention and time at our chosen thresholds [i.e., no voxels survived thresholding when comparing the difference image in time (T0 – T1) between the two groups].
Difference in T0 – T1 in FA of the control dogs. We found significant decreases in white matter FA after 1 year in control dogs, but not in the tacrolimus-treated dogs. Colored regions show regions where the FA at T0 was significantly different from the FA at T1 for control dogs (red–yellow, T0 > T1; teal–blue, T0 < T1). Dogs treated with tacrolimus are not pictured here as there were no significant voxels of difference when comparing the two time points.
Interestingly, no other diffusion metric studied exhibited reliable differences over time in either group, suggesting a very specific age-related decrease in the control dogs in only the a priori regions and a distinct protection against this effect by the drug. This finding bolsters our theory that NDI and ODI are sensitive to specific microstructural changes associated with age, and may be early predictors of medial temporal lobe pathology in these specific regions.
The hippocampal volume of both groups decreased over time, but not in the other a priori regions
We observed no significant relationship between age and hippocampal or parahippocampal volume at baseline as measured by the T1w image (hippocampus: R2 = 0.003, p = 0.771; parahippocampal gyrus: R2 = 0.027, p = 0.404), further demonstrating that these diffusion metrics may be detecting microstructural changes well before more large-scale volumetric changes present themselves. However, we observed a main effect of time on hippocampal volume for both groups (F = 9.986, p = 0.0041, ANOVA), suggesting that while the drug may protect against specific cytoarchitectural changes, it may not be able to protect against overall age-related volumetric atrophy in the hippocampus. As expected, no cross-sectional relationship between age and prefrontal volume was observed at baseline, and prefrontal volume did not change over the year in either group. There was also no global reduction in volume in either group, suggesting that age-associated atrophy is limited to the hippocampus at this stage in the life span. These results support the theory that volumetric changes in these regions occur further down the life span and that diffusion metrics may be earlier indicators of future pathologic and cognitive decline and may be more sensitive to measure interventional changes.
Limited cognitive changes were observed over time
We also assessed the effect of age on baseline cognition as well as the interaction between intervention and cognition after a year. We observed no significant relationship between age and discrimination learning (R2 = 0.014, p = 0.442) or reversal learning (R2 = 0.017, p = 0.393). However, age had a negative effect on spatial accuracy at 20 s (R2 = 0.172, p = 0.007) and spatial accuracy at 70 s (R2 = 0.109, p = 0.036), but not with spatial accuracy at 110 s (R2 = 0.045, p = 0.186). After 1 year of intervention, there was no significant difference in either group with respect to discrimination learning (control, p = 0.821; tacrolimus, p = 0.628) or spatial accuracy (control: p = 0.151, 0.796, 0.504; tacrolimus: p = 0.999, 0.471, 0.625 for 20, 70, and 110 s accuracy versions, respectively). Performance on reversal learning trended toward a decrease in error scores (i.e., better function) over time in the control dogs (p = 0.057), but not in the dogs treated with tacrolimus (p = 0.112). After the removal of an outlier in the tacrolimus group, this effect of time was significant in both groups (control, p = 0.041; tacrolimus, p = 0.029). These data, in conjunction, suggest that while these dogs are not exhibiting major cognitive decline, continued treatment will allow for more opportunities to see improvements. To that end, continued treatment may also be able to reveal whether the structural protection that tacrolimus grants to the study group translates to cognitive benefits as well. None of the cognitive scores studied were significantly correlated with the diffusion metrics, possibly because the middle-aged dogs do not yet show significant decline but are already displaying signs of microstructural deterioration.
Discussion
In this study, we used the drug tacrolimus to test the hypothesis that calcineurin inhibitors can prevent aging-related pathology, as measured by neuroimaging, in the middle-aged canine. We observed a positive correlation between hippocampal and parahippocampal NDI with age at baseline, a relationship that agreed with our observations in humans from previous studies (Radhakrishnan et al., 2020; Venkatesh et al., 2020). Interestingly, 1 year treatment with tacrolimus resulted in a decrease in both hippocampal and parahippocampal NDI, while the control dogs did not exhibit this effect. The drug also protected against an increase in parahippocampal ODI and a decrease in prefrontal FA, both consistently recognized as negative consequences of aging. We also showed that these changes precede most widespread volumetric changes and all cognitive changes and are specific to the a priori regions studied. These data, put together, suggest that (1) calcineurin inhibitors may rescue negative microstructural outcomes associated with age and (2) advanced diffusion imaging measures may be valuable biomarkers for predicting aging-associated pathology well before other symptoms are present.
The overexpression of calcineurin helps drive neuroinflammation and astrogliosis, which are commonly observed in aging (Rusnak and Mertz, 2000; Norris et al., 2005; Reese and Taglialatela, 2011). Although neuroinflammation is ultimately a systemic consequence of age, the dogs we studied are not old enough to exhibit these changes globally (subsection Whole-brain exploratory analysis revealed disorganized decreases in white matter of the control dogs, but not of the dogs treated with tacrolimus). However, the hippocampus and nearby regions are thought to be some of the initial hotspots of such inflammation (Akiyama et al., 2000; Verbitsky et al., 2004; Gavilán et al., 2007; Head, 2011), which could potentially be captured in our middle-aged model. Older dogs display an increase in GFAP immunoreactivity and protein levels in the hippocampus and neighboring regions, as well as increased astrogliosis and astrocyte hypertrophy (Borràs et al., 1999; Pugliese et al., 2006; Hwang et al., 2008). While there are currently no effective methods to measure such inflammatory changes noninvasively, several mouse studies have demonstrated reliable positive correlations between both the NDI and the ODI with immunoreactivity, astrocyte reactivity, and microglia count (Colgan et al., 2016; Grussu et al., 2017; Wang et al., 2019). Here, we showed that tacrolimus reduces hippocampal and parahippocampal NDI after just a year of treatment. Though neuropathological outcomes have not yet been obtained, increases in NDI in gray matter regions could be a consequence of inflammation, as microglial and astrocyte swelling cause the cells to expand, resulting in an increase in intracellular volume fraction, which is estimated by NDI (Colgan et al., 2016; Garcia-Hernandez et al., 2020). The hypothesis that tacrolimus may prevent neuroinflammation is further supported by our finding that it protects against an increase in parahippocampal ODI, which could be a marker for microglial density (Colgan et al., 2016; Yi et al., 2019; Garcia-Hernandez et al., 2020). Region-specific increases in microglial densities in the hippocampus and parahippocampal regions also precede plaque formation and are suppressed in mouse models of AD by inhibition of calcineurin signaling pathways (Furman et al., 2012; Sompol and Norris, 2018), again suggesting that tacrolimus administration in dogs may be protecting against aging-related pathologic changes through calcineurin inhibition (Marlatt et al., 2014; Fakhoury, 2018). However, these theories must be handled cautiously, as we are yet to find adequate histologic evidence for the neurobiological specificity of NODDI metrics.
We also found that tacrolimus protects against an age-associated decrease in prefrontal FA, suggesting that the drug may be capable of preventing, or at least delaying, the formation of amyloid plaques (Andrews-Hanna et al., 2007; Kantarci et al., 2017; Nasrabady et al., 2018), which, again, is consistent with the effects of calcineurin inhibition in rodent models (Hong et al., 2010). Moreover, the dogs showed no adverse effects on kidney function as a consequence of the drug, as measured by BUN, creatine, and phosphorous levels in the blood, reducing concerns that tacrolimus might cause nephrotoxicity in this model.
Perhaps most striking is the fact that these effects are specific to the prefrontal and hippocampal regions in both groups of dogs. Other than the drug protecting against global neurodegeneration in white matter (reflected as a decrease in FA in the control dogs, but not those treated with tacrolimus), the diffusion metrics in no other brain regions, except those considered a priori aging hotspots, changed after a year. This specificity and the fact that these protections are displayed before cognitive decline is promising. These results strongly support the potential of tacrolimus to prevent age-related pathologic decline and suggest that similar drugs could be used as middle-aged preventative care in humans. More research on the neurobiological mechanisms of calcineurin inhibitors would help to indicate a more specific time frame in the human life span in which these drugs could be most effective in preventing neuropathology.
The results from this study also suggest a compelling case for using higher-order diffusion imaging measures. NDI, ODI, and even tensor metrics like FA computed on multishell data, all show potential to be early biomarkers for aging-related pathology. They may be sensitive to microstructural alterations preceding other measurable pathologies and capture these changes well before gross atrophy or cognitive decline is present. Acquisition of higher-order, multishell data allows for both forms of analyses, and the complexity and the tissue-agnostic approach of NODDI (Zhang et al., 2012) makes it far more applicable to the study of gray matter microstructure and longitudinal change that may result from AD-associated neuropathological changes (e.g., inflammation and astrogliosis).
This study provides novel outcomes that include (1) evidence for treatment benefits of tacrolimus on brain structure before cognitive decline; and (2) support for a canine model that shows changes in NODDI metrics that can be detected both in both cross-sectional and longitudinal studies. However, this study is not without limitations. The advanced diffusion metrics, specifically NDI and ODI, have not been adequately histologically validated, and though some studies suggest that they might be sensitive to inflammation, these results must be interpreted cautiously. The male/female ratio prevents us from assessing sex differences, and these results may not be as significant in male beagles. However, all male dogs studied had diffusion metrics well within the range of their female counterparts, with no significant outliers. Moreover, dogs were middle aged without signs of significant cognitive decline, posing a challenge to detect structure–behavior relationships. Unfortunately, both groups showed a significant decrease in hippocampal volume after 1 year, suggesting that the drug may not be able to protect against more macrostructural atrophy. The study will continue for another year, and our hypothesis that structural brain changes occur before cognitive decline may be testable at the next time point. Also, future neuropathology outcome measures will help us determine whether our speculations regarding FA and Aβ, and NDI and glial activation/inflammation are valid.
In summary, treatment with low doses of tacrolimus in the canine model of aging protects against age-associated structural changes, as shown by neuroimaging and presents no observable adverse effects. It is intriguing to consider that the structural neuroimaging outcomes noted here may precede cognitive decline in control dogs and may predict benefits in treated animals, which will be evaluated as the study continues.
Footnotes
This study was funded by National Institutes of Health Grant R01-AG-056998, given to E.H. and C.M.N. We thank Aperus Pharma for supporting one arm of the baseline data of the study. We also thank Frederick Bresch at the University of Kentucky for support with the cognitive testing software.
The authors declare they have no conflict of interest.
- Correspondence should be addressed to Elizabeth Head at heade{at}uci.edu
References
In this issue
