Detection of memory- and learning- related brain connectivity changes following trace eyeblink-conditioning using resting-state functional magnetic resonance imaging in the awake rabbit

Abstract

Animal imaging studies have the potential to further establish resting-state fMRI (rs-fMRI) and enable its validation for clinical use. The rabbit subjects used in this work are an ideal model system for studying learning and behavior and are also an excellent established test subject for awake scanning given their natural tolerance for restraint.

We found that analysis of rs-fMRI conducted on a cohort of rabbits undergoing eyeblink conditioning can reveal functional brain connectivity changes associated with learning, and that rs-fMRI can be used to capture differences between subjects with different levels of cognitive performance.

rs-fMRI sessions were conducted on a cohort of rabbits before and after trace eyeblink conditioning. MRI results were analyzed using independent component analysis (ICA) and network analysis. Behavioral data were collected with standard methods using an infrared reflective sensor aimed at the cornea to detect blinks. Behavioral results were analyzed, and a median split was used to create two groups of rabbits based on their performance.

The cohort of rabbits undergoing eyeblink conditioning exhibited increased functional connectivity in cingulate cortex, retrosplenial cortex and thalamus consistent with brain reorganization associated with increased learning. Differences in the striatum and right cerebellum were also identified between rabbits in the top or bottom halves of the group as measured by the behavioral assay. Thus, rs-fMRI can provide not only a tool to detect and monitor functional brain changes associated with learning, but also to discriminate between different levels of cognitive performance.

Graphical Abstract

Mammalian brain can adapt to life experiences and reorganize its functional network to learn a new task. In our study we were able to detect and characterize rs-fMRI signals from alert rabbits to track the functional reorganization of the neural networks that mediate acquisition and retention of eyeblink conditioning.

Introduction

Pre-stimulus functional connectivity in the awake rabbit can be used to predict the behavioral response to a conditioning stimulus during a session of eyeblink conditioning (Schroeder, Weiss, Procissi, Wang, & Disterhoft, 2016b). The present experiment was done to determine if resting state connectivity could reflect persistent changes that occur during eyeblink conditioning, and the learning process in general.

One of the most salient features of the mammalian brain lies in its functional and structural plasticity which provides it with the exceptional ability to remodel and recover after trauma, injury, or insult (Carey, Nilsson, & Boyd, 2019; Maguire, Eleanor A., Katherine Woollett, 2006; Nasrallah, To, Chen, Routtenberg, & Chuang, 2016; Tavazzi et al., 2018). Being able to detect and monitor plasticity-related brain changes using non-invasive imaging could offer a longitudinal assessment and insight into CNS disorders and to aid therapeutic assessments (Barban et al., 2017). In this respect changes in brain connectivity and blood oxygen level dependent (BOLD) activity often occur before any anatomical and morphological structural changes in specific brain regions become detectable (L. Wang et al., 2006), and are ideal candidates for early predictors of CNS disease progression. Resting state fMRI (rs-fMRI), which relies on slow, non-task related fluctuations of brain activity offers the ability to detect whole brain functional connectivity and provides a direct way to capture regional functional brain activity changes that occur early during several neurodegenerative diseases. Moreover, advanced analysis of rs-fMRI data using network analysis and graph theory provides additional insight, not only into regional functional brain changes, but also on the complex set of network interactions among different brain components (K. Wang et al., 2007).

In this study we aimed to demonstrate how rs-fMRI techniques can help gain insight into the dynamic interplay between the natural positive functional plasticity of the brain and cognitive performance. We collected sets of rs-fMRI data on a cohort of rabbits undergoing trace eyeblink conditioning (EBC) with the purpose of comparing reorganization of brain functional connectivity between baseline and conditioned states. Trace EBC incorporates a stimulus free interval between the end of the conditioning stimulus (CS) and onset of the unconditioned stimulus (US). Lesions of the hippocampus or prefrontal cortex in the rabbit (Moyer, Deyo, & Disterhoft, 2015; Solomon, Vander Schaaf, Thompson, & Weisz, 1986; Weible, McEchron, & Disterhoft, 2000) prevent acquisition of trace EBC, but not delay conditioning (where the CS and US overlap and co-terminate in time), These results support the use of trace EBC as a simple cognitive task for the study of forebrain substrates of learning and memory.

Through group-ICA, dual regression, and network analysis we detected multiple relevant functional connectivity changes occurring with task acquisition (i.e. learning) as confirmed by the behavioral assay. More importantly, the results were able to differentiate between rabbits in the top and bottom halves of overall performance, in terms of their percent of conditioned responses. These results, together with the fact that the regions and networks we identified (default mode network, thalamus, motor cortex and cerebellum) have analogues in other species including humans, indicate that the awake rabbit is a good model system to explore the use of rs-fMRI. This suggests that the proposed methods have great potential for use as tools for development of novel therapeutic approaches.

Methods

Animals

All experimental procedures involving animals were approved by the Northwestern University IACUC. New Zealand White female rabbits (n=10) were obtained from Envigo and maintained in the Northwestern University vivarium under normal conditions. Since our objective was to test the reliability of rs-fMRI to predict differential learning and memory performance we selected a sub-group of animals from a set of aging retired breeder rabbits (32.5±0.6 months at time of arrival) that were beyond the 24 month threshold for age-related learning impairments (Thompson, Mover, & Disterhoft, 1996), and which were expected to exhibit variable acquisition rates. Those rabbits were also used for a diet-based study with the aim of maximizing the probability of heterogeneous learning rates (diet-based data not shown\discussed here). Each animal was implanted with a premade “headbolt” for head fixation during behavioral training and MRI. The headbolt was made of dental acrylic and encased the head of 4 nylon bolts (6–32 x ¾”) that fit into a crossbar within the restrainer. Head fixation during behavioral training allowed proper aiming of an infrared reflective optical sensor to transduce movement of the nictitating membrane (third eyelid) across the eye, and head fixation during imaging minimized motion artifacts during MRI sessions. The longitudinal set of experiments involved initial MRI sessions, four 5-day weeks of daily sessions of eyeblink conditioning (which further habituated the rabbits to restraint), and post-conditioning MRI sessions within 2–7 days (5.2 ± 1.9 days) after the completion of EBC. The general experimental scheme and its timeline are graphically depicted in Fig. 1 a.

FIG. 1:

Panel a) shows a scheme of experimental pipeline depicting the various steps and their timeline. Shown in panel b) are representative pictures of the rabbit in its holder with the rf coil fixed to the headbolt.

As reported previously (Schroeder, Weiss, Procissi, Disterhoft, & Wang, 2016) a one hour habituation session was sufficient to prevent movement during restraint and prevent struggling during initial swaddling of the rabbit. Rabbits were transported from the vivarium to the magnet in a small animal kennel. Rabbits were removed from the kennel and swaddled inside a snug cloth sack that tied at the neck and behind the rear of the animal. Foam earplugs were inserted into the ear canals and rabbits were placed in a prone position inside a Lomir Bunny Snuggle that was secured with Velcro by placing the animals inside the scanner with the relevant MR sequences running. Each rabbit underwent a one hour session for habituation to the MRI holder and environment (~ 1 week prior to image collection).

Surgery

Sterile surgery was performed to implant a restraining bolt assembly onto the skull to fix the head in the stereotaxic plane in our custom-built MR cradle. Methods were as described previously (Weiss, Procissi, Power, & Disterhoft, 2018).

Eyeblink Conditioning

Rabbits underwent 80 trials of tEBC each day. Conditioning was performed in a light and sound attenuating chamber (Med-Associates, Inc.). In order to minimize any contextual differences between the two training environments, rabbits were restrained in a similar manner and MRI scanner acoustic noise (~80dB) played inside the conditioning chambers to mimic the environmental stimuli of the MRI.

Trials consisted of a 250 ms conditioned stimulus (CS; whisker stimulation, ~100 μm dorsal-ventral deflections of Row B whiskers at 62.5 Hz) followed by a 500 ms, stimulus-free trace interval and a 150 ms unconditioned stimulus (US; corneal air puff). An intertrial interval (ITI; average = 45 s; range: 30–60 s) followed each trial. The right eyelid was held open with tailor hooks and an infrared emitter and detector was positioned ~1 cm in front of the right eye to detect blinks. Eyeblink data were sampled at 1 kHz. A computer running custom Labview software (National Instruments, Inc.) controlled stimulus delivery and behavioral data collection. Extension of the nictitating membrane (i.e., a ≥15 ms voltage increase 4 standard deviations above the mean of a 250 ms pre-CS baseline amplitude) prior to US onset was defined as a conditioned response (CR). Each conditioning session lasted approximately 1 hour. The percentage of trials with conditioned responses was used as a summary statistic for analysis. An increase in the percentage across training sessions is evidence of learning.

MRI Acquisition

Acquisition was performed using a 7T ClinScan MRI scanner (Bruker, Germany) equipped with a 20 cm inner diameter gradient coil (gradient strength = 300 mT/m; slew rate = 1040 T/m/s). A volume coil was used as a radio frequency transmitter with reception of the signal accomplished through a three-channel phased array coil custom designed to enable fixation of the head through an aperture on the top of the coil (RAPID MR International, Columbus, Ohio). The animal cradle was modified to integrate this 3-channel receiver coil using a crossbar fixed to the holder as shown in Fig. 1 b. Before each MRI session the rabbit was secured in a cotton sac, then wrapped in a Lomir Snuggle Sack and then placed in the acrylic cradle with the head locked in the crossbar using the implanted bolts. Body temperature and respiration were monitored throughout the experiment inside the magnet using the SA instruments physiological monitoring unit. The acquisition protocol included a coronal 3D-GRE multi-echo scan (TR = 68 ms; Echos = 2.7, 6.83, 11.26, 16, 10.13, 25 ms; Flip Angle = 15; voxel size = 0.29 × 0.29 × 0.5 mm³; FOV = 29.6 × 55.6 × 24 mm³) and a transverse EPI (TR = 1800 ms; TE = 25 ms; Flip Angle = 70; voxel size = 0.65 × 0.65 mm²; slice thickness = 1.5 mm; number of slices = 20; matrix = 52 × 68; GRAPPA = 2; echo spacing = 0.25 ms; volumes = 500). The rs-fMRI EPI sequence was repeated twice. Repositioning of the same animal was achieved in all three directions (X, Y, and Z) with great accuracy (<500 um) across sessions (Schroeder, Weiss, Procissi, Disterhoft, et al., 2016), as also confirmed by post-acquisition analysis.

Behavioral Data Analysis

Behavioral data were dichotomized to the first and last five days of training. T-tests were performed between the average percent CRs on the first 5 days versus the average percent CRs on the last 5 days to assess task learning. Using a threshold of CRs > 50 % over the last 5 days average we split the rabbits into two groups of 5 animals and a second t-test was run to assess difference between the top 5 ranked and bottom 5 ranked rabbits.

Resting state fMRI Processing and Analysis

Data analysis was performed using FMRIB Software Library version 6.0 (Analysis Group, FMRIB, Oxford, UK), FSLNets 0.6 and MatLab R2017a (The Mathworks Inc).

A high-resolution 3D GRE image of each brain was generated from the average of 6 echos. The 3D images from each rabbit were co-registered (non-rigid 12 degree-of-freedom transformation) and averaged to generate a rabbit brain template.

Visual inspection of EPI volumes was performed to ensure that no ghosting artifacts were present inside the brain image. For each fMRI acquisition we computed the average absolute head displacement from a reference volume position (middle volume) calculated as the Euclidian norm of the translational and rotational rigid-body movements and the better of the two fMRI scans acquired for each rabbit was selected. No sequence used in the analysis shows an absolute average displacement > 1.5 mm. For the resting state analysis, EPIs were first pre-processed: i) all volumes were registered by a rigid transformation to the central volume and motion regressors were removed from the data to limit the effect of motion artifacts, ii) a brain mask was generated starting from the bias field corrected mean of the volumes and used as an inclusive mask for EPI, iii) a common origin was selected for all subjects’ EPIs, iv) the time course was high-pass filtered with a threshold of 0.02 Hz and vi) images were smoothed using a 0.7 mm gaussian kernel. Functional volumes were registered to high-resolution 3D images and then to the common template before ICA. ICA group-analysis was run using a multi-session temporal concatenation that pre-selected 30 desired components, and the output was inspected to identify resting state and spurious components. By dual-regression analysis (Erhardt et al., 2011) (corrected for multiple comparison using 5000 permutations) of the 30 ICs spatial maps we tested functional connectivity differences before and after training, and between rabbits assigned to the top and bottom halves of performance, based on the percent of conditioned responding.

Network analysis was performed on time courses resulting from stage 1 of dual regression analysis to assess correlations among the resting state components across all rabbits. For each rabbit we computed the correlation between the rs-fMRI time course of each pair of components and a matrix of Pearson correlation coefficients was calculated for every animal. The correlation matrices were transformed using Fisher z-transform to make the values normally distributed and a cross-subject GLM analysis was performed to explore differences among correlation values between groups (5000 permutations multicomparison correction).

One sample t-tests were run on z-transformed correlation matrices within each group to compute p-values group matrices. An adjacency matrix for each group was then generated thresholding p-values matrices (p_uncorrected < 0.05) and consequently the graph plot representing brain functional networks. In the graphs the dots (vertices) represent the functional components and the lines (edges) the connections between them.

Results

The behavioral data showed that the average percentage of conditioned responses increased across the days of conditioning (Fig. 2 a). The results were dichotomized to the first and last five days of training to create a group of rabbits that learned the task rapidly, and a group that learned the task more slowly. In order to relate changes in learning to changes in MR signal, we compared the percentage of CRs during the last 5 days of training to the percentage of CRs during the first five days of conditioning. The percentage of responses increased significantly across sessions (53.3±18.5 vs 17.6±10.5, p<0.05) (Fig. 2 c). Using a threshold of conditioned responses > 50 % we also identified subgroups of 5 rabbits with significantly different performance at the end of EBC training (78.6 ± 4.8 % vs 38 ± 12.7 % CRs, p < 0.05) (Fig. 2 b). The repeated measure ANOVA confirmed a statistically significant effect of sessions in learning (F(19, 152) = 16.5, p = 4.8×10–28) and a significant group-time interaction (F(19, 152) = 6.2, p = 1.4×10–11). There was no significant difference in the number of days between the end of EBC training and MRI acquisition between the top 5 ranked and lower 5 ranked rabbits (5.8±1.6 vs 4.6 ± 2 days). This result was the basis for using a median split of the data to compare the rs-fMRI ICA findings.

FIG. 2:

Panel a) shows the average CRs for the two groups (top 5 ranked and Lower 5 ranked rabbits) for each timepoint. Shown in panel b) are the box plots of the percentage of conditioned responses averaged over the last 5 days for the 5 rabbits performing at > 50 % CRs (top 5 ranked) or <50% CRs (lower 5 ranked). Shown in panel c) are the box plots of the percentage of conditioned responses averaged over the first 5 days of training (Early Conditioning) and over the last 5 days of training (Late Conditioning). On each box, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers (> 2 standard deviation), and the outliers are plotted individually using the ‘ + ‘ symbol.

Among 30 independent components (ICs) resulting from group ICA analysis including all animal scans, we identified 10 functional brain components (Fig. 3); the remaining components were related to CSF, white matter, or physiological/ scanner artifacts and were discarded. T-contrasts following dual regression analyses showed significant differences (p_corrected < 0.05) when comparing the groups before and after learning in the cingulate cortex, retrosplenial cortex and thalamus (Fig. 4).

FIG. 3:

The 10 functional brain components maps identified from the group ICA analysis run on all fMRI acquisitions (all rabbits and both timepoints). The ICs are shown with Z-score calibration bar for: cingulate cortex, retrosplenial cortex, occipital cortex, hippocampus, thalamus, right motor cortex, left motor cortex, striatum, cerebellum left, and cerebellum right.

FIG. 4:

Group differences obtained by dual regression analysis for the t-contrast of after conditioning – before conditioning. The p-value is corrected for multiple sample comparison (5000 permutations).

Statistical analysis on correlation matrices showed significant differences (p_corrected < 0.05) in the retrosplenial/cingulate and motor/retrosplenial nodes (connection between two components) before and after conditioning (Fig. 5 a,b,c). Visual inspection of the correlation matrices and graphs shows a connected network with a larger number of edges for the group of animals after the conditioning process as compared to the same group before conditioning, i.e. fewer edges and 2 isolated vertices are seen prior to learning (Fig. 6 a). The main difference observed in the brain network between the top 5 and lower 5 ranked rabbits is the higher degree of connectivity for right cerebellum and striatum (Fig. 6 b).

FIG. 5:

The p-values matrix in panel a) shows the results from t-tests run on the correlation matrices (for each node) comparing the rabbits after and before conditioning. The values on the diagonal representing the self-correlation of the components, are artificially set to 1. The box plots in panels b) and c) show the Fisher z-transformed correlation values of the two nodes (connections between two functional brain components) resulting significantly different (p_corrected < 0.05) before and after eyeblink conditioning.

FIG. 6:

Matrices of p-values obtained from within groups one sample t-tests on correlation matrices, and below them the corresponding graph obtained from the adjacency matrices generated by p-values thresholding (p < 0.05). Panel a) depicts the matrices and corresponding graphs representing the functional brain networks before and after conditioning for all 10 subjects grouped together, regardless of behavioral performance. Panel b) depicts the matrices and corresponding graphs representing the functional brain networks for the bottom 5 learners (right) and the top 5 learners (left) in regard to eyeblink conditioning. In the graphs the dots (vertices) represent the functional components (Cg = Cingulate cortex, Rs = Retrosplenial cortex, Occ = Occipital Cortex, Hip = Hippocampus, Tha = Thalamus, Motror R = Right Motor Cortex, Motor L = Left Motor Cortex, Str = Striatum, Cereb L = Left Cerebellum, Cereb R = Right Cerebellum) and the lines (edges) represent the connections above the threshold value between them. The size of the vertices is proportional to the number of connections of the component with the others. The values on the diagonal of the matrices, representing the self-correlation of the components, are artificially set to 1.

Discussion

In this study we employed a rabbit model of learning and memory which is ideally suited for awake functional MRI studies (Schroeder, Weiss, Procissi, Disterhoft, et al., 2016). Previously we demonstrated how task-based pretrial fMRI activity can be used to predict the upcoming behavioral performance of each subject (Schroeder, Weiss, Procissi, Wang, et al., 2016b). In this study we demonstrated how non-task based rs-fMRI can predict connectivity changes in the brain as a consequence of learning. The current results extend the earlier findings by suggesting that rs-fMRI can predict and differentiate between different degrees of learning performance. By investigating these changes using eyeblink conditioning as the learning paradigm we were able to observe neuroplasticity-related alterations in a wide network of brain regions that interact during memory and learning tasks (Weiss & Disterhoft, 2011). By comparing rs-fMRI data before the rabbits underwent behavioral training with rs-fMRI data after the end of conditioning sessions, we were able to detect several differences involving specific brain components and a more general reorganization of the functional brain network, and to correlate those results with the behavioral data. Specifically, we found significantly enhanced spontaneous neural activity in cingulate cortex, retrosplenial cortex and thalamus following learning sessions. This increased activity in the cingulate and retrosplenial cortex is interesting because these two brain areas are part of the rabbit default mode network (DMN) (Schroeder, Weiss, Procissi, Disterhoft, et al., 2016), which is made up of several regions that have been shown to exhibit reduced functional connectivity during task performance and which are related to intrinsic functional brain network organization.

The DMN exists in humans as well as in non-human primates, cats and rodents (Lu et al., 2012; Raichle, 2015) and is impaired by several neurodegenerative diseases (Agosta, Galantucci, & Filippi, 2017). In fact, a neuroimaging study comparing early Alzheimer disease patients to elderly healthy subjects concluded that the DMN may be closely involved in episodic memory processing (Greicius et al., 2016). Our experimental findings (i.e. increased cingulate cortex activity in rabbits after conditioning) seems to partially confirm the role of the DMN in long-term functional brain reorganization after acquiring a cognitive memory (Teixeira, Pomedli, Maei, Kee, & Frankland, 2006). Note, that we did not observe differences in the somatosensory or cerebellar cortex, as was observed when BOLD activity was recorded during active conditioning sessions (Miller et al., 2003; Schroeder, Weiss, Procissi, Wang, & Disterhoft, 2016a). In one study investigating increased brain connectivity after maze learning activities in rats (Nasrallah et al., 2016), similar results showed strong differences the day after training. These differences were observed to fade away after ~5 days, but in agreement with our study, the enhancement in cingulate cortex outlasted the connectivity changes by several days.

The other component with enhanced activity after conditioning was the thalamus, which operates as a integrative hub in the mammalian brain (Hwang, Bertolero, Liu, & D’Esposito, 2017). Along with the structures that comprise the DMN region, the thalamus is an important brain area given its involvement in several types of neuropathology such as migraine and depression (Wang et al., 2016; Kang et al., 2018).

By conducting network analyses on the time courses of functional brain components we were able to explore global effects, and we showed, as already observed in other studies (Lewis, Baldassarre, Committeri, Romani, & Corbetta, 2009; Nasrallah et al., 2016), that the process of learning is accompanied by a general reorganization of functional brain networks, with enhanced network connectivity. The strongest increase in conditioning related node connectivity was detected in the cingulate cortex, the retrosplenial cortex, and in the striatum. This is consistent with the conditioning-induced functional reorganization of neuronal responsiveness in the anterior cingulate gyrus that we detected with single-neuron recording (Hattori, Yoon, Disterhoft, & Weiss, 2014). More importantly, we identified a difference in degree of connectivity between the superior sub-group of rabbits that learned well as compared to rabbits that did not learn as well based on the behavioral assay. This result validates the described method as a reliable and quantitative tool that is related to cognitive performance. The network analysis results were consistent with the dual regression analysis results in the case of the of cingulum and retrosplenial cortex, but particularly interesting was the observation of an increased number of functional connections in the striatal node, which is also part of rodent default mode network (Lu et al., 2012). This increase suggests its centrality and role in the functional plasticity of the resting brain, and supports the involvement of the caudate in trace eyeblink conditioning, as we have shown previously (Flores & Disterhoft, 2013, 2009). Review of the statistical correlations among different nodes revealed a significantly increased correlation between the left motor cortex and retrosplenial cortex following learning which is probably driven by the fact that whisker vibration and corneal air puffs delivered exclusively to the right side of the face were used as the CS and US during behavioral conditioning. This same laterality is also detected in the network comparison between the top and bottom half of learners looking at the increased connectivity of the right cerebellar node, further strengthening our hypothesis that the described imaging approaches do detect differences in cognitive abilities (note that both the inputs and outputs of the cerebellum decussate to the forebrain, so that the ipsilateral side of the body is affected) (Miller et al., 2003).

Functional MRI is very sensitive to motion artifacts requiring most animals to be anesthetized or physically restrained to acquire stable images. Unfortunately, anesthesia, prolonged restraint during acquisitions and habituation sessions, and the MRI scanner noise negatively affect most animals, altering results in unpredictable ways (Paasonen, Stenroos, Salo, Kiviniemi, & Gröhn, 2018; Reed, Pira, & Febo, 2013). Our proposed preclinical model includes the advantage of an easier implementation of awake imaging due to the rabbits’ innate habit to live in burrows, and to become immobile in the presence of a perceived threat. This makes the rabbit particularly suitable for MRI studies that allow shorter habituation sessions to the MRI procedures, as compared to other animals (Schroeder, Weiss, Procissi, Disterhoft, et al., 2016; Weiss et al., 2018) and ultimately provides a more ideal model to test translational neuroimaging approaches and analysis methods for clinical use.

In our study we employed a data-driven ICA approach, often used in similar human studies, to identify the network nodes instead of using seed-based ROIs. The reason for this choice relies on the well-established use of ICA for decomposing functional brain components and on the fact that functional ROIs could be inaccurately identified because of imperfect co-registration between anatomical and EPI images, and of the nontrivial overlap between functional components and anatomical structures (Smith et al., 2011).

The translational impact of our current results is further supported by the consistency of the spontaneous BOLD fluctuations detected by rs-fMRI among mammals (Pawela et al., 2008) and the similar network topography between rodents and humans (Liang, King, & Zhang, 2011). Importantly, our work was conducted on the same group of subjects which underwent conditioning with the rabbits being scanned twice during the study, and each rabbit served as its own control suggesting the ability to differentiate between subjects with different levels of cognitive performance (i.e. learning ability).

Preclinical MR neuroimaging has provided information on altered physiological, functional, and anatomical brain features which inevitably accompany several types of neurodegenerative conditions (Albanese, Greco, Auletta, & Mancini, 2018; Katsuno, Sahashi, Iguchi, & Hashizume, 2018). Thus, MR neuroimaging has the potential to enable early diagnosis and allows for longitudinal tracking of CNS disease progression in both laboratory animals and patients by identifying relevant and quantitative imaging biomarkers associated with diseases related to pathophysiological processes. This is an invaluable feature with high translational potential particularly when one considers that initial changes underlying the disease process likely become manifest years before onset of observable neurological deficits which lead to inevitable negative outcomes and missed opportunities for preventive or ameliorative therapies (Sancesario & Bernardini, 2018). In this framework, in our study we showed the important ability to detect cognitive capabilities from network connectivity data.

We expect to employ these methods to study differences between cohorts of animals with clinically relevant differences such as Alzheimer-like conditions with the aim of investigating how the observed and described functional reorganization of brain networks may reflect brain plasticity. Based on these results, we suggest that the described methods could potentially be used in humans to capture network connectivity alterations occurring early during progression of neurodegenerative diseases and possibly preceding even mild cognitive impairment. Such an approach which might serve as a biomarker for individuals that would benefit with therapeutic interventions as they become available.

We did not include a group of rabbits that were given random unpaired presentations of the conditioning stimuli. Although such a group is sometimes used to test for non-associative pseudoconditioning, the percentage of trials that exhibit “CRs” under those conditions is low. We have used pseudoconditioning control rabbits to demonstrate that physiological changes mediating EBC are learning specific, rather than training specific (Hattori, Chen, Weiss, & Disterhoft, 2015), but the current study was designed to compare the same group of rabbits before and after conditioning. The within animal control did not require a group of pseudoconditioned animals to support the conclusions.

Conclusions

In conclusion, the awake rabbit provides an excellent animal model system in which to observe learning related changes in resting state brain functional connectivity. An overall reorganization of brain connectivity occurs with learning, is detectable at individual components of the network, and can differentiate between subjects showing different degrees of learning capability. This latter result suggests that the described method may be useful for investigating the pathogenesis of neurodegenerative diseases and responses to treatments in preclinical rabbit models and it may be translated to human patients.