Adult female Sprague Dawley rats (n = 68), 90–100 days of age were purchased from Charles River Laboratories (Wilmington, MA, United States). Rats were kept in Plexiglas cages, with two rats per cage, and maintained at a room temperature of 22–24°C under a 12:12 reverse light:dark cycle, with lights turning off at 09:00 h. All experimental activities were conducted between 10:00 and 18:00 h under dim red illumination to prevent disturbances due to transitions between light–dark cycles. The rats had access to food and water without restriction. All animal acquisition and care procedures adhered to the guidelines outlined in the NIH Guide for the Care and Use of Laboratory Animals. All the methods and procedures described in this study had been approved in advance by the Northeastern University Institutional Animal Care and Use Committee, with the protocol assigned the number 20-0626R. Northeastern University’s animal care and housing facilities have full accreditation from AAALAC, International. The study protocols employed in this research followed the ARRIVE guidelines for reporting in vivo experiments in animal research as recommended by Kilkenny et al. (2010). Animals were monitored daily over the duration of the study for general health, food and water consumption. A 15% loss in body weight was set as a humane endpoint.

PEA was provided by RAD Wellness (Valleyview, TX). On the day of imaging PEA was taken up in a solution of 2% gum Arabic in water. To deliver drug remotely during the imaging session, a poly-ethylene tube (PE-20), approximately 30 cm in length, was positioned in the peritoneal cavity. There were four experimental groups: vehicle, 3.0 mg, 10 mg, and 30 mg/kg PEA. The range of doses of PEA were taken from the literature (Jaggar et al., 1998; Conti et al., 2002; Farquhar-Smith and Rice, 2001; Costa et al., 2002; Ahmad et al., 2012). Rats were randomly assigned to one of four experimental groups, six per group. Due to motion artifact five rats were excluded from the study: two from vehicle, two from 3.0 mg PEA and one from 30 mg PEA. The final subject distribution was: (1) gum Arabic vehicle, n = 4; (2) 3.0 mg PEA, n = 4; (3) 10 mg PEA, n = 6; and (4) 30 mg PEA, n = 5. The behavioral studies included 24 rats again randomly divided into four experimental groups of six each, while the lipidomics studies included 20 rats randomly divided into four groups of five in each. These rats were independent of those used for imaging.

To mitigate the stress associated with head restraint, rats underwent an acclimation protocol to familiarize them with the restraining system, which consisted of a head holder and body tube. This system was designed to include a cushioned head support, eliminating the need for ear bars and, in turn, reducing discomfort to the animals while minimizing any unintended motion artifacts. These acclimation sessions were conducted daily for four to five consecutive days. During these sessions, rats were briefly anesthetized with 1–2% isoflurane and securely placed into the head holder, with their forepaws fastened using surgical tape.

Once fully conscious, the rats were positioned within an opaque black box, essentially a “mock scanner,” for 30 min. Inside the mock scanner, a tape recording of the MRI pulse sequence was played to simulate the environment of the magnet bore and the imaging protocol. It is worth noting that significant decreases in respiration, heart rate, motor activity, and plasma corticosterone levels were observed when comparing the first and last acclimation sessions, as reported by King et al. (2005). This reduction in autonomic and somatic indicators of arousal and stress contributed to the enhancement of signal resolution and image quality.

The effectiveness of this passive restraining system can be evaluated by the minimal levels of motion artifacts observed during the 20-min imaging sessions. Supplementary Figure S1 illustrates the average displacement in all orthogonal directions throughout the entire 20-min scanning period for all rats, whether they received PEA or a vehicle (n = 19). At no time point did the average motion exceed 200 μm in any orthogonal direction.

Five to six rats were imaged in a day. Each day had a mix of the different experimental groups known by all the investigators. A detailed description of the awake rat imaging system is published elsewhere (Ferris et al., 2014). Notably, we used a quadrature transmit/receive volume coil (ID = 38 mm) that provided both high anatomical resolution and high signal-to-noise ratio for voxel-based BOLD fMRI (Ekam Imaging Boston MA). The design of the coil provided complete coverage of the brain from olfactory bulbs to brain stem with excellent B1 field homogeneity. A video of the rat preparation for imaging is available at www.youtube.com/watch?v=JQX1wgOV3K4. The experiments were carried out with the use of a Bruker Biospec 7.0 T/20-cm USR horizontal magnet, accompanied by a 2 T/m magnetic field gradient insert with an inner diameter of 120 mm, capable of a rapid 120-μsec rise time. At the outset of each imaging session, a high-resolution anatomical dataset was obtained employing the rapid acquisition, relaxation enhancement (RARE) pulse sequence (RARE factor 8). This involved acquiring 25 slices, each with a thickness of 1 mm, within a field of view (FOV) of 3.0 cm². The data matrix used was 256 × 256, with a repetition time (TR) of 3 s, an echo time (TE) of 12 milliseconds, and an effective TE of 48 milliseconds. The number of excitations (NEX) was set to 3, resulting in an acquisition time of approximately 4.48 min, and an in-plane resolution of 117.2 μm².

Functional images were captured using a multi-slice Half Fourier Acquisition Single Shot Turbo Spin Echo (HASTE) pulse sequence. This involved collecting 22 slices, each 1.1 mm thick, with the same FOV of 3.0 cm². The data matrix was 96 × 96, with a TR of 6 s, TE of 3.75 milliseconds, and an effective TE of 22.5 milliseconds. The acquisition took approximately 20 min, with an in-plane resolution of 312.5 μm². It should also be emphasized that high neuroanatomical fidelity and spatial resolution are critical in identifying distributed neural circuits in any animal imaging study. Many brain areas in a segmented rat atlas have in-plane boundaries of less than 400 μm2 and may extend for over 1,000 μm in the rostral/caudal plane. With the development of a segmented, annotated 3D MRI atlas for rats (Ekam Solutions, Boston, MA, United States), it is now possible to localize functional imaging data to precise 3D “volumes of interest” in clearly delineated brain areas. For example, this spatial resolution was sufficient to identify the bilateral habenula of the thalamus, with approximately 4–5 voxels on each side, but not to differentiate between the lateral and medial habenula (Kawazoe et al., 2022). Using the spin-echo technique for functional imaging was essential to ensure high anatomical accuracy for data registration to the rat MRI atlas.

Each functional imaging session entailed continuous data acquisition for 250 scan repetitions, covering a total time of 25 min. The first 5 min of this period constituted a baseline, consisting of the initial 50 image acquisitions, followed by the administration of either a vehicle or a drug, and then another 200 acquisitions over the subsequent 20 min. The 20 min imaging protocol is standard for lipid soluble drugs that rapidly cross the blood brain barrier (Sadaka et al., 2021). The order of drug doses was randomized across the scanning sessions.

Approximately 5 min after the 20-min functional imaging scan the rats were imaged for connectivity. The functional connectivity images were collected over a 15-min period. These functional connectivity scans were acquired using a spin-echo triple-shot EPI sequence with specific imaging parameters: a matrix size of 96 × 96 × 20 (height × width × depth), TR/TE of 1,000/15 milliseconds, voxel size of 0.312 × 0.312 × 1.2 mm, slice thickness of 1.2 mm, 150 repetitions, and a total acquisition time of 15 min.

The fMRI data analysis consisted of three main steps: pre-processing, processing, and post-processing. All these steps were executed using SPM-12 (available at https://www.fil.ion.ucl.ac.uk/spm/). In the pre-processing stage, several operations were performed, including co-registration, motion correction, smoothing, and detrending. Co-registration was carried out with specific parameters: Quality set at 0.97, Smoothing at 0.6 mm, and Separation at 0.4 mm. Additionally, Gaussian smoothing was applied with a Full Width at Half Maximum (FWHM) of 0.8 mm.

The processing step involved aligning the data to a rat atlas, followed by segmentation and statistical analysis. To achieve registration and segmentation, all images were initially aligned and registered to a 3D Rat Brain Atlas^©, which included 173 segmented and annotated brain regions. This alignment was performed using the GUI-based EVA software developed by Ekam Solutions (Boston, MA). The image registration process encompassed translation, rotation, and scaling adjustments, performed independently in all three dimensions. All spatial transformations applied were compiled into a matrix [Tj] for each subject. Each transformed anatomical pixel location was tagged with its corresponding brain area, resulting in fully segmented representations of individual subjects within the atlas.

In the equation, Pi represents the p-value derived from the t-test conducted at the i-th pixel within the region of interest (ROI), comprising V pixels, with each pixel ranked according to its probability value. For our analysis, we set the false-positive filter value q at 0.2, and we fixed the predetermined constant c(V) at unity, following a conservative approach for assessing significance (as per Sathe et al., 2023). Pixels that achieved statistical significance retained their relative percentage change values, while all other pixel values were set to zero. Our analysis employed a 95% confidence level, two-tailed distributions, and assumed heteroscedastic variance for the t-tests.

To create composite maps displaying the percent changes in the Blood Oxygen Level Dependent (BOLD) signal for each experimental group, we mapped each composite pixel location (in terms of rows, columns, and slices) to a voxel within the j-th subject using the inverse transformation matrix [Tj]-1. A trilinear interpolation method was used to determine the contribution of subject-specific voxel values to the composite representation. The use of inverse matrices ensured that the entire composite volume was populated with subject inputs. The average of all contributions was assigned as the percent change in the BOLD signal at each voxel within the composite representation of the brain for the respective experimental group.

In the post-processing phase, we compared the number of activated voxels in each of the 173 brain regions between the control and PEA doses using a Kruskal–Wallis test statistic. The data were ranked in order of significance, as detailed in Table 1. We generated probability heat maps, depicted in Figure 1, showing brain areas with significant differences when comparing two or more groups. The saturation of the pink color in these maps indicates a lower p-value, signifying higher confidence in the observed differences in those brain areas.

The dose-dependent effect of PEA on brain activity was quantified by assessing positive and negative percent changes in the BOLD signal relative to the baseline. Initial analyses of signal changes in individual subjects compared image acquisitions 150–200 to the baseline period of 5–45. The statistical significance of these changes was evaluated for each voxel, approximately 15,000 per rat in their original reference system, using independent Student t-tests. A threshold of 1% was applied to account for normal fluctuations in the BOLD signal in the awake rodent brain. To mitigate false positive detections resulting from multiple t-tests, a control mechanism was introduced to maintain the average false positive detection rate below 0.05, as defined by a specific formula.

We conducted all statistical analysis for the graph theory assessment using GraphPad Prism version 9.1.2 for Windows (GraphPad Software, San Diego, California United States). To decide whether parametric or non-parametric assumptions were appropriate for different group subregions, we performed normality tests. We used Shapiro–Wilk’s tests to assess the normality assumption. Subregion degree centrality p-values exceeding 0.05 were considered to exhibit a normal distribution. Once the normality assumptions were confirmed, we employed paired t-tests to compare the degree centrality between the PEA and vehicle groups in various subregions. In cases where there was evidence against the normality assumption, we conducted a non-parametric Wilcoxon signed-rank (WSR) test.

Tail flick was used to assess peripheral nociception. Rats were acclimated to a restrainer that restricts movement while also exposing the tail. The tail was placed on a hot plate set to 50°C. The time it took for the rat to flick its tail was recorded, with a ceiling time of 10 s. Tail flick was conducted within the first hour. of PEA treatment. Each measure for the four experimental groups was compared with a one-way ANOVA using GraphPad Prism.

Plasma and whole brains were shipped on dry ice to the Bradshaw lab where they were stored at −80°C until processed. Brains were thawed for 5 min on an ice-cold dissection plate, dissected into 9 brain regions, individual brain areas flash frozen in liquid nitrogen, then stored at −80°C as previously described. Plasma samples (75 μL) and brain areas [hypothalamus (HYP); cerebellum (CER), thalamus (THAL), right side anterior cortex (CTX), hippocampus (HIPP), and striatum (STR)] were processed as previously described (Bradshaw and Johnson, 2023; Leishman et al., 2019). In brief, methanolic extracts were partially purified using C18 solid phase extraction columns (SPEs; Agilent, Santa Clara, CA, United States). Final elutions (i.e., fractions) of 65, 75, and 100 percent methanol were collected and stored at −80°C until MS analysis.

Methanolic elutions were analyzed as previously described (Leishman et al., 2019) with the exception that the API 7500 (Sciex, Framingham, MA 01701, United States) was used for analysis instead of the API 3000. The API 7500 is coupled to a Shimadzu LC system LC-40DX3 (Kyoto, Japan). Examples of chromatograms of PEA standards as well as plasma and CNS samples from this instrumentation are illustrated in Supplementary Figures S2, S3. All parent and fragment pairs are identical to those previously described from the Bradshaw lab (Tortoriello et al., 2013). Standard curves were generated by using purchased standards (Cayman Chemical, Ann Arbor, MI, United States), and those made in-house were validated through NMR and MS analysis as previously described (Tan et al., 2006). Sicex Analyst peak matching software, Analyst (Sciex, Framingham, MA 01701, United States) was used to validate standard peaks and sample peaks.

Statistical analyses for the plasma PEA dose curve were completed in IBM SPSS Statistics 29 (Chicago, IL, United States). One-way ANOVAs followed by Fisher’s Least Significant Difference post-hoc analyses were used to determine statistical differences between the average concentration of PEA measured in the plasma samples. Analysis of individual endogenous lipids in plasma and individual brain areas comparing the levels of a specific lipid in vehicle or after 30 mg/kg PEA injection were analyzed using Students t-tests set to 2-tails and Type 2. For the CNS combined analyses, levels of an individual lipid measured in the six different brain regions were summed to generate an overall value for each subject, therefore, the number of subjects remained the same at n = 5 each and analyzed via Students t-tests set to 2-tails and Type 2.

Samples with an endogenous lipid concentration outside of 2 standard deviations from the group mean were omitted from statistics for that compound. Statistical significance for all tests was set at p < 0.05, and trending significance at 0.05 < p < 0.10. Descriptive and inferential statistics were used to create heatmaps for visualizing changes in the concentration of each lipid analyte for every condition as previously described (Leishman et al., 2016). Briefly, the direction of changes for each analysis group compared to are depicted by color, with green representing an increase and orange representing a decrease. Level of significance is shown by color shade, wherein p < 0.05 is a dark shade and 0.05 < p < 0.1 is a light shade. Direction of the change compared to vehicle is represented by up (increase) or down (decrease) arrows. Effect size is represented by the number of arrows, where 1 arrow corresponds to 1–1.49-fold difference, 2 arrows to a 1.5–1.99-fold difference, 3 arrows to a 2–2.99-fold difference, 4 arrows a 3–9.99-fold difference, and 5 arrows a difference of tenfold or more (Stuart et al., 2013). Supplementary Figure S4 illustrates how these analyses are applied to the heatmaps. An abbreviation of ‘BDL’ indicates that the lipid concentration that was present in the sample was below the detectable levels of our equipment while ‘BAL’ indicates below analytical levels. Finally, bar graphs which show data as mean ± SE mean were made using GraphPad Prism Software (La Jolla, CA, United States) for key lipid families.

Figure 2 presents the dose-dependent variations (mean ± SD) in various behavioral measures recorded in the open field during a 5-min observation period. There was a significant dose-dependent decrease in distance traveled [F_{(3, 20)} = 9.02, p = 0.0006]. The highest 30 mg/kg dose of PEA was significantly less than vehicle (p < 0.0001), 3.0 mg/kg and 10 mg/kg (p < 0.05) treatments. Time spent along the walls was significantly different across treatments [F_{(3, 20)} = 7.47, p = 0.0015]. The 3 mg/kg dose was significantly less than vehicle (p < 0.05), while the high dose of 30 mg/kg showed the lowest mean time of 39 s and was significantly less than vehicle (p < 0.01) and the 10 mg/kg dose (p < 0.05). Time spent in the corner was also significantly different across treatments [F_{(3, 20)} = 2.46, p = 0.0918]. In this case, the high 30 mg/kg dose spent 260 s out of the total 600 s observation period standing in the corners of the open field box, a duration greater than vehicle (p < 0.01) and the 10 mg/kg dose (p < 0.05). There were no differences in time spent in the center (F = 2.466, p < 0.0918). Indeed, all rats, with the exception of two, spent less than 10/600 s exploring or crossing the center of the open field.

Shown in Figure 3 are results (mean ± SD) from the Novel Object Recognition test assessing memory (Figure 3A), and tail flick (Figure 3B) to the stimulus of hot water assessing pain sensitivity. The investigation ratio or time spent in the vicinity of the novel object over the 5 min observation period was only greater than chance (0.50 horizontal line) in vehicle treated rats (p < 0.05) as determined by a one-sample two-tailed t test. Rats treated with the different doses of PEA failed to perform better than chance. Indeed, one of the rats in the high dose 30 mg/kg group had a zero investigation ratio as it failed to attend to the novel object and just sat in a corner for the 600 s observation period. This rat was omitted from the group analysis. With respect to pain sensitivity, there were no significant differences between treatments. All rats, with the exception of one, withdrew their tail under 15 s.

Table 1 is a list of brain areas (Mean ± SE) that showed a dose-dependent change in negative volume of activation, i.e., number of negative BOLD voxels. The brain areas are ranked in order of their significance using a critical value of p < 0.05. Shown are their p-value and effect size given as omega square (Ω Sq). The false discovery rate (FDR) was p = 0.079. There were 69/173 brain areas shown to be significantly different with a Kruskal Wallace multiple comparisons test. For all PEA doses there is an increase in negative volume of activation over vehicle. When comparing the areas (mean values), note that many show an inverse dose–response, i.e., the 3 mg/kg dose has the highest voxel numbers and the 30 mg/kg dose the lowest. See for example, the ventral medial striatum, accumbens core and retrosplenial ctx. Certain brain regions like the thalamus, somatosensory cortices, hippocampus and basal ganglia containing the striatal areas and accumbens, are very sensitive to PEA (Figure 1).

Table 2 lists brain areas that showed a significant change in positive volume of activation, i.e., number of positive BOLD voxels in response to different doses of PEA. In this case there were only 26/173 brain areas that showed a decrease in positive volume of activation (FDR; p = 0.030). In all PEA doses the positive voxels were less than vehicle with the exception of the entorhinal ctx. The response to PEA was primarily blunting of the positive BOLD signal as many areas showed zero to few numbers of activated voxels such as the somatosensory ctx, anterior thalamus, and prelimbic ctx. The decrease in positive BOLD and increase in negative BOLD (Table 1) would suggest PEA is reducing activity in these brain areas. Tables for all 173 brain areas, negative and positive volume of activation, are provided in Supplementary material.

Figure 1 shows the anatomical location of the brain areas listed in Table 1 for negative BOLD volume of activation presented as 2D statistical heat maps. The coronal sections are labeled (a) through (h) and arranged from rostral (top) to caudal (bottom). Areas in blue are significantly different between VEH and 3 mg/kg dose of PEA. Areas in yellow denote the location of white matter tracts. The annotation is organized to report brain regions on the left and specific brain areas on the right. Brain section (a) shows no activity in the olfactory bulb. Section (b) highlights the prefrontal cortex comprised of the anterior cingulate, prelimbic, infralimbic and orbital cortices. Sections (c-e) highlight the many areas of the primary somatosensory ctx. Section (c) highlights the basal ganglia comprised of all striatal zones, i.e., dorsal medial, dorsal lateral, ventral medial, and ventral lateral together with the accumbens. Section (d) highlights the thalamus noting the ventral posterior lateral (VPL), posterior, medial dorsal and ventrolateral thalamic area all of which project to the primary somatosensory cortices. Sections (d-e) highlight the hippocampus noted by CA1, CA3, and dentate gyrus. Sections (e, f) denote the visual cortex. Note that the hindbrain sections (f-h) comprising the pons, cerebellum and hindbrain are not significantly affected by PEA. The essential findings taken from the 2D images, sections (b-e) are summarized in color-coded 3D reconstructions to the right. A dorsal view of the brain shows the somatosensory, motor and visual cortices as a yellow translucent cover over all of the brain minus the olfactory bulbs and cerebellum/brainstem. When this transverse perspective is viewed from below (ventral) or from the side (sagittal view) the prefrontal cortex (blue) basal ganglia (red) thalamus (light blue) and hippocampus (green can be seen).

Results from graph analysis looking at the number of connections (degrees) between the 173 different brain areas in the 3D MRI rat atlas for each PEA dose and vehicle are summarized by brain regions in Figure 4A. In all cases the 30 mg/kg dose increased brain connectivity. The dose-dependent nature of this effect is shown in the cerebellum where all doses are greater than vehicle and the 30 mg dose is greater than the 3 and 10 mg/kg doses [F_{(1.728, 34.18)} = 25.198, p < 0.0001]. The connectivity of the cerebellum to specific brain areas is shown in Figure 4B. All efferent information leaving the cerebellum essentially goes through the three deep cerebellar nuclei, e.g., dentate, fastigial and interposed highlighted in red. Following vehicle treatment their local connections are limited to surrounding cerebellar and brainstem areas (black circles), and a few extended connections, e.g., superior colliculus, CA3 and dentate gyrus of the hippocampus (green circles). Following the 30 mg/kg dose of PEA there is a significant increase in the number of degrees to local areas within the cerebellum and brainstem and to several more extended connections that include the accumbens, bed nucleus of the stria terminal (BNST), thalamus, hypothalamus, and visual cortex. Areas that were connected with vehicle treatment but not PEA treatment are shown as open white circles. These data are summarized in the 3D reconstructions in Figure 4C.

Figure 5, 6 illustrate changes in the plasma and CNS lipidome after acute PEA administration. Figure 6 is a selected heatmap (41 of 84) of the endogenous lipids analyzed in 6 brain regions and plasma after the 30 mg PEA/kg dose. See Supplementary Figure S5 for full heatmap. Figure 5 provides bar graphs with individual data points of selected data to illustrate the types of data used for generating the heatmap in Figure 6. Figure 5A shows the dose–response values of plasma PEA after 3, 10, and 30 mg/kg injections. These data show that PEA levels in the 30 mg/kg dose are 50-fold higher than those in the vehicle-injected group. This is illustrated in Figure 6 as five arrows denoting the largest fold increase in our scale (see Methods: MS analysis and Figure 3). Levels of PEA in hypothalamus (HYP) and cerebellum (CER) are also significantly increased when compared to vehicle control within brain region. When brain regions are combined, the overall levels of PEA are significantly higher than vehicle though the levels in the PEA-injection group show significantly more variability within group than the vehicle with some individual values that are more than 3-fold higher than vehicle (Figure 5B). Conversely, the levels of the endogenous cannabinoid, Anandamide (N-arachidonoyl ethanolamine; AEA) are significantly decreased in the striatum (STR), thalamus (THAL), and anterior cortex (CTX; Figure 6), which also translated in an overall decrease across the CNS (Figures 5C, 6). Likewise, in addition to individual brain regions (Figure 6), levels of N-oleoyl alanine (Ol-Ala; Figure 5H) were significantly reduced across the brain with some levels being more than 3-fold lower than the vehicle levels. Conversely, levels of 2-AG and N-palmitoyl serine (Pal-Ser; Figure 5G) were moderately but significantly increased across the brain with low variability in the levels in both groups, whereas levels of arachidonic acid (AA) that were significantly reduced in THAL and CTX did not demonstrate and overall reduction across the brain and exhibited low variability (Figure 5F).

Addition endolipids were modulated both in individual brain regions and across the brain (Figure 6; Supplementary material), with the most notable changes being that N-acyl alanine, N-acyl GABA, and N-acyl glycines all had multiple lipid species within those families that had significant decreases in the STR, THAL, and CTX. Supplementary Figure S3 likewise shows that additional lipids in the N-acyl tyrosine and N-acyl valine families were also reduced in STR, THAL, and CTX. For the purposes of discussion and insights into potential mechanisms of action, we will focus on the modulations of the 7 endolipids highlighted here (e.g., PEA, AEA, 2-AG, A-GABA, AA, Ol-Ala, and Pal-Ser; Figures 5, 7).