Bovine milk sEVs (BEVs) were isolated from skim milk from a local grocery store by using sequential ultracentrifugation and authenticated by using Nanosight NS300 nanoparticle size analysis, scanning electron microscopy and transmission electron microscopy as previously described (Supplementary Figure 1) (32). The antibodies and their dilutions used in immunoblot analysis were the same as previously described (32). Protocol details were deposited in the EV-Track database (ID EV210338). BEVs were suspended in sterile phosphate-buffered saline (PBS) and kept at −80°C until use. For transport studies in cell monolayers, BEVs were labeled with FM 4-64 (Molecular Probes, Inc., Eugene, OR, United States) or by labeling RNA cargos by using the ExoGlow-RNA™ EV Labeling Kit (System Biosciences, Inc., Palo Alto, CA, United States) following the manufacturers’ recommendations. For transport studies in dual-chambers, BEVs were loaded with synthetic IRDye-labeled miR-34a as previously described (14).

Murine brain endothelial bEnd.3 cells [American Type Culture Collection (ATCC) CRL-2299, passages 21 – 30)] and C8-D1A astrocytes (ATCC CRL-2541, passage unknown to ATCC) were purchased form ATCC. BV2 microglia (passage 15 – 25) were a gift from Dr. Sanjay Maggirwar (University of Rochester Medical Center, Rochester, NY, United States). Cells were cultured following ATCC recommendations. In monolayer studies, uptake of BEVs by bEnd.3 cells and BV2 microglia was assessed as previously described using times, concentrations and competitors shown in Results (13). Transport kinetics was modeled using the Michaelis–Menten equation and non-linear regression; modeling was conducted using GraphPad Prism 6.0 (GraphPad Software). Confocal Z-stacks were collected at 60-fold magnification using 300 nm z-spacing on an A1R-Ti2 confocal system (Nikon) and used to determine whether bEnd.3 cells internalized BEVs or whether BEVs adsorbed to the cell surface. Dual chamber assays as a model of transport across the blood-brain barrier (BBB) were conducted as previously described with the following modifications (13). bEnd.3 cells were seeded on the semiporous membrane in co-culture with astrocytes in the bottom chamber; the integrity of the bEnd.3 cell monolayer was assessed by using trans endothelial electrical resistance in an Epithelial Volt/Ohm meter equipped with STX2 electrodes (EMD Millipore Corporation). Uptake of BEVs labeled with FM 4-64 and RNA cargos labeled with ExoGlow-RNA™ was quantified by using a microplate fluorescence reader (BioTek Instruments, Inc., Winooski, VT, United States) and confocal microscopy imaging, respectively. The transfer of IRDye-labeled miR-34a across a bEnd.3 cell monolayer was measured in dual chamber assays by using an Odyssey^® imaging system (LI-COR, Inc., Lincoln, NE, United States).

We developed an sEV and cargo tracking (ECT) mouse on the C57BL/6J genetic background that enables studies of exosome and cargo trafficking among tissues, as well as studies of the transfer of sEVs from dam to pup (14). Briefly, ECT mice express an open reading frame (ORF) coding for the exosome marker, CD63 (33) fused with enhanced green fluorescent protein (eGFP) flanked by loxP sites. In the presence of cre recombinase, the CD63-eGFP ORF is removed, and mice express an open reading frame coding for a fusion protein of CD63, near-infrared protein (iRFP), transmembrane domain and a second iRFP. The second iRFP localizes to the outer exosome surface and can be used to collect exosomes for cargo analysis.

Wild-type (WT) newborn C57BL/6J pups were fostered to ECT dams or WT dams from synchronized pregnancies and nursed for 17 days. Pups were euthanized and brains were fixed via transcardial perfusion of 4% paraformaldehyde, stored in phosphate-buffered saline and shipped to the Whole Brain Microscopy Facility at the University of Texas Southwestern Medical Center for analysis by serial two-photon tomography (STPT) (34–37), and immunostaining of brain slices with anti-GFP antibodies and confocal imaging. Eleven whole brains from WT pups fostered to ECT dams across three separate litters and two brains from WT pups fostered to WT dams were used for STPT and anti-GFP immunostaining of isolated coronal brain sections. STPT is a high-resolution, high-throughput volumetric imaging strategy for assessing the regional distribution of native fluorescent labels throughout entire uncleared mouse brains via serial vibratome sectioning and mosaic two-photon imaging (34–37). Vibratome sections in the coronal plane at 75 μm thickness were generated via STPT. A subset of the sections were selected for immunoamplification of the native GFP signals via immunostaining with anti-GFP antibodies (Invitrogen # A11122; 1:500 dilution) and confocal imaging. Two sections (one at the level of the dorsal hippocampus and one at the level of the cerebellum) were selected from each brain for immunostaining and confocal imaging. Two 20X fields of view, at the cortex and hippocampal dentate gyrus, were acquired from the rostral section and one 20X field of view including cerebellar molecular, Purkinje and granule cell layers was acquired from the caudal section. Immunostained coronal brain sections were imaged using a Zeiss LSM 780 confocal microscope (Live Cell Imaging Facility, UTSW) at 20X and with the same acquisition parameters across samples. All animal studies in this paper were approved by the Institutional Animal Care Program at the University of Nebraska-Lincoln (protocols 1229 and 1713).

We assessed BEV-dependent gene networks in the left hippocampus in male and female C57BL/6J mice (Jackson Labs., stock 000664). Briefly, C57BL/6 mice were fed ERD or ERS diets starting at age 3 weeks for 7 weeks when mice were mated (15). Pups born to these breeders were continued on parental diets until age 7 weeks. Pups were euthanized via transcardial perfusion with phosphate-buffered saline and brains were excised. The left hippocampus was dissected and flash frozen in liquid nitrogen for storage at −80°C. Total RNA was extracted by using miRNeasy Kit (Qiagen, Inc., Hilden, Germany) according to the manufacturer’s instructions, and the concentration, quality and integrity of RNA was analyzed as previously described (38, 39). cDNA libraries were prepared by using a proprietary kit and sequenced by using a paired-end 150 base-pair protocol and the NovaSeq platform (Illumina, Inc., San Diego, CA, United States) in the Beijing Genomic Institute. RNA-seq data were analyzed as previously described (16). RNA-seq data were validated for seven differentially expressed genes by using quantitative real-time PCR (qRT-PCR) with primers shown in Supplementary Table 1 (40–47). Briefly, total RNA was reverse transcribed using the QuantiTec Reverse Transcription kit (Qiagen, Inc., Hilden, Germany) and qPCR was performed with SsoAdvanced Universal SYBR Green Supermix in a 10-μl reaction volume by following the manufacturer’s instructions (Bio-Rad Laboratories, Inc., Hercules, CA, United States) using the following conditions: 94°C for 3 min, followed by 40 cycles at 94°C for 30 s, 60°C for 30 s, and 72°C for 30 s. The relative gene expression were calculated using 2^–ΔΔCt method with GAPDH as internal reference gene. Three biological repeats were analyzed and each biological repeat was analyzed three times.

Neuronal dendrites are highly branched, tree-like structures, the morphological complexity of which is linked to signal integration and firing pattern of individual neurons and the functionality of neural circuitry (48). Mice were fed ERD and ERS diets as described above. Hippocampal neurons from the left hemispheres of mice were stained using the Golgi-Cox method as previously described (49, 50). Brightfield images of dentate granule cells in the suprapyramidal blade were collected in the Z plane at 20× magnification in 0.3-μm steps with an Olympus IX-81 inverted spinning disk confocal microscope using MetaMorph Advanced software version 7.1 (Molecular Devices). Three-dimensional dendritic structures of 3 – 5 granule cells were manually traced and reconstructed in each hippocampus using NeuroLucida version 2019.1.2 (MBF Bioscience). Dendritic complexity and other morphological features were quantified via Sholl analysis through NeuroLucida Explorer version 2019.1.2 (MBF Bioscience). Research staff was blinded regarding the treatment of mice.

Phenotyping studies focused on the assessment of spatial learning and memory (SLM), kainic acid-induced seizures, acoustic startle response and prepulse inhibition. These endpoints were chosen because the hippocampus has been implicated in SLM, kainic acid-induced seizures and acoustic startle response in mice and rats (51–53). The choice of phenotypes is also consistent with our observations that milk sEVs accumulated in the hippocampus (in addition to other regions) and altered the expression of genes implicated in axon guidance and calcium signaling in murine hippocampi in mice; dietary depletion of sEVs led to a decrease in neuronal branching in murine dentate granule cells (see the section “Results”).

Spatial learning and memory was assessed at two ages by using the Barnes maze (54). The Barnes maze measures the ability of a mouse to learn and remember the location of an escape hole on a circular surface with the help of a visual cue; low values represent strong test performance (54). Mice were trained on six consecutive days prior to measuring the time needed to locate the escape hole on day 7. Training and measurement were stopped after 5 min if a mouse failed to enter the escape hole. Measurements were take by using the ANY-maze video tracking system (Stoelting Co., Wood Dale, IL, United States). SLM experiments were conducted using the same mice that were subsequently used in RNA-sequencing analysis, except that additional mice were included in tests of SLM. SLM was also assessed in adult mice ages 12 – 15 weeks fed ERD and ERS diets starting at age 3 weeks. Mice were randomly assigned to diet groups and both sexes were studied.

In tests of seizure severity, C57BL/6J mice ages 3 weeks were fed ERD or ERS diets for 18 weeks when seizures were triggered by subcutaneous administration of kainic acid (25 mg/kg body weight) (55). Kainic acid is a non-hydrolyzable glutamate analog that binds to five glutamate receptors in the brain (56, 57), thereby causing neuronal excitotoxicity and seizures (55). The severity of seizures was scored using a modified Racine scale (58, 59). In the modified Racine scale, seizures are scored at timed intervals for 2 h and the highest score in each 5-min block are reported: 0, no seizure; 1, immobility; 2, forelimb and/or tail extension; 3, automatisms; 4, forelimb clonus, rearing, and/or falling; 5, repetition of stage 4; 6, tonic–clonic seizures; and 7, death.

Acoustic startle response and pre-pulse inhibition were assessed as previously described using the mice from the Barnes maze experiments 1 day after conducting the studies in the maze (60, 61). Pre-pulse inhibition of the acoustic startle response was measured using prepulses of 68, 74, and 80 dB for 10 ms with 65-dB background white noise, followed by a pulse of 105 dB for 20 ms; intervals between stimuli were random 10 – 30 ms. The percent prepulse inhibition of acoustic startle response was calculated as [(1 – (startle response at 105 dB with prepulse stimuli)/startle response for startle response at 105 dB without prepulse)] × 100. The acoustic startle response was scored ten times per mouse for each of the following conditions: 105-dB startle stimulus without prepulse, 105-dB startle stimulus with prepulses of 68, 74, 80 dB, and no startle stimulus with pulses of 80 dB. Test were performed by using an SR-LAB Startle Response System (San Diego Instruments, San Diego, CA, United States).

The F-test was used to assess the homogeneity of variances (62). Some variances were heterogeneous, e.g., Racine scale scores. Log transformation of these data resulted in homogenous data variation. Data from time-dependent transwell studies were analyzed by using repeated measures one-way ANOVA followed by Dunnett’s multiple comparisons test. The Kruskal-Wallis test was used for the analysis of data from acoustic startle response experiments. Data from both seizure studies and the Sholl analysis were analyzed by using repeated measures ANOVA and mixed procedure. The distance to soma was used as the within-subjects repeated measure. qRT-PCR data were analyzed by t-test. The model includes treatment as the fixed effect and mouse nested in treatment as the random term. Data analysis was conducted by using SPSS 27, SAS 9.4 and GraphPad Prism 9.0. P < 0.05 was considered statistically significant. Data are reported as mean ± SEM.

bEnd.3 cells internalized BEVs by using a saturable process and secreted microRNA cargos across the basal membrane. The uptake of FM4-64-labeled BEVs was modeled using the Michaelis-Menten equation (Figure 1A): transporter capacity (maximal velocity, V_max) = 0.77 ± 0.18 × 10¹¹ BEVs/(10,000 cells × 45 min) and affinity (Michaelis–Menten constant, K_m) = 1.8 ± 2.0 × 10¹¹ BEVs/mL). All subsequent studies in bEnd.3 cells were conducted under conditions when BEV concentrations (6 × 10¹¹ BEVs/mL) and incubation times (45 min) do not limit BEV uptake (Supplementary Figure 2). Z-stack confocal imaging confirmed that bEnd.3 cells internalized ExoGlow-RNA™ labeled BEVs, as opposed to BEVs adsorbing to the cell surface (Supplementary Figure 3). Upon internalization, the ExoGlow-RNA™ labeled BEVs localized to the cell cytoplasm (Supplementary Figure 4).

In a dual chamber model of transfer across the BBB, bEnd.3 cells secreted IRDye-labeled miR-34a, loaded into BEVs, across the basal membrane into the bottom chamber in co-cultures with astrocytes (Figure 1B). The integrity of the bEnd.3 cell monolayer on the semiporous membrane was assessed by measuring the trans endothelial electrical resistance and reached a plateau approximately 4 days after seeding the cells (Supplementary Figure 5); absence of astrocytes in the bottom chamber caused a loss of monolayer integrity. BEVs that crossed the BBB, were internalized by brain cells, using brain macrophages, BV2 microglia as model. BEVs uptake by BV2 microglia followed saturation kinetics (Figure 1C): V_max = 0.66 ± 0.14 × 10¹¹ BEVs/(10,000 cells × 45 min) and K_m = 1.9 ± 1.9 × 10¹¹ BEVs/mL.

eGFP-positive sEVs accumulated primarily in the brain but also in the liver and small intestinal mucosa in WT pups fostered to ECT dams (Figure 2). No eGFP fluorescence was detected in the brain and liver in WT pups fosterd to WT dams, and the signal in the small intestine represents autofluorescence (control). Brains from WT pups fostered to either ECT or WT dams were subjected to serial two-photon tomography and anti-GFP immunostaining in isolated coronal brain sections. Accumulation of GFP labeled sEVs was evaluated in cortex, hippocampus and cerebellum using confocal imaging. Out of 11 brains from pups fostered to ECT dams across three litters, we observed positive anti-GFP staining in at least one of these three regions in seven of the brains. In total, brains from seven male pups and 4 female pups fostered to ECT dams were imaged. 2/7 male and 2/4 female brains did not show positive GFP signal. GFP signal was not observed in two brains from WT pups fostered to WT dams (one male and one female). Representative 2D coronal section images at the level of the dorsal hippocampus from the brain of a WT pup fostered to an ECT dam (Figure 3A) and that of a WT pup fostered to a WT dam (Figure 3B) shows robust accumulation of GFP labeled sEVs throughout the section from the brain of the pup which was fostered to the ECT dam (Figure 3A), but no GFP signal was observed in the pup fostered to the WT dam (Figure 3B). 3D renderings of the hippocampus (Figure 3C) and the entire brain (Figure 3D) from the pup fostered to the ECT dam demonstrate robust accumulation of GFP labeled exosomes throughout the hippocampus and many other brain regions. Figures 3E–G show confocal imaging of anti-GFP immunostaining in the brain of a different WT pup fostered to an ECT dam in the hippocampus, cerebellum and cortex. Positive GFP labeling was observed in each of these three brain regions. GFP labeled sEVs were detected in the hippocampal granule cell layer in the dentate gyrus (Figure 3E). In contrast, no GFP labeling was seen in brains of pups fostered to WT dams as shown in Figure 3H.

Dietary intake of BEVs altered gene expression in the brain. Two hundred ninety-five genes were differentially expressed in the left hippocampus in male mice fed ERD or ERS diets for 7 weeks (Figure 4A). Three genes (Dcn, decorin; Nos1, nitric oxide synthase 1; Ndn, necdin) were differentially expressed in females. Raw sequencing data can be accessed in the NCBI-BioProject database through accession number PRJNA783128. KEGG pathway analysis revealed 45 BEV-dependent biological pathways in males, each with at least three BEV-dependent genes (Figure 4B). Pathways of calcium signaling and axon guidance were among the three top-ranked pathways. No KEGG pathways emerged in the analysis of mRNA expression data in females.

qRT-PCR analysis of mRNA expression in hippocampi confirmed the patterns obtained by RNA-sequencing analysis (Supplementary Figure 6). For example, the expression of BTAF1 and Dscaml1 was significantly higher and lower, respectively, in males fed the ERS diet compared to males fed the ERD diet in both qRT-PCR analysis and RNA-sequencing analysis. Expression of Ndn, Nos2 and Dcn was lower in females fed the ERS diet compared to females fed the ERD diet in both qRT-PCR analysis and RNA-sequencing analysis. Expression of Cplx3 and Gpr27 was significantly higher and lower, respectively, in males fed the ERS diet compared to males fed the ERD diet in both qRT-PCR analysis and RNA-sequencing analysis but diet effects were not statistically significant in qRT-PCR analysis.

Bovine milk small extracellular vesicle-dependent changes in gene expression and neuronal growth were associated with impaired SLM and increased the severity of kainic acid-induced seizures in mice fed the ERD diet compared to mice fed the ERS diet. For example, when female pups were nursed by dams fed the ERS diet for 3 weeks and continued on the maternal diet for 1 week, the mice performed nine times better in the Barnes maze test of SLM compared to mice nursed by ERD dams and fed the ERD diet for 1 week (Figure 5A). Diet effects on SLM were more modest in males than in females and in mice older than 4 weeks (Supplementary Figure 7); the only exception are females ages 7 weeks in which diet effects were similar to females ages 4 weeks. Path length patterns matched the time it took mice to locate and enter the escape hole (Supplementary Figure 7). Beneficial effects of BEVs on brain function were not limited to SLM but extended to seizure activity which was five times higher in male mice fed the ERD diet 30 min after administration of kainic acid compared to ERS controls (Figure 5B and Supplementary Table 2). Note that the effect of diets on kainic acid-induced seizure activity were also detectable in females but effects were modest when compared to males and not statistically significant (Supplementary Table 2). Dietary effects on pre-pulse inhibition of acoustic startle response were modest and not statistically significant in most age groups and both sexes (data not shown).

Given the preliminary evidence of neuronal accumulation of BEV-derived microRNAs, as well as alterations in hippocampal pathways implicated in neuronal development and cognitive deficits in SLM after dietary BEV depletion, we evaluated its impact on dendritic complexity of dentate granule cells in the hippocampus. Representative samples of traced three-dimensional dendritic architecture of dentate granule cells were shown in Figures 6A,B. Sholl analyses evaluated dendritic complexity by quantifying the number of interactions between dendrites and soma-oriented concentric spheres of increasing diameters (Figure 6C). There is a significant main effect of diet on dendritic complexity. This is evidenced by the smaller number of dendritic intersections in DG neurons from mice fed the ERD diet compared to mice fed the ERS diet (P < 0.05), suggesting that milk sEV deficiency resulted in underdevelopment of neuron dendritic architecture in the developing brain. This difference was primarily due to a greater number of branch nodes of granule cells in ERS mice compared to ERD mice (Table 1). In addition, there was a trend toward a greater number of dendritic tips in ERS mice compared to ERD mice (P = 0.08). The number of primary dendrites and total dendritic length of dentate granule cells were not significantly different between diet groups (Table 1).