This study used indoor-housed sexually mature, female rhesus macaques (n=8) maintained on a standard chow diet (TestDiet, St. Louis, Missouri) with an additional cookie with either THC (n=4) or placebo (n=4) daily. Although there are two primary cannabinoids in cannabis, THC and cannabidiol (CBD), we initially focused on the direct effects of THC alone to increase rigor and reproducibility and minimize potential confounders. CBD exhibits low binding affinity for CB1 and CB2 receptors(22), and there is variability in CBD product composition (e.g., can also contain THC, pesticides) and poor labeling accuracy (i.e., half of CBD product labels are inaccurate)(23).

Cookies containing THC (THC edible) were made using research-grade THC obtained directly from the National Institute on Drug Abuse (NIDA) Drug Supply Program (24, 25). Edibles were administered prior to the animal’s daily chow to ensure consumption on an empty stomach and to confirm complete ingestion. Animals were slowly titrated up to 2.5mg/7kg/day of THC, equivalent to a heavy human medical cannabis dose, using published weight-based medical cannabis acclimation recommendations (26) approximately 4 months prior to undergoing time-mated breeding as previously published (24). THC-exposed pregnant animals continued to consume a daily THC edible (2.5mg/7kg/day) throughout pregnancy and postnatally. We previously established that the proposed THC dosing in the rhesus macaque recapitulates the same THC pharmacokinetics reported in humans following a similar edible weight-based THC dose. In addition, in medical cannabis dispensaries in California and Washington states, the median THC dose of edible products is approximately 54mg, thus the dose proposed in this study (25mg for a 70kg adult) reflects contemporary dosing (27, 28). Lung Blood-oxygen level dependent (BOLD)-MRI consisting of T₂* measurements was performed near gestational days 110 and 150. All animals underwent natural delivery at term (~168 days). At 6 months, infants underwent pulmonary function testing prior to necropsy (Figure S1).

All protocols were approved by the Oregon National Primate Research Center (ONPRC) Institutional Animal Care and Use Committee and conformed to all guidelines for humane animal care (IP0001389). Methods are reported in accordance with the ARRIVE guidelines (https://arriveguidelines.org)(29). All methods of euthanasia utilized at the ONPRC are consistent with the American Veterinary Medical Association Guidelines for the Euthanasia of Animals(30). Sodium pentobarbital (Vortech Pharmaceuticals) was delivered intravenously followed by exsanguination under a deep surgical plane of anesthesia.

Lung Blood-oxygen level dependent (BOLD)-MRI consisting of T2* measurements was performed at approximately 110 and 150 days of gestation. Animals were maintained on 1–2% isoflurane and intravenously administered a neuromuscular blocker (cis-atracurium at 0.1mg/kg) prior to fetal magnetic resonance imaging, for temporary arrest of maternal motion to improve image quality. Maternal respirations were controlled by mechanical ventilation and images captured during expiratory breath-holding. MRI scans were conducted using an NHP-dedicated Siemens Tim Trio (3T) MRI system (Erlangen, Germany) using a circularly-polarized transmit, 15-channel receive radiofrequency “extremity” coil (QED, Cleveland, OH). Following localization of the uterus and acquisition of single-shot turbo spin-echo (HASTE) anatomic images in the coronal and axial planes, axial 2D multislice spoiled gradient echo (SPGR) images spanning the entire uterus were acquired at six in-phase echo times with monopolar readout gradients(31). Fetal oxygen perfusion was quantified using BOLD-MRI as previously published(31). Fetal lung volumes were estimated separately for left and right lung by voxel volume weighted numerical integration over lung regions of interest (ROIs) drawn on maps of T2*. Right and left lung volumes were summed for each gestational age measured.

PFTs were performed in intubated, anesthetized infant monkeys as previously published (32, 33). Measurements included total lung capacity (TLC), vital capacity (VC), functional residual capacity (FRC), inspiratory capacity (IC), residual volume (RV), and expiratory reserve volume (ERV). In addition, spirometry measurements included forced vital capacity (FVC), forced expiratory flows at 50% FVC (FEF50), and forced expiratory volume in 200 msec (FEV200). After spirometry, carbon monoxide lung diffusion capacity (DLCO) and alveolar volume (VA) were measured. The lung was inflated with the test gas (0.5% Ne; 0.5% CO; 21% O2; balance nitrogen) using a volume that was two-thirds of vital capacity. Following 4-seconds for neon equilibration and CO uptake, the same volume was withdrawn, and the final concentrations of Ne and CO were measured with a gas chromatograph (3000 Micro GC, INFICON, East Syracuse, NY). Measurements were performed 2–3 times with at least 3 minutes between replicates. VA was calculated using the lung inflation volume of test gas and the concentrations of neon in the test gas and the volume withdrawn following the 4-second equilibration; there were no corrections for dead space volume. DLCO was calculated from VA and the CO concentrations in test gas and the volume withdrawn following 4-seconds of CO uptake. Values within 10% were averaged and expressed under standard temperature and pressure conditions.

At necropsy, lungs were removed en bloc with as much trachea as possible. Using a 60cc syringe with attached sterile tubing, 50mL of PBS was used to inflate the lungs through the trachea, with the tube inserted approximately 2cm above the carina. The PBS was immediately pulled back into the same 60cc syringe, with a return volume typically between 35 and 45mL. Three 1mL aliquots of BAL were aliquoted and saved for additional analysis. BAL sample was centrifuged at 2000rpm for 10 minutes to pellet cells. Supernatant was collected for Luminex analysis and assessed for concentrations of immune mediators using the R&D 36-plex NHP XL Cytokine Premixed Kit (Cat# FCSTM21–36, Bio-Techne, Minneapolis, MN, USA) following manufacturer’s instructions. Samples were acquired using the MAGPIX xMAP instrument (Luminex Corporation, Austin, TX, USA). Data was analyzed using the Luminex Xponent software.

Following BAL, infant lung samples were collected, wrapped in foil, and snap frozen in liquid nitrogen. Frozen lung tissue was later broken into small pieces on a stainless-steel plate on dry ice. For each infant, approximately 30 mg of tissue was transferred to a tube containing 750 μl of RLT buffer with 10% β-Mercaptoethanol and a 5mm stainless steel bead. Samples were sonicated in a Tissue Lyser at 25hz in 1 minute bursts until fully homogenized. The homogenate was centrifuged at max RCF for 30 seconds and supernatant was then extracted using the Qiagen All Prep DNA/RNA/miRNA universal kit and the QIAcube for automated sample processing.

DNA extracted from frozen lung samples was sent to Novogene for DNA fragmentation, methylation library preparation, bisulfite conversion, and sequencing on an Illumina NovaSeq 6000 with 150bp paired-end reads at ≥60X. Raw FASTQ sequence files were processed with our standard in-house WGBS alignment protocol. Briefly, paired readsets were preprocessed with Trimmomatic(34) and aligned to the Mmul_10 reference genome using Bismark(35) to map bisulfite converted reads and determine cytosine methylation status.

Differentially methylated regions between THC and control samples were identified using the DMRichR package in R and are defined as having a 10% difference or greater between groups (36, 37). Default parameters were used with the following exceptions: “Percent of samples per a group for CpG coverage cutoff”, which was set to 0.5 (50%); number of permutations for block and DMR analyses was set to the total number of samples (n=8); and cutoff value for single CpG coefficient (effect size) used to discover testable background regions was set at 0.1 (10%). We used Ensembl transcript annotations instead of the default Bioconductor annotations as recommended for non-model organism genomes. The GOfuncR option was selected for gene ontology analysis, which performs permutation based enrichment testing for the DMRs relative to the background regions and accommodates for gene length bias. By default, DMRs are only mapped to genes if they are between 5 kb upstream and 1 kb downstream.

RNA extracted from frozen lung samples was sent to Novogene for RNA QC, library preparation, and sequencing. Messenger RNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, the first strand cDNA was synthesized using random hexamer primers followed by the second strand cDNA synthesis, A-tailing, adapter ligation, size selection, amplification, and purification. Libraries were quantified with Qubit and real-time PCR and assessed for size distribution with a bioanalyzer. Quantified libraries were pooled and sequenced on an Illumina NovaSeq 6000 with ≥30M (150bp paired-end) reads per sample. FASTQ sequence data was preprocessed with Trimmomatic(34) and aligned to Mmul_10 reference genome using STAR(38) stranded alignment to generate a table of counts. After trimming and alignment the raw RNA-seq count table consisted of 35,432 Ensembl annotated genes. Count data was transformed to counts per million (cpm) and we filtered out genes with low coverage (<=0.1 cpm in >=5 samples, based on the minimum number of samples per group). We next performed principal component analysis (PCA) and sample clustering on the remaining genes (n= 17,464) to check for potential outliers. We estimated fractions of 10 immune cell types within the bulk lung RNAseq using the immunedeconv package and the quanTISeq method (39).

The edgeR package(40) was used for differential expression analysis between THC-exposed and control groups and results visualized using heatmaps with ComplexHeatmap (41), volcano and boxplots with ggplot2(42). We used the STRINGdb, enrichplot, and clusterProfiler packages(43, 44) to visualize interaction networks among differentially expressed genes (DEGs) and to examine enrichment of Gene Ontology (GO) terms, KEGG Pathways, Keywords, PMIDs, Tissues, etc. We also used Weighted Gene Co-expression Network Analysis (WGCNA) (45) to identify correlated modules of expression and to relate expression to phenotype data, visualized using functions from the comethyl package(46).

To examine the correlation between log normalized expression of nominal DEGs with methylation, we first filtered the WGBS Bismark object to only include CpGs with 10X coverage in 50% or more of samples and performed Pearson correlation between expression and individual filtered CpGs within 25kb. We next examined the correlation between DMRs and expression using the smoothed methylation value per DMR per sample and expression of transcripts within 25kb.

The 10x Genomics CytAssist-enabled Visium assay uses probe sets representing 18,085 human genes to spatially profile transcription across tissues. Hybridized probes were captured on a 5,000 spot array over a 6.5×6.5mm area of tissue. Briefly, 5 μm sections of FFPE tissue were mounted on microscope slides, deparaffinized, stained with hematoxylin and eosin and imaged on a Nikon Ni-E microscope with motorized stage as per kit guidelines (10x Genomics, Demonstrated Protocol CG000520). Decrosslinked tissue was hybridized with the Visium Human Transcriptome Probe Kit v2 and ligated probes captured on Visium slides using the CytAssist instrument. Libraries were generated from captured probes as per manufacturer’s instructions (10x Genomics, User Guide CG000495). Paired end sequencing was performed by Novogene at a depth of >25,000 read pairs per spot. Space Ranger (v2.0.1) was used on the University of Kentucky HPC to align and count reads. Data was assessed for quality based on web summary files from Space Ranger and based on distribution of features.

Output files from Space Ranger were loaded into Seurat (v4.4.0) for data pre-processing and analysis (47) in R (v4.2.2). We first filtered each dataset to remove spots with less than 500 features detected. We then normalized count data using the SCTransform function (vst.flavor=v2) (48) prior to integration after selecting a list of informative features (49). We next performed joint dimension reduction on the integrated Seurat object (RunPCA, RunUMAP, FindNeighbors, and FindClusters) to identify clusters of spots with co-expression that are representative of distinct structures and cell mixtures, with approximately 1–10 cells per 55-micron spot. Next, we overlaid cluster annotation on spatial images, extracted average expression (i.e. per cluster, gene, and treatment), and performed differential expression analysis between treatments using the non-parametric Wilcoxon rank sum test for each cluster. Additionally, we used a human reference (LungMAP_HumanLung_CellRef_Seed.v1.1.2), composed of 48 normal lung cell types with cell identities derived from the LungMAP CellCards(50) to identify enriched cell populations within clusters. We used these estimated cell proportions and cluster gene markers to manually annotate clusters along with information regarding underlying structural features.

To confirm the results from differential expression analysis between treatment groups using bulk RNAseq, we performed a pseudobulk analysis of differential expression using the spatial RNAseq dataset. To do so, we aggregate the counts per gene per sample across all spatial clusters and then perform filtering and differential expression analysis between groups using edgeR and the same process used for bulk RNAseq. We directly compared the log counts per million between bulk and pseudobulk datasets from the same animals using scatterplots and correlation analysis. We also compared the logFC between the two platforms to validate the direction and magnitude of change in the top DEGs and performed pathway enrichment analysis on all candidate DEGs.

Statistical comparison between treatment groups for individual outcome measurements (i.e. measurements at MRI, PFT, and necropsy) was performed as follows: means and standard errors for each measurement were calculated using the emmeans package. P-values are from the ANOVA between treatment groups using unadjusted models and the car package, and models adjusted for crown-rump length at PFT when applicable. Additional integration of datasets was performed using functions from comethyl and WGCNA to reduce ‘omic data into co-regulated modules, which are then correlated with sample traits and outcome measurements.

As described in the methods, pregnant dams with chronic THC consumption (THC, n=4) or placebo (CON, n=4) were allowed to deliver naturally at term. There was an equal ratio of male to female offspring per group. Offspring were kept with their mothers for normal rearing after delivery. There were no significant differences in baseline characteristics between THC- and CON-exposed dams, or offspring characteristics at birth or PFT as shown in Table 1.

Blood-oxygen level dependent (BOLD)-MRI consisting of T₂* measurements was performed near gestational days 110 (G110) and 150 (G150). In THC exposed fetuses there were non-significant decreases in lung volume (p=0.108; p=0.268), lung perfusion, and oxygen availability (p=0.089; p=0.328) at ~G110 and G150, respectively (Figure 1). All animals underwent natural delivery at term and at 6 months of age, offspring were sedated, intubated, and PFTs were performed. The lung measures of functional residual capacity (FRC) and total lung volume (TLC) were significantly reduced in the THC exposed animals (p = 0.015 and 0.047, respectively; Figure 2A). After adjusting for crown-rump length at PFT, the difference in FRC remained significant (p=0.029) and TLC association was slightly attenuated (p=0.076). Static lung compliance and gas exchange, as measured by diffusing capacity of the lungs for carbon monoxide (DLCO), also tended to decrease (p = 0.11 and 0.13 respectively). Forced expiratory flows (FEFs) and volumes (FEVs) were not significantly different between groups (Table S1).

Quantification of cytokines in the bronchoalveolar lavage (BAL) collected at necropsy was performed to explore potential differences between THC versus control animals and correlation with respiratory outcomes. We observed significant decreases in the lung growth factors brain-derived neurotrophic factor (BDNF), fibroblast growth factor-basic (FGFb), and vascular endothelial growth factor (VEGF) and in the pleotropic cytokine interleukin-15 (IL-15) in THC animals (Figure 2B). The full list of cytokines quantified in BAL is provided in Table S2. Additionally, lung volume measured at different stages of development and by different methods were significantly correlated with each other and with BAL cytokines (Figure 2C).

We next performed differential expression analysis between THC-exposed and control lungs using bulk-RNAseq. There were 730 genes nominally different between treatment groups (0 FDR significant; Figure 3A and Table S3) and pathway analysis revealed enrichment of gene ontology (GO) terms related to immune signaling (Figure 3A and Table S4) and activation of KEGG pathways involved in cytokine-cytokine receptor interaction (NES= 1.853; p=0.0017). Given this finding, we estimated fractions of 10 immune cell types within the bulk lung RNAseq and found significantly decreased proportion of natural killer (NK) cells in the THC exposed animals (Figure S2). Adjusting for cell fractions in differential expression analysis resulted in 1,453 nominally significant DEGs (Figure 3B and Table S5). Cell fraction adjusted DEGs were enriched for GO terms related to extracellular matrix organization and development including “lung development”, “respiratory tube development”, and “kidney development” (Figure 3B and Table S4) and gene set enrichment analysis revealed suppression of the “retrograde endocannabinoid signaling” KEGG pathway (NES= −2.015; p=0.0007).

We used WGCNA to cluster all genes into correlated modules and identify gene networks associated with treatment and biological measurements (e.g., FRC, TLC, etc.). Using this approach, we identified a total of 74 modules, two of which were significantly correlated with THC treatment (MEdarkorange and MEorangered3; (Figure 3C). The MEdarkorange module eigenvalues were also correlated with BDNF and FGFbasic protein concentration in BAL, DLCO, and FRC (Figure 3C). Pathway analysis of the MEdarkorange genes (n=124; Table S3) identified enrichment for immune tissues and cell types, which is similar to results of pathway and KEGG enrichment of all DEGs without cell type adjustment. The MEorangered3 module eigenvalues were also correlated with BDNF and (BOLD)-MRI T₂* measures at ~G110, however this module only 19 contained genes and therefore we did not perform pathway analysis.

DNA was isolated simultaneously from the same samples used for RNAseq. We identified 3,558 nominally significant differentially methylated regions (DMRs) with 10% or greater difference in methylation between THC-exposed and control lungs (70% hypermethylated, 30% hypomethylated) (Figure 4; Table S5). The top 3 DMRs were annotated to phosphatidylinositol glycan anchor biosynthesis class Z (PIGZ; difference= −54%), transient receptor potential cation channel subfamily M member 8 (TRPM8; difference= −28%), and PX domain containing serine/threonine kinase like (PXK; difference= −34%). GO enrichment testing suggests that DMRs are enriched for genes located in cellular junction and involved in cell junction organization (Figure 4C). Enrichment testing for gene region (genic) annotations identified positive enrichment of DMRs in promoter, exon, intron, and 3’UTR genomic features and depletion in intergenic regions. DMR annotations were also enriched for CpG Islands, Shores, and Shelves, and depleted in Open Sea regions (Figure S3).

For each candidate differentially expressed gene we performed correlation with DNA methylation (DNAm) using CpGs within 25kb to examine epigenetic dysregulation of gene expression following maternal THC use during pregnancy. Out of 730 DEGs in our unadjusted model, 673 (over 92%; Table S3) were correlated with DNAm at one or more CpG located within 25kb of one or more measurable transcripts, and 97 unique DMRs were correlated with gene expression (included in Table S6). Similarly, over 95% of DEGs following adjustment for cell types were correlated with DNA methylation at one or more CpG (Table S5).

Bulk-RNAseq represents the average of gene expression across all cell types within a sample and can mask differences between and within cell populations. Therefore, we next determined spatial niches of differential expression and underlying lung structures using the 10x Genomics Visium Spatial Gene Expression platform which provides an average resolution of 1 to 10 cells per 55-micron spot. We measured expression of 17,318 genes (>95% of human capture probes), demonstrating that the Visium platform is highly compatible with rhesus macaque DNA. Data integration and dimension reduction revealed 13 distinct clusters of spots (Figure 5A) and the top marker genes for each cluster are visualized in Figure S4. Clusters were manually annotated based on 1) cell type mapping per spot to the Human LungMap reference (Figure 5B), 2) underlying tissue morphology (Figure 5C), and 3) functional enrichment analysis of marker genes for each cluster (Figure 5D) (51). The only significant difference in the percentage of spots between treatments was in the “0 cluster” (p=0.008), although there were trends for decreased percentage of cluster 4, and increased percentage of spots in clusters 3, 5, and 11, in the THC-exposed lungs (Figure S5).

Differential expression analysis within each cluster identified 186 unique spatial cluster DEGs (sDEGs) with Bonferroni p<0.05 between THC and control lungs, although several sDEGs were identified in more than one cluster resulting in 484 total instances of differential expression (Table S7). The majority of sDEGs were from the clusters enriched for vascular smooth muscle cells (cluster 10), ciliated cells lining airways (cluster 6), and immune cells (cluster 11) (Figure 6A). Functional analysis of all sDEGs identified enrichment of genes involved in extracellular matrix organization, regulation of peptidase activity, muscle contraction, and wound healing (Table S8). Results from enrichment analysis of sDEGs in each cluster separately were similar to results for all sDEGs, with response to transforming growth factor beta, and response to hyperoxia enriched in clusters 6 and 11, and sDEGs in cluster 11 uniquely enriched for terms related to immune system processes and regulation of tube size (Figure 6B; Figure S6). Pseudobulk analysis of spatial transcriptomics, which aggregates expression per gene across spatial clusters, support results from bulk-RNAseq, as shown by significant positive correlation of effect sizes (logFC) with bulk DEG candidates (r=0.68; p<2.2e-16).