Multiple regression analysis of a comprehensive transcriptomic data assembly elucidates mechanically- and biochemically-driven responses to focused ultrasound blood-brain barrier disruption

Abstract

Background: Focused ultrasound (FUS) blood brain barrier disruption (BBBD) permits the noninvasive, targeted, and repeatable delivery of drugs to the brain. FUS BBBD also elicits secondary responses capable of augmenting immunotherapies, clearing amyloid-β and hyperphosphorylated tau, and driving neurogenesis. Leveraging these secondary effects will benefit from an understanding of how they correlate to the magnitude of FUS BBBD and are differentially affected by the mechanical and biochemical stimuli imparted during FUS BBBD. Methods: We aggregated 75 murine transcriptomes in a multiple regression framework to identify genes expressed in proportion to biochemical (i.e. contrast MR image enhancement (CE)) or mechanical (i.e. harmonic acoustic emissions from MB-activation (MBA)) stimuli associated with FUS BBBD. Models were constructed to control for potential confounders, such as sex, anesthesia, and sequencing batch. Results: MBA and CE differentially predicted expression of 1,124 genes 6 h or 24 h later. While there existed overlap in the transcripts correlated with MBA vs CE, MBA was principally predictive of expression of genes associated with endothelial reactivity while CE chiefly predicted sterile inflammation gene sets. Over-representation analysis identified transcripts not previously linked to BBBD, including actin filament organization, which is likely important for BBB recovery. Transcripts and pathways associated with neurogenesis, microglial activation, and amyloid-β clearance were significantly correlated to BBBD metrics. Conclusions: The secondary effects of BBBD may have the potential to be tuned by modulating FUS parameters during BBBD, and MBA and CE may serve as independent predictors of transcriptional reactions in the brain.

Trial registration: ClinicalTrials.gov NCT02986932 NCT03739905 NCT03119961 NCT03671889 NCT04118764.

Keywords: blood-brain barrier; drug delivery; focused ultrasound; neurogenesis; transcriptomics.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interest exists.

© The author(s).

Figures

Figure 1

MBA and CE are not collinear. (A) Scatter-plot for all samples with paired CE and MBA data. The solid blue line and text represent linear regression, while the dashed red line represents the MBA mean, simulating a null linear fit. (B) Paired T1-weighted contrast-enhanced 3T MRI images (top) and PCD traces in the Fourier domain around the 2nd harmonic (bottom) for 2 different mice (M11 and M12) during FUS BBBD treatment within a single experiment. Comparison of M11 and M12 illustrates that the relative magnitudes of MBA and CE can vary markedly from treatment to treatment.

Figure 2

Overview of dataset processing and variability. (A) Flow chart describing computational processing pipeline. Untreated and FUS treated samples from multiple studies were pooled and analyzed for contrast enhancement (CE) and microbubble activation (MBA). Linear and exponential models were fit for each prediction metric (CE or MBA) and timepoint (6 h vs 24 h post treatment), followed by bioinformatics analyses. (B) Principle components analysis of RNA-seq transcript counts after variance stabilizing transformation. Each dot represents a single sample, color coded according to the sample characteristics including sex, anesthetic, and harvest timepoint.

Figure 4

CE and MBA predict expression of BBB associated transcripts. (A-D) Heatmaps of significance of correlation (red) or anti-correlation (blue) for selected BBB-associated genes predicted by CE or MBA at 6 or 24 h post-FUS (columns). Selected categories include (A) tight junctions, (B) leukocyte adhesion, (C) transporters, and (D) transcytosis/miscellaneous.

Figure 3

CE and MBA predict significant gene expression 6 h and 24 h after FUS BBBD. (A) Scatter-plots of TPM normalized expression for the top 3 genes predicted by CE or MBA at 6 h or 24 h after treatment. (B) Tile chart representing the top 15 genes predicted in each pool. Note that the absence of a tile for a particular pool-gene combination does not necessarily mean the gene is not significantly correlated, just that it is not in the top 15. (C) Tile chart representing the top 11 anti-correlated genes from each pool, with the same conditions as in B. (D) Upset plot indicating gene identity overlaps of positively correlated genes from each pool. (E) Upset plot indicate gene identity overlaps of anti-correlated genes from each pool.

Figure 5

CE and MBA robustly predict enrichment of genes associated with actin filament organization. Integrated gene concept network for the actin filament organization pathway, which was significantly enriched across all 4 pools. Each dot, representing a contributing gene, is color coded as a pie-chart representing pools in which that gene is significantly correlated.

Figure 6

Over-representation analysis reveals gene-sets that are most strongly associated with MBA and CE. (A-D) Gene concept networks of the top 5 over-represented gene sets expressed 6 h after FUS proportional to CE (A), 6 h after FUS proportional to MBA (B), 24 h after FUS proportional to CE (C), and 24 h after FUS proportional to MBA (D). Redundant pathways were removed by semantic similarity analysis. Supporting genes within each network are colored in proportion to the significance of their correlation with the specified metric at the specific timepoint.

Figure 7

Over-representation analysis of gene-sets associated with neurogenesis and amyloid-beta clearance. Integrated gene concept networks are shown for the extracellular matrix organization, regulation of ERK1 and ERK2 cascade, amyloid-beta clearance, and microglial cell activation pathways. Each dot, representing a contributing gene, is color coded as a pie-chart representing pools in which that gene is significantly correlated.