A single-cell map of antisense oligonucleotide activity in the brain

Abstract

Antisense oligonucleotides (ASOs) dosed into cerebrospinal fluid (CSF) distribute broadly throughout the central nervous system (CNS). By modulating RNA, they hold the promise of targeting root molecular causes of disease and hold potential to treat myriad CNS disorders. Realization of this potential requires that ASOs must be active in the disease-relevant cells, and ideally, that monitorable biomarkers also reflect ASO activity in these cells. The biodistribution and activity of such centrally delivered ASOs have been deeply characterized in rodent and non-human primate (NHP) models, but usually only in bulk tissue, limiting our understanding of the distribution of ASO activity across individual cells and across diverse CNS cell types. Moreover, in human clinical trials, target engagement is usually monitorable only in a single compartment, CSF. We sought a deeper understanding of how individual cells and cell types contribute to bulk tissue signal in the CNS, and how these are linked to CSF biomarker outcomes. We employed single nucleus transcriptomics on tissue from mice treated with RNase H1 ASOs against Prnp and Malat1 and NHPs treated with an ASO against PRNP. Pharmacologic activity was observed in every cell type, though sometimes with substantial differences in magnitude. Single cell RNA count distributions implied target RNA suppression in every single sequenced cell, rather than intense knockdown in only some cells. Duration of action up to 12 weeks post-dose differed across cell types, being shorter in microglia than in neurons. Suppression in neurons was generally similar to, or more robust than, the bulk tissue. In macaques, PrP in CSF was lowered 40% in conjunction with PRNP knockdown across all cell types including neurons, arguing that a CSF biomarker readout is likely to reflect ASO pharmacodynamic effect in disease-relevant cells in a neuronal disorder. Our results provide a reference dataset for ASO activity distribution in the CNS and establish single nucleus sequencing as a method for evaluating cell type specificity of oligonucleotide therapeutics and other modalities.

Graphical Abstract

Figure 1.

Clustering and assignment of brain cell types. Single-cell gene expression profiles projected into two dimensions using Uniform Manifold Approximation and Projection (42) (UMAP) and characterized using dot plots. In dot plots, gray to blue color gradient represents higher expression while small to large dot size gradient represents broader expression. Thus, a large blue dot indicates a marker widely and highly expressed by cells within the indicated cluster; a small gray dot indicates little to no expression by those cells. A small blue dot can indicate a marker highly expressed by only a subset of cells within the cluster, while a large gray dot can indicate a marker broadly but lowly expressed. MLI = molecular layer interneuron, UBC = unipolar brush cell, PLI = Purkinje layer interneuron, OPC = oligodendrocyte progenitor cell. For breakdown by weeks post-dose and active/inactive treatment group see Supplementary Figure S1.

Figure 2.

Single-cell distribution of ASO activity in mouse cerebellum at 12 weeks post-dose. Mice received 50 μg Malat1 ASO (N = 4), 500 μg ASO 6 (N = 4), or PBS (N = 4) ICV and cerebella were harvested 12 weeks later. (A) Malat1 knockdown after Malat1 ASO treatment, assessed by bulk qPCR (x axis) versus aggregation of single cell sequencing regardless of cell type (y axis), for individual animals (points) and groups (crosshairs indicate means and 95% confidence intervals). (B) Histogram of the number of Malat1 UMIs per single cell across all samples. Arrow indicates a second peak observed only for treated animals. (C) Breakdown across 14 cell types. Top panels are histograms of single cells as in (B), but broken down by cell type. Bottom panels are scatterplots showing total UMIs per single cell versus Malat1 UMIs per cell and best fits by linear regression (Supplementary Table S12). Percentages indicate residual Malat1. Note the ‘key’ panels at the far right of the middle row. (D) Negative binomial (NB) modeling of single cell data as point estimates of knockdown for each animal and each cell type (points) and means and 95% confidence intervals (bars) for each treatment group and cell type. (E) Correlation across cell types of Prnp knockdown by ASO 6 and Malat1 knockdown by Malat1 ASO. Each point is a cell type, colors are from Figure 1A, and point sizes are logarithmically scaled with number of cells sequenced. (F) Histogram of Prnp UMIs per single cell in astrocytes for PBS (gray) and ASO 6-treated animals (cyan). Shaded gray bars indicate the distribution predicted by a NB model fit to the PBS data. Shaded red bars indicate the distribution if the observed 56% residual Prnp RNA in astrocytes corresponded to 56% of cells following the original NB distribution and 44% being set to zero. Shaded cyan bars indicate the distribution if the observed 56% residual corresponded to the NB parameter mu being reduced by 44%.

Figure 3.

Atlas of ASO activity across cell types in mouse brain at 3 weeks post-dose. Mice received a single 500 μg dose of ASO 6 or PBS and were harvested 3 weeks later. N = 4 ASO 6 versus N = 7 PBS for cortex, N = 3 ASO 6 versus N = 4 PBS for thalamus, N = 4 ASO 6 versus N = 4 for cerebellum. (A–C) Concordance of bulk qPCR (x axis) and single cell (y axis) measurement of total Prnp knockdown in cortex (A), thalamus (B) and cerebellum (C). Points are individual animals, crosshairs are 95% confidence intervals on both dimensions. (D) Residual Prnp expression across cell types in three brain regions. Error bars are 95% confidence intervals of the mean. (E) Cortical excitatory neurons from panel A were reclustered into 15 clusters ranked by expression of four excitatory layer markers. Residual Prnp (top) is shown as individual animals (points), means (lines), and 95% confidence intervals of the mean (shaded areas). Each marker's expression (bottom) is normalized to the cluster with the highest expression; points are normalized values and curves are loess fits. (F) Cortical inhibitory neurons from panel A, reclustered and plotted as in panel B. (G–I) Scatterplots of residual Prnp expression (y axis) versus candidate covariates (x axes). Each point represents a region and cell type combination from panel D (cortical fibroblasts, the only cell type with nominally >100% residual expression, are not visible). Colors correspond to cell type colors in Figure 1, and dot sizes scale logarithmically with the number of cells sequenced. Candidate covariates examined are Rnaseh1 expression in UMIs per million (UPM; G), total UMIs per cell (H) and basal Prnp expression (UPM; I).

Figure 4.

Duration of activity across different ASO chemistries in mouse cortex from 2 to 12 weeks post-dose. Mice received PBS (N = 4 per timepoint), 500 μg ASO 1 (N = 4 per timepoint), or 500 μg ASO 6 (N = 4 per timepoint) and cortex was harvested after 2 or 12 weeks. (A) Overall residual Prnp assessed by bulk qPCR (x axis) versus aggregate single cell data without regard to cell type (y axis). Points represent individual animals and crosshairs represent means and 95% confidence intervals on both axes. 2 week data (lower left) and 12 week data (upper right) are connected by arrows indicating washout for both ASO 1 (dark blue) and ASO 6 (cyan). (B) Washout from 2 weeks (left) to 12 weeks (right) for each cell type in mouse cortex for ASO 1 and ASO 6. Lines represent means and shaded areas represent 95% confidence intervals. (C–E) Scatterplots of candidate variables (x axis) basal Prnp expression (UMIs per million, UPM; C), Rnaseh1 expression (UPM; D), and 2 week residual Prnp (E) versus percentage points of recovery (washout at 12 versus 2 weeks, y axis) for each cell type (cell types are points, sized logarithmically by number of cells) for both ASO 1 (dark blue outlines) and ASO 6 (cyan outlines), with symbol fill color indicating cell type (color scheme according to Figure 1B).

Figure 5.

Cell type activity distribution and biomarker response in non-human primates. Cynomolgus macaques received 20 mg ASO N (N = 4) or aCSF (N = 4) at weeks 0, 4, 8 and 12, and cortex and cerebellum were evaluated at week 13. (A, B) PRNP knockdown assessed by bulk qPCR (x axis) versus aggregation of single cell sequencing regardless of cell type (y axis), for individual animals (points) and groups (crosshairs indicate means and 95% confidence intervals) in cortex (A) and cerebellum (B). (C) PrP protein measured by in-house ELISA (56) in brain parenchyma (cortex, cerebellum) and in cerebrospinal fluid (CSF). (D) Residual PRNP by cell type in cortex and cerebellum. (E, F) Scatterplot of cynomolgus 13-week residual PRNP by cell type (x axis) versus mouse residual Prnp at 2 weeks after a single dose of ASO 6 (data from Figure 3A) in cortex (E) and cerebellum (F). The data point for pericyte/fibroblast reflects the weighted average of residual Prnp in these two cell populations in mouse. Each point is a cell type, sized logarithmically by total number of cells (cynomolgus + mouse datasets) and colored as in Figure 1.

Figure 6.

Comparison of cell type target engagement profiles across all conditions examined. (A) Top: differences from overall residual (each cell type's residual target minus the overall residual target quantified from single-nucleus data; y axis) for every combination of tissue, ASO, timepoint, and species (shared x axis). Each point is a cell type, sized logarithmically by number of cells and colored as in Figure 1. Bottom: weighted standard deviations (weighted by number of cells) in percentage points of residual target (y axis) for each condition (shared x axis). (B, C) Correlograms (weighted Pearson's correlations) between every pair of datasets for cortex (B) and cerebellum (C). Colors represent the value of the correlation coefficient rho and the outline represents the nominal P value, see legend at far right. (D) Differences from overall residual for all neuronal subtypes in all conditions. Error bars are 95% confidence intervals from the negative binomial models for the subtype-specific residuals.