Abstinence from cocaine self-administration promotes microglial pruning of astrocytes, which drives cocaine-seeking behavior

Abstract

Rodent drug self-administration leads to a compromised ability of nucleus accumbens astrocytes to maintain glutamate homeostasis as well as to reductions in surface area, volume, and synaptic colocalization of astrocyte membranes. However, the mechanisms driving astrocyte responses to drug administration are unknown. Here, we report that long-access rat cocaine self-administration followed by prolonged home cage abstinence results in decreased branching complexity of nucleus accumbens astrocytes, characterized by the loss of peripheral processes. Using a combination of confocal fluorescence microscopy and immunoelectron microscopy, we show that these alterations in astrocyte structural features are driven by microglial phagocytosis, as virally labeled astrocyte membranes are found within microglial phagolysosomes. Inhibition of complement C3-mediated phagocytosis using the neutrophil inhibitory peptide (NIF) rescued astrocyte structure and decreased cocaine-seeking behavior following cocaine self-administration and abstinence. Collectively, these results provide evidence for microglial pruning of nucleus accumbens astrocytes across cocaine abstinence, which mediates cocaine craving.

Keywords: CP: Neuroscience; astrocytes; cocaine; drug seeking; microglia; nucleus accumbens; phagocytosis; relapse.

Figure 1.. The cocaine-induced atrophic phenotype of NAc core astrocytes is characterized by decreased structural complexity

(A) Experimental timeline. (B) Three-dimensional reconstructions of isolated masks of Lck-GFP-positive astrocytes (left) were used for filament tracing to construct wire models of each astrocyte (middle) to be used for Sholl analysis (right). Scale bar: 20 μm. (C) Number of Sholl intersections from all four groups plotted as a function of distance from the nucleus two-way interactions (distance × model: F(93, 2,604) = 11.54, p < 0.0001; distance × drug: F(93, 2,604) = 10.64, p < 0.0001). Three-way repeated measures (RM) ANOVA. Asterisk signifies statistically significant group difference confirmed with two-way RM ANOVA (expanded on in E and F). (D) The total number of Sholl intersections (left; F(3, 28) = 10.75, p < 0.0001), the peak intersections (maximum intersection value) (middle; F(3, 28) = 6.062, p = 0.0026), and the distance at the peak complexity (right; F(3, 28) = 4.436, p = 0.0113) are all decreased in the LgA-Ab cocaine group. Each data point represents an individual cell (nested factor). One-way nested ANOVA with Bonferroni post hoc test. (E) Sholl intersections, plotted as a function of distance for ShA-Ex (top) and LgA-Ab (bottom), and their respective AUC. ShA-Ex group (top) main effect of the drug: F(1, 14) = 3.883, p = 0.0689; AUC p = 0.0079. LgA-Ab model (bottom) main effect of the drug: F(1, 14) = 20.62, p = 0.0005; AUC p < 0.0001. Two-way RM ANOVA with Bonferroni post hoc test (distance plots) and two-tailed unpaired t test (AUC). (F) Sholl intersections, plotted as a function of distance for saline-saline and cocaine-cocaine groups compared from both behavioral models, and their respective AUC. No effect of the self-administration protocol was observed for the saline control groups (top; main effect of the model: F(1, 13) = 0.0013, p = 0.9719; AUC p = 0.3682), while cocaine-cocaine comparisons revealed a larger magnitude of decrease in complexity in the LgA-Ab cocaine group compared to ShA-Ex cocaine (bottom; main effect of the model: F(1, 15) = 9.734, p = 0.0070; AUC p < 0.001). Two-way RM ANOVA with Bonferroni post hoc test (distance distribution plots) and two-tailed unpaired t test (AUC). Data are represented as the mean ± SEM. *p < 0.05; ns, not significant. Group sizes: ShA-sal, N = 7, n = 6–9; ShA-coc, N = 9, n = 8; LgA-sal, N = 8, n = 8–11; and LgA-coc, N = 8, n = 8–10 (N is the number of animals and n is the number of cells per animal). Detailed statistics for all the interactions in three- and two-way ANOVA are reported in Table S1. All group comparisons for the one-way nested ANOVA are reported in Table S2.

Figure 2.. Cocaine-induced decrease in structural complexity of accumbal astrocytes results from decreased branching of peripheral processes

(A) Wire model of astrocyte processes showing arborizing (left), continuing (middle), and terminating (right) bifurcation nodes. Scale bar: 10 μm. (B) Number of bifurcations, plotted as a function of distance for both behavioral models separately. ShA (top): arborizing (left)—main effect of the drug, F(1, 14) = 6.050, p = 0.0275; continuing (middle)—main effect of the drug, F(1, 14) = 5.992, p = 0.0282; terminating (right)—main effect of the drug, F(1, 14) = 8.144, p = 0.0128. LgA (bottom): arborizing (left)—main effect of the drug, F(1, 14) = 19.20, p = 0.0006; continuing (middle)—main effect of the drug, F(1, 14) = 22.02, p = 0.0003; terminating (right)—main effect of the drug, F(1, 14) = 18.48, p = 0.0007. Two-way RM ANOVA with Bonferroni post hoc test. (C) Number of bifurcations, plotted as a function of distance for drug conditions separately. Saline (top): arborizing (left)—main effect of the model, F(1, 13) = 2.858, p = 0.1148; continuing (middle)—main effect of the model, F(1, 13) = 2.291, p = 0.1541; terminating (right)—main effect of the model, F(1, 13) = 4.217, p = 0.0607. Cocaine (bottom): arborizing (left)—main effect of the model, F(1, 15) = 10.87, p = 0.0049; continuing (middle)—main effect of the model, F(1, 15) = 11.31, p = 0.0043; terminating (right)—main effect of the model, F(1, 15) = 10.71, p = 0.0051. Two-way RM ANOVA with Bonferroni post hoc test. ns, not significant. (D) Type-specific bifurcations of all four groups plotted together as a function of distance. Arborizing (left)—distance × model, F(91, 2,548) = 9.9515, p < 0.0001; distance × drug, F(91, 2,548) = 9.096, p < 0.0001. Continuing (middle)—distance × model, F(91, 2,548) = 9.706, p < 0.0001; distance × drug, F(91, 2,548) = 10.890, p < 0.0001. Terminating (right)—distance × model, F(91, 2,548) = 6.475, p < 0.0001; distance × drug, F(91, 2,548) = 10.32, p < 0.0001. Three-way RM ANOVA. Asterisks signify statistically significant group difference confirmed with two-way RM ANOVA. Insets (total number of type-specific bifurcations): arborizing (left), F(3, 28) = 11.07, p < 0.0001; continuing (middle), F(3, 28) = 11.70, p < 0.0001; terminating (right), F(3, 28) = 10.44, p < 0.0001. One-way nested ANOVA with Bonferroni post hoc test. Data are represented as the mean ± SEM. Significance was set at p < 0.05. Group sizes: ShA-sal, N = 7, n = 6–9; ShA-coc, N = 9, n = 8; LgA-sal, N = 8, n = 8–11; LgA-coc, N = 8, n = 8–10 (N is the number of animals and n is the number of cells per animal). Detailed statistics for all the interactions in three- and two-way ANOVA are reported in Table S1. All group comparisons for the one-way nested ANOVA are reported in Table S2.

Figure 3.. The cocaine-induced decrease in branching of accumbal astrocytes is a result of a loss of peripheral astrocyte processes

(A) Wire model of reconstructed astrocyte showing terminal points of most peripheral segments (a proxy for astrocytic terminals). Scale bar: 10 μm. (B) Total number of terminal points for all groups (F(3, 28) = 10.72, p < 0.0001). One-way nested ANOVA with Bonferroni post hoc test. (C) Number of terminal points for all four groups plotted as a function of distance from the nucleus (distance × model, F(91, 2,548) = 7.776, p < 0.0001; distance × drug, F(91, 2,548) = 10.53, p < 0.0001). Three-way RM ANOVA. Asterisks signify statistically significant group difference confirmed with two-way RM ANOVA. (D) Number of terminal points, plotted as a function of distance for both behavioral models separately (ShA [left]—main effect of the drug, F(1, 14) = 5.533, p = 0.0338; LgA-Ab [right]—main effect of the drug, F(1, 14) = 19.71, p = 0.0006). Two-way RM ANOVA with Bonferroni post hoc test. (E) Number of terminal points, plotted as a function of distance for both drug conditions separately (saline [left]—main effect of the model, F(1, 13) = 2.031, p = 0.1777; cocaine [right]—main effect of the model, F(1, 15) = 10.80, p = 0.0050). Two-way RM ANOVA with Bonferroni post hoc test. ns, not significant. (F) A complete wire model of an isolated astrocyte, used to measure the combined length of the filament and count its segments (inset: a single segment, defined as a part of the filament between two bifurcation nodes). Scale bar: 5 μm. (G) Total length of the filament (F(3, 28) = 15.56, p < 0.0001), number of segments (F(3, 28) = 11.11, p < 0.0001), and average length of the segments (F(3, 28) = 0.4864, p = 0.6945) for all groups. One-way nested ANOVA with Bonferroni post hoc test. ns, not significant. (H) Principal-component (PC) analysis of all the measured astrocyte complexity parameters shows clear separation and clustering between the cocaine and the saline groups for LgA-Ab (bottom), with the PC1 contributing to the separation of the groups (PC1, percentage of variance [POV] = 83.06%, eigenvalue = 9.137; PC2, POV = 9.56%, eigenvalue = 1.052; PCR [active lever presses; ALPs], F(2, 13) = 10.14, p = 0.0022), while in the ShA-Ex model (top), the separation is not clear (PC1, POV = 71.32%, eigenvalue = 7.846; PC2, POV = 17.33%, eigenvalue = 1.906; PCR [ALPs], F(2, 13) = 2.837, p = 0.0949). PCs used in PCR were selected by the Keiser rule. (I) Regression analysis for the LgA-coc group, showing correlation between the number of ALPs during SA and the most relevant complexity parameters of astrocytes: number of segments (left; R2 = 0.6734, F(1, 6) = 12.37, p = 0.0126), number of terminal points (middle; R2 = 0.6720, F(1, 6) = 12.29, p = 0.0127), and total number of all bifurcation nodes (right; R2 = 0.6736, F(1, 6) = 12.38, p = 0.0125). The results are plotted as a simple linear regression with the 95% confidence interval. (J) Simple linear regression analysis for the LgA-sal group, showing the lack of correlation between ALPs during SA and the most relevant complexity parameters of astrocytes: number of segments (R2 = 0.4482, F(1, 6) = 4.841, p = 0.0694), number of terminal points (R2 = 0.4466, F(1, 6) = 4.841, p = 0.0701), and total number of all bifurcation nodes (R2 = 0.4508, F(1, 6) = 4.925, p = 0.0683). The results are plotted as a simple linear regression with the 95% confidence interval. Data are represented as the mean ± SEM. Significance was set at p < 0.05. Group sizes: ShA-sal, N = 7, n = 6–9; ShA-coc, N = 9, n = 8; LgA-sal, N = 8, n = 8–11; LgA-coc, N = 8, n = 8–10 (N is the number of animals and n is the number of cells per animal). Detailed statistics for all the interactions in three- and two-way ANOVA are reported in Table S1. All group comparisons for the one-way nested ANOVA are reported in Table S2. See also Figure S1.

Figure 4.. There were no detected microglial morphological changes at AD1 or AD45 following LgA cocaine self-administration

(A) Whole-brain slice showing the nucleus accumbens, with expression of AAV5-GfaABC1D-Lck-GFP (top). Microglia were examined from within the white square. Timeline of self-administration and tissue collection points for microglia analysis (bottom). (B) Microglial cells were counted at AD1 and AD45. The images on the right depict 20× representative fields of view (FOV) images of Iba1-stained microglia. (C) Cell counts were measured at each time point by looking at cells/FOV. AD1 sal vs. coc, F(1, 14) = 2.115, p = 0.0528; AD45 sal vs. coc, F(1, 14) = 2.272, p = 0.0394. (D) Individual microglial cell stained with Iba1 (top). Three-dimensional reconstruction of the same cell using Bitplane Imaris software showing the soma in purple and reconstruction of processes in green (middle). The reconstructed cell shows branchpoints (magenta) and terminal points (white, bottom). (E) Microglial morphometric measurements were collected, and a few chosen graphs are shown from left to right: surface area, volume, number of segments, number of terminal points, and soma sphericity. Surface area sal vs. coc: AD1, F(1, 14) = 0.8313, p = 0.4197; AD45, F(1, 18) = 0.06991, p = 0.9450. Volume sal vs. coc: AD1, F(1, 14) = 0.5609, p = 0.5837; AD45, F(1, 18) = 0.1614, p = 0.8736. Number of segments sal vs. coc: AD1, F(1, 14) = 1.017, p = 0.3265; AD45, F(1, 18) = 0.5116, p = 0.6152. Number of terminal points sal vs. coc: AD1, F(1, 14) = 0.9016, p = 0.3825; AD45, F(1, 18) = 0.4314, p = 0.6713. Soma sphericity sal vs. coc: AD1, F(1, 14) = 1.735, p = 0.2089; AD45, F(1, 18) = 0.9626, p = 0.3485. (F) Sholl analysis was also conducted with the shown graphs of AD1 and AD45. (G) The accompanying quantification for Sholl analysis is shown in the graph, depicted by total number of Sholl intersections. (H) PCA of all the measured microglia morphometric and complexity parameters fails to show clear clustering between cocaine and saline groups at AD1 (PC1, POV = 65.48%, eigenvalue = 7.858; PC2, POV = 16.90%, eigenvalue = 2.029). (I) PCA of all the measured microglia morphometric and complexity parameters fails to show clear clustering between cocaine and saline groups at AD45 (PC1, POV = 61.48%, eigenvalue = 7.378; PC2, POV = 17.48%, eigenvalue = 2.098). *p < 0.05 between groups; all error bars are standard error of the mean (SEM). Data are represented as the mean ± SEM. Significance was set at p < 0.05. Group sizes: AD1-sal, N = 8, n = 8–14; AD1-coc, N = 8, n = 7–17; AD45-sal, N = 10, n = 4–14; AD45-coc, N = 10, n = 4–26 (N is the number of animals and n is the number of cells per animal). Detailed statistics for all microglial morphological parameters analyzed are reported in Table S3.

Figure 5.. Cocaine self-administration and abstinence lead to increased astrocyte-microglia interactions

(A) Representative images of isolated microglia (Iba1) from two self-administration models and microglia-associated astrocyte processes (Lck-GFP) as well as microglia-colocalized CD68 and C3 signal. Scale bar: 10 μm. (B) Colocalization analysis for Iba1/CD68 (top; AD1, F(1, 14) = 0.6292, p = 0.4409; AD45, F(1, 18) = 7.215, p = 0.0151), Iba1/Lck-GFP (middle; AD1, F(1, 14) = 1.950, p = 0.1844; AD45, F(1, 18) = 13.98, p = 0.0015), and Iba1/C3 (bottom; AD1, F(1, 14) = 0.9589, p = 0.3441; AD45, F(1, 18) = 5.242, p = 0.0343). Two-tailed nested t test. ns, not significant. Data are represented as the mean ± SEM. Significance was set at p < 0.05. Group sizes: AD1-sal, N = 8, n = 8–14; AD1-coc, N = 8, n = 7–17; AD45-sal, N = 10, n = 4–14; AD45-coc, N = 10, n = 4–26 (N is the number of animals and n is the number of cells per animal).

Figure 6.. Increase in the microglia-astrocyte colocalization following cocaine SA and prolonged abstinence is a result of microglial phagocytosis

(A) Orthogonal projections of isolated microglia showing Iba1/Lck-GFP, Iba1/Lck-GFP/C3, and Iba1/Lck-GFP/C3/CD68 colocalization (left half). Scale bar: 5 μm. The same isolated microglia, 3D reconstructed with transparent surface, showing that colocalized GFP and CD68 signals are located inside the microglial cytoplasm and not merely colocalized with the cell surface/membrane (right half). Scale bar: 5 μm. CD68⁻ and GFP-positive microglial inclusions can be found in soma (red boxes) as well as in the peripheral microglial processes (blue boxes). Box size: 10 × 10 μm. (B) The number of total GFP-positive inclusions per microglia (I): AD1, F(1, 14) = 0.0294, p = 0.8663; AD45, F(1, 18) = 6.076, p = 0.0240. The number of GFP-positive inclusions per microglial soma (II): AD1, F(1, 14) = 5.970, p = 0.0284; AD45, F(1, 18) = 4.837, p = 0.0412. Total volume of GFP-positive inclusions per microglia (III): AD1, F(1, 14) = 0.0133, p = 0.9098; AD45, F(1, 18) = 17.26, p = 0.0006. Total volume of GFP-positive inclusions per microglial soma (IV): AD1, F (1, 14) = 0.1626, p = 0.6928; AD45, F(1, 18) = 14.61, p = 0.0012. Percentage of the total microglia volume occupied by the GFP-positive inclusions (V): AD1, F(1, 14) = 1.067, p = 0.3192; AD45, F(1, 18) = 16.00, p = 0.0008. Percentage of the total microglia soma volume occupied by the somatic GFP-positive inclusions (VI): AD1, F(1, 14) = 0.0002, p = 0.9890; AD45, F(1, 18) = 35.01, p = 0.001. Two-tailed nested t test. ns, not significant. (C) Three-dimensional reconstruction of a representative microglia (AD45-coc group) showing opaque surface of the soma and primary microglial processes (I). The same reconstruction with transparent surface, showing CD68-positive phagosome with opaque surface inside microglial soma (II). The same reconstruction with the transparent surface of the CD68-positive phagosome, showing GFP-positive astrocyte inclusions inside the microglial phagosome (III). SEM micrograph of a microglia (yellow) from the AD45-coc group containing a phagosome with a GFP-positive inclusion (IV). A nearby neuron is indicated in red. Scale bar: 3 μm. Inset (blue box): cropped image of the GFP-positive inclusion (red arrow) inside the microglial phagosome. Scale bar of inset: 1 μm. (D) Principal-component analysis of all the measured microglial astrocyte-derived inclusion parameters shows clear separation and clustering between the cocaine and the saline groups for the AD45 group (bottom), with both PCs contributing to the separation of the groups (PC1, POV = 51.74%, eigenvalue = 5.691; PC2, POV = 30.42%, eigenvalue = 3.346; PCR [infusions], F(2, 17) = 17.47, p < 0.0001), while at the AD1 time point (top), the separation is not clear (PC1, POV = 62.88%, eigenvalue = 6.916; PC2, POV = 17.72%, eigenvalue = 1.949; PCR not performed). PCs used in PCR were selected by the Keiser rule. (E) Linear regression analysis for the AD45-cocaine group (left), showing correlation between the number of active lever presses (ALPs) during self-administration and astrocyte inclusion-related parameters of microglia: CD68/Iba1 colocalization (R2 = 0.4804, F(1, 8) = 7.397, p = 0.0263) and C3/Iba1 colocalization (R2 = 0.5015, F(1, 8) = 8.049, p = 0.0219) were positively correlated with ALPs (top), while the total number of astrocyte-derived inclusions (R2 = 0.5933, F(1, 8) = 11.670, p = 0.0091) and the total volume of inclusions (R2 = 0.4639, F(1, 8) = 6.922, p = 0.0301) were correlated inversely (bottom). None of the inclusion measurements were correlated in the AD45-saline group (right): CD68/Iba1 colocalization (R2 = 7.465 × 10⁻⁵, F(1, 8) = 0.0006, p = 0.9811), C3/Iba1 colocalization (R2 = 0.0004, F (1, 8) = 0.0033, p = 0.9550), the total number of astrocyte-derived inclusions (R2 = 0.1386, F(1, 8) = 1.287, p = 0.2894), and the total volume of inclusions (R2 = 0.1940, F(1, 8) = 1.925, p = 0.2027). Plotted as a simple linear regression with the 95% confidence interval. Data are represented as the mean ± SEM. Significance was set at p < 0.05. Group sizes: AD1-sal, N = 8, n = 8–14; AD1-coc, N = 8, n = 7–17; AD45-sal, N = 10, n = 4–14; AD45-coc, N = 10, n = 4–26 (N is the number of animals and n is the number of cells per animal).

Figure 7.. Blocking microglia-mediated astrocyte pruning restores astrocytic synaptic coverage and decreases cocaine seeking

(A) Timeline for self-administration and intra-NAc NIF infusions. (B) Representative image of the NAc core showing location of cannula placement. Scale bar: 100 μm. (C) The number of cocaine infusions across 10 days of LgA self-administration, showing no difference between groups (main effect of treatment: F(1, 15) = 1.069, p = 0.3176). (D) The number of lever presses during cocaine seeking test on AD24. Animals that received intra-NAc infusions of NIF peptide once weekly during abstinence had a significant reduction in the number of lever presses (Mann Whitney U = 14.50, p = 0.0380). (E) Representative images of an Lck-GFP-labeled astrocyte (left), PSD95 (middle left), and resulting merge of Lck-GFP and PSD95 (middle right) from vehicle- (top left) or NIF- (bottom left) treated animals. Representative colocalization of PSD95 within the boundaries of Lck-GFP astrocytes (right). Scale bars: 10 μm. (F) Astrocyte surface area quantified in vehicle- and NIF-treated animals. Nested t test t(17) = 4.343, p < 0.001. (G) Astrocyte volume quantified in vehicle- and NIF-treated animals. Nested t test t(17) = 5.002, p < 0.001. (H) Number of PSD95 spots colocalized to astrocytes in vehicle- and NIF-treated animals. Nested t test t(17) = 5.360, p < 0.001. Data are represented as the mean ± SEM. Significance was set at p < 0.05. Group sizes: vehicle, N = 9–10, n = 7–9; NIF, N = 8–9, n = 8 (N is the number of animals and n is the number of cells per animal).