Morphological characterization of cortical reactive astrocytes in response to traumatic brain injury

In this study, we employed the StarTrack genetic lineage tracing method to investigate the heterogeneous reactive responses of cortical astrocytes in a model of traumatic brain injury (TBI). The StarTrack method allows the stochastic expression of six different fluorophores expressed in each of the two main cellular compartments, the cytoplasm, and the nucleus. This gives rise to a total of 12 possible combinatorial expressions of colors, providing a unique ID color code to identify astrocyte clones [24]. Here, we analyzed the structural modifications of control and reactive astrocytes subpopulations and established morphology-based clusters to categorize astrocytes reactive responses (Fig. 1A).

Fig. 1

StarTrack reactive astrocyte subpopulations in a model of traumatic brain injury (TBI). A. Experimental design. Mice at embryonic day 14 (E14) were in utero electroporated with StarTrack. At postnatal day 50 (P50, n = 2) or 90 (P90, n = 2) mice were submitted to a model of TBI at the somatosensory cortex. Two control mice at postnatal day 30 were included (P30, n = 2). Seven days post-injury (7dpi) we analyzed structural modifications and established a morphology-based clustering method to categorize the reactive responses of astrocytes. B. Representative images of the somatosensorial cortex of control (1,2) and TBI (3,4) StarTrack mice. C. Representative images of CUX1/DAPI immunohistochemistry. CUX1 was used to delineate upper cortical layers (II-IV; 1,2). Representative images of StarTrack astrocytes, and GFAP immunohistochemistry (3,4,5). The images are from TBI mice. D and E. Astrocyte subpopulations along the corpus callosum (cc), cortex (Cx) and pia mater (PM) from control (D) and TBI (E) mice. We identified pial (1), protoplasmic upper and lower layers (1,2,3), juxtavascular (3) and fibrous (4). TBI Traumatic brain injury, Pt. Up. Protoplasmic upper layers, Pt. Low. Protoplasmic lower layers. Scale bar B1-4, 200 μm; C1-5, 200 μm and 4–6, 100 μm; D1-4, E1-4, 100 μm

First, we identified distinct astrocyte subpopulations in both control (Fig. 1B, 1–2) and TBI (Fig. 1B, 3–4) groups across the corpus callosum (cc), cortex (Cx) and pia mater (PM). To differentiate these subpopulations throughout the Cx, we used CUX1/GFAP immunolabeling to distinguish between upper (layers II–IV) and lower (layers V–VI) cortical astrocytes (Fig. 1C, 1–5). Additionally, we delineated their location within the injury site using GFAP immunolabeling that separate the reactive gliosis area into the “Area 1: contusion core”, and the surrounding normal-appearing tissue termed “Area 2: pericontusional” (Fig. 4A, 1–3). Subsequently, we classified StarTrack labelled astrocytes from both control (Fig. 1D) and TBI (Fig. 1E) groups, based on their localization and morphological properties of their soma and primary branches as follows: (1) pial astrocytes (Fig. 1D, 1 and Fig. E, 1, arrowheads), also known as marginal or perimeningeal astrocytes, located at the surface and in direct contact with the PM; (2) protoplasmic astrocytes (Fig. 1D, 1–2, and Fig. 1E, 2–3, arrowheads), distributed across layers I to VI; (3) juxtavascular astrocytes (Fig. 1D, 3, and Fig. 1E, 2, arrowheads), attached to cortical blood vessels; and (4) fibrous astrocytes in the cc (Fig. 1D, 4, and Fig. 1E, 4, arrowheads). Our analyses included morphological reconstructions of 118 StarTrack reactive astrocytes (n = 4, Fig. 2A, 1–5 TBI), and 45 StarTrack astrocytes that were used as controls (n = 2, Fig. 2A, 1–5 Control). For each cell, we identified major astrocytic components, soma, and primary and secondary branches (branchlets), to analyze their morphological profile (Fig. 2B). Thus, we employed two- and three-dimensional (2D and 3D) image projections (Fig. 2C and 2D) to assess size- and shape-related parameters such as “cell body area” (μm2; Fig. 2C, 1), “convex hull area” (μm2; Fig. 2C, 2), “perimeter” (μm; Fig. 2C, 3), “circularity” (4π(“area”)/(“perimeter”)2; Fig. 2C, 4), “solidity” (“area”/”convex hull area”; Fig. 2C, 5), total “thickness” (μm3; Fig. 2D, 1), “1&2 branches” (unit; Fig. 2D, 2), total branches “length” (μm; Fig. 2D, 3), “intersections” (unit; Fig. 2D, 4), “radius” or “3D distance” (unit, μm; Fig. 2D, 5) and “complexity” (Fig. 2D, 6). Altogether, this dataset let us to explore astrocyte territories and branching: parameters such as “area”, “perimeter” and “convex hull area”, measured the size of the soma and branches of astrocytes (Fig. 2C, 1–3), while parameters such as “circularity” and “solidity” (Fig. 2C, 4–5) measured their general shape variation. These shape-related parameters utilized ratios to compare the “area” of each astrocyte with its “perimeter”, in the case of “circularity”; and with its “convex hull area”, in the case of “solidity”. Higher values of “perimeter” and “convex hull area” resulted in lower “circularity” and “solidity” ratios, that indicated “less regular” and “less dense or spongiform” astrocyte morphologies. The parameter “thickness” (Fig. 2D, 1) measured the total astrocyte volume, although this parameter is mainly affected by astrocyte branching, as their processes constitute up to 95% of total cell volume [64]. Similarly  parameters “1&2 branches” and “length” (Fig. 2D, 2–3) determined cell branching. In addition, we included Sholl analysis metrics such as “intersections”, “radius” or “3D distances”, and “complexity” or “Sholl intersections profile” (Fig. 2D, 4–6).

Fig. 2

Two- and three-dimensional (2D and 3D) morphometric analysis of astrocyte subpopulations. A. Representative images of control and reactive astrocytes subpopulations: pial (1; TBI, n = 15 cells; control, n = 10 cells); protoplasmic upper layers (2; TBI, A1, n = 19 cells, A2, n = 15 cells; control, n = 10 cells); protoplasmic lower layers (3; TBI, A1, n = 13 cells, A2, n = 22 cells; control, n = 10 cells); juxtavascular (4; TBI, n = 19 cells; control, n = 10 cells); and fibrous (5; TBI, n = 15 cells; control, n = 10 cells). Sample size (n) across astrocyte subpopulations. B. Representative images of single-cell major astrocytic components -soma, primary and secondary branches, and end feet- across astrocyte subpopulations. C. Graphic representation of 2D size- and shape-related parameters: area (1; µm2), convex hull area (2; µm2), and perimeter (3; µm); 2D-shape related parameters: circularity (4; 4π(“area”)⁄(“perimeter”)2) and solidity (5; “area”⁄ “convex hull area”). D. Graphic representation of 3D-size and shape-related parameters: thickness (1; relative to 0.05 threshold, µm3), 1&2 branches (2; unit), length (3; µm), intersections (4; unit), and radius or 3D distance (5; unit, µm); and 3D-shape related parameter, complexity (6; “intersections” as a function of “3D distance”). E. Coefficient of variation (CV) radar graphs across astrocyte subpopulations: pial (1), protoplasmic upper layers (2), protoplasmic lower layers (3), juxtavascular (4), and fibrous (5). Astrocytes subpopulations and parameters were identified as equally significant sources of variability. TBI Traumatic brain injury, Pt. Up. Protoplasmic upper layers, Pt. Low. Protoplasmic lower layers. Scale bar 100 μm

By analyzing these datasets of parameters, we constructed radar graphs with coefficient of variation values (CVs) for each parameter to gain insights into the biological trends among control and TBI groups (Fig. 2E). Interestingly, we found that both, astrocyte subpopulations and morphological parameters contributed equally to variability. Astrocyte variability ranged from 7.70% (CV < 10%, indicating low variability) in control protoplasmic “solidity” (lower layers, Fig. 2E, 3) to 121.8% (CV > 35%, indicating high variability) in pial “thickness” (Fig. 2E, 1). Specifically, subpopulations exhibiting acceptable interindividual variability included control protoplasmic (upper and lower layers; CV < 35%, acceptable; Fig. 2E, 2–3), and juxtavascular astrocytes (Fig. 2E, 4), as well as TBI fibrous reactive astrocytes (CV < 35%, acceptable; Fig. 2E, 5). Parameters with acceptable CV included “perimeter”, “circularity”, “solidity”, “length”, “intersections”, and “radius”, which were efficient in detecting significant changes between control and TBI groups (Fig. 2E).

Comparisons between control and TBI groups revealed that lower layer protoplasmic astrocytes exhibited significant changes in nearly all size- and shape-related parameters, including “area”, “convex hull”, “perimeter”, “circularity”, “solidity”, “length”, “branches”, “intersections”, and “radius” (Fig. 3A, B, 1–10). In contrast, upper layer protoplasmic astrocytes only exhibited significant changes in the size-related parameter “radius” (also interpreted as the “3D distance” (Fig. 3A, and Fig. 3B, 1–10). Distinct responses were also observed among juxtavascular, pial and fibrous astrocytes. Juxtavascular astrocytes showed only significant changes in shape, as measured by “circularity” (Fig. 3A and Fig. 3B, 1–10). Pial and fibrous reactive astrocytes, however, showed changes in both, shape- and size-related parameters: pial astrocytes showed alterations in “circularity”, “solidity”, “thickness” and “radius”, while fibrous astrocytes showed alterations in “area”, “convex hull area”, “solidity”, “length”, and “branches” (Fig. 3A and, Fig. 3B, 1–10). Upon a more comprehensive analysis of the datasets from both the control and TBI groups, we found that under physiological conditions (control group), only the pial lineage exhibited a distinct profile among astrocyte subpopulations. However, following TBI, lower layers protoplasmic astrocytes displayed differential profiles in shape- and size-related parameters across astrocytes subpopulations (data not shown).

Fig. 3

Comparison between control and reactive (TBI) astrocytes. A. Proportion of parameters with significant changes across astroglial subpopulations. Lower layer protoplasmic astrocytes exhibited a significant reactive response in nine out of the ten parameters evaluated. The X axis indicates the number of parameters (absolute values) and pie charts their proportion (relative values). B. Graph bar of 2D and 3D size- and shape-related parameters in control and TBI reactive astrocytes: area (1), convex hull area (2), perimeter (3), circularity (4), solidity (5), length (6), 1&2branches (7), thickness (8), intersections (9), radius (10). The graph shows the sample´s distribution, mean and 95% CI. Comparison between control and reactive astrocytes was performed by unpaired T test, **** p ≤ 0.0001, *** p ≤ 0.001; ** p ≤ 0.01; * p ≤ 0.05. C. XY graph of complexity (Sholl intersections profile) shows the number of intersections along the 3D soma distance. Sholl profile include three segments: an initial growth segment extending approximately 12 to 24 μm from the soma, mainly composed by primary branches; followed by a peak segment ranging between 16 to 28 μm from the soma (dashed lines) that include the highest number of intersections and their critical -distance from the soma- value; and a final decline segment, between 24 to 110 μm to the edge of the last branch. The graph shows the mean and standard deviation. Datasets on graphs in absolute values; pial, protoplasmic upper layers, protoplasmic lower layers, juxtavascular and fibrous, control n = 10 cells; pial TBI n = 15 cells; protoplasmic upper layers, TBI n = 19 cells; protoplasmic lower layers, TBI n = 13 cells; juxtavascular TBI n = 19 cells; fibrous TBI n = 15 cells. Comparison between control and reactive astrocytes was performed by two-way ANOVA and Bonferroni´s multiple comparisons test, **** p ≤ 0.0001, *** p ≤ 0.001; ** p ≤ 0.01; * p ≤ 0.05. TBI Traumatic brain injury, Pt. Up. Protoplasmic upper layers, Pt. Low. Protoplasmic lower layers

Further analysis of “complexity” (Sholl intersections profile) allowed us to identify three distinct segments in astrocyte branching (Fig. 3C, 1–5). These segments included: (1) an initial growth segment extending approximately 12–24 μm from the soma, mainly composed by primary branches; (2) a peak segment ranging between 16 and 28 μm from the soma (dashed lines) that included the highest number of intersections and their critical -distance from the soma- value; and (3) a final decline segment, with a size ranging between 24 and 110 μm to the edge of the last branch. Comparisons between control and TBI groups revealed changes in astrocyte branching density and spatial distribution for protoplasmic (upper: Fig. 3C, 2 and lower: Fig. 3C, 3) and fibrous (Fig. 3C, 5) astrocytes. In particular, protoplasmic reactive astrocytes increased their complexity while fibrous reactive astrocytes showed an opposite response (Fig. 3C). Collectively, our findings highlight distinct profiles among astrocytes subpopulations, revealing specific size- and shape- related alterations in response to the brain damage. The morphological changes observed in upper- and lower-layers protoplasmic astrocytes prompted us to further investigate the relationship between astrocyte reactivity and their spatial distribution.

Influence of the distance to the lesion in reactive astrocytes morphologies

Protoplasmic astrocytes were analyzed based on their location within the cortical layers (see above, Fig. 1C, 1–2) and their distance from the injury, as determined by GFAP immunolabeling. This analysis separated the area of reactive gliosis, named as “Area 1: contusion core”, from the surrounding normal-appearing tissue, named as “Area 2: pericontusional” (Fig. 1C, 3–5 and Fig. 4A). Areas 1 and 2 were stablished using image analysis extending 1000 μm from the injury site, revealing GFAP fluorescence-intensity profiles. This led to define two main areas: Area 1 (A1), characterized by an increased GFAP intensity profile extending approximately 250 μm from the injury site, and Area 2 (A2), immediately adjacent to A1, with a lower GFAP intensity profile (Fig. 4A, 1–2). A1 was identified as a reactive gliosis region, while A2 had a GFAP-normal-appearing profile (Fig. 1C, 3–5 and Fig. 4A, 3).

Fig. 4

Protoplasmic reactive astrocyte distribution along TBI areas and morphometric analysis. A. Representative image of TBI main areas according to GFAP immunolabeling; Pt. Up and Pt.Low (1). Sample size and GFAP fluorescence-intensity profile (2); and representative images of GFAP/StarTrack reactive astrocytes at contusional core (A1; a, b) and pericontusional (A2; c, d) areas (3). The images are from the somatosensory cortex of TBI mice. Fluorescence-intensity profiles revealed an enriched GFAP area (A1) extending approximately 250 μm along the core, followed by a moderate GFAP-normal appearing area (A2). B. Graph bar of 2D and 3D size- and shape-related parameters in control, A1 and A2 protoplasmic (upper and lower layers) reactive astrocytes. The graph shows the sample distribution, mean and 95% confidence interval (CI). Comparison between control, A1 and A2 reactive astrocytes (Pt. Up and Pt. Low.) was performed by ordinary one-way ANOVA test and Tukey´s multiple comparisons test, **** p ≤ 0.0001, *** p ≤ 0.001; ** p ≤ 0.01; * p ≤ 0.05. C. XY graph of complexity (Sholl intersections profile) shows the number of intersections along the 3D soma distance. The graph shows the mean and standard deviation (SD). Comparison between control and reactive astrocytes (Pt. Up and Pt. Low.) was performed by two-way ANOVA and Bonferroni´s multiple comparisons test, **** p ≤ 0.0001, *** p ≤ 0.001; ** p ≤ 0.01; * p ≤ 0.05. D. T-distributed Stochastic Neighbor Embedding (T-SNE) plot and clustering analysis (HDBSCAN) of upper- (1) and lower- (2) layers protoplasmic astrocytes. Datasets on graphs in absolute values; protoplasmic upper layers, control, n = 10 cells, A1, n = 19 cells, A2, n = 15 cells; protoplasmic lower layers, control, n = 10 cells, A1, n = 13 cells, A2, n = 22 cells. Scale A1, 250 μm and A2, 50 μm. Pt. Up. Protoplasmic upper layers, Pt. Low. Protoplasmic lower layers

Our analysis revealed that only A1 lower-layers protoplasmic astrocytes showed distinct reactive responses compared to A2 astrocytes (Fig. 4B). These responses included morphological alterations in both size  and shape-related parameters, such as “area”, “perimeter”, “circularity”, “solidity”, “length”, “intersections”, and “radius” (Fig. 4B). Interestingly, the analysis of “complexity” (Fig. 4C) indicated a group effect for both—control and reactive astrocytes in upper and lower cortical layers, with a particular emphasis on lower-layers astrocytes that displayed larger 3D distances, and complexity-curve shapes with a more regular decline as they moved away from the soma (Fig. 4C).

Pairwise analysis of each shape- and size-related parameters, along with visualization of the cell population in t-distributed Stochastic Neighbor Embedding (t-SNE. Figure 4D) plots, revealed that A1 and A2 upper layers protoplasmic control and reactive astrocytes were randomly distributed in the plot (Fig. 4D, 1). In contrast, most A1 lower layers protoplasmic astrocytes were distinctly clustered. On the other hand, A2 lower layers protoplasmic reactive astrocytes were distributed next to controls, indicating similar morphological profiles (Fig. 4D, 2). Consistent with these observations, further clustering analysis based on the TSNE distribution (Fig. 4D, 2) grouped A2 lower-layers protoplasmic reactive astrocytes with controls.

Reactive response landscape among astrocyte lineages

Lastly, we integrated size- and shape-related parameters to construct morphology-based clusters of reactive responses and explored astrocyte lineage participation (Fig. 5). Initially, we applied a multimodal index analysis (MMI, Suppl. Figure 2A) to guide the selection of parameters for cell clustering, focused on those with MMI´s values greater than 0.55, such as “convex hull area”, “perimeter”, “length”, “thickness”, “intersections”, and “radius” (Suppl. Figure 2A). Subsequent hierarchical clustering (HC, Fig. 5A and Suppl. Figure 1), and principal component analysis (PCA, Fig. 5B) enabled the partition of the dataset into clusters labeled from A to H, which were then categorized based on their reactive responses (Fig. 5A, B and Suppl. Figure 2B). Analysis of variance (ANOVA) confirmed that clusters A-B and C-D exhibited similar parameter profiles, while clusters E, F, G, H displayed greater variability (Fig. 5C and Suppl. Figure 2B). Principal component 1 (PC1, 63.6%) and Principal component 2 (PC2, 17.6%) collectively explained 81.2% astrocyte variation in a bidimensional distribution, with optimal correlation values among parameters (Fig. 5B and Suppl. Figure 2C, 1). PC1 included measurements of the size-related parameters “length”, “intersections”, and “perimeter”; PC2 predominantly received contributions from “thickness” and “radius”. The PC biplot displayed how selected morphological parameters influenced and contributed to the direction of the datasets (Suppl. Figure 2C, 1). Upon conducting a PC biplot on relevant datasets, it became evident that the distribution of the plot was influenced by upper- and lower-layers protoplasmic astrocytes (Suppl. Figure 2C, 2).

Fig. 5

Clustering of reactive astrocytes responses. A. Heatmap showing the hierarchical clustering (HC) of reactive astrocytes subpopulations. B. Principal component analysis (PCA) plot with confidence ellipses of reactive astrocytes and clusters distribution; graphical representation of selected parameters. C. Plot profile (normalized mean) of selected parameters among clusters. D. Logical tree of reactive astrocytes response-clusters according to their somatic and branching complexity. E. Floating-bars showing the mean and 95% CI for each cluster. F. Bubble graph of astrocytes subpopulation frequency among reactive responses. The size of the bubble corresponds to the relative frequency (%). G. Representative images of reactive astrocytes responses for juxtavascular subpopulations. Datasets on PCA (B), parameters profile (C) and cell enrichment graphs (F) in relative values, data sets on somatic and branching complexity graphs (E) in absolute values. Pial n = 15 cells; protoplasmic upper layers, n = 19 cells; protoplasmic lower layers, n = 13 cells; juxtavascular n = 19 cells; fibrous n = 15 cells. Scale bar 50 μm

Next, we built a dendrogram (logical tree) to establish diversification nodes for astrocyte reactive responses based on astrocyte clusters, individual features, and type of reactive responses (Fig. 5A–D). The dendrogram was constructed using a hierarchical clustering approach that identified diversification nodes based on a 95% confidence interval (CI) for each parameter to suggest cut-off values (Fig. 5D, E). Parameters such as “perimeter”, “intersections”, and “radius” defined the first-level nodes of the tree. Parameters like “length” and “convex hull area” were used to determine second- and third-level branches nodes, while “thickness” defined fourth-level branches and terminal nodes (Fig. 5D). For the first-level nodes, reactive astrocytes responses were classified into 3 group nodes: “moderate” node, which included astrocytes from clusters C and D (C + D); “strong” node, included clusters A and B (A + B); and “very strong” node that clustered the E to H astrocyte groups (E, F, G, H) (Fig. 5D). Clusters A and B could be further subdivided into in “medium” and “large” second level nodes based on their “convex hull area” values, while clusters E to H, (E, F, G, H) could be subdivided into “short”, “medium”, and “long” based on their processes “length”, and into “large” and “medium” based on their “thickness” values. Interestingly, the reactive response type “very strong” consisted of clusters with greater variability (E, H), primarily characterized by extreme “length” and “thickness” values (Fig. 5D).

We also examined astrocyte diversity within each reactive response (Fig. 5F). The “strong” reactive response (A + B) had the highest proportion of protoplasmic astrocytes from both upper and lower layers, while the “very strong” type exhibited similar lineage participation (Fig. 5F and Suppl. Figure 3D). This diversity was evident through the StarTrack method and astrocytes reconstructions, which allowed us to group different responses based on similar morphologies (Fig. 5G and Suppl. Figure 2D, 1–2). For example, reconstructions and categorizations of reactive astrocytes in two TBI mice (A147 and A117 mice; aged 50, and 90 days respectively) revealed 2 main response models: Model 1 (enriched with “strong” and “very strong” reactive response, primarily from pial, upper-layers protoplasmic and fibrous lineages (Suppl. Figure 2D, 1), the Model 2 displayed proportional types of reactive responses, enriched with pial (moderate), protoplasmic low (strong) and juxtavascular (very strong) lineages (Suppl. Figure 2D, 2).

Finally, we conducted an exploratory analysis to characterize the morphology of different astrocyte responses grouped in our dendrogram using one TBI mouse (A147). This analysis integrated clonal analysis, categorization of reactive responses in astrocyte subpopulations, and their spatial distribution. Our findings revealed that similar reactive responses were localized proximally and at similar distances from the injury (Fig. 6A). Clonal analysis of 281 astrocytes, corresponding to 35 clones with identical fluorophore combinations in both the nucleus and cytoplasm (Fig. 6B), showed that all types of astrocyte subpopulations -pial, upper- and lower- layers protoplasmic, juxtavascular, and fibrous- were present around the injury site. Among these, upper- and lower- layers protoplasmic astrocytes were the most prevalent, followed by fibrous astrocytes, while juxtavascular and pial astrocytes represented smaller fractions (Fig. 6C). Quantitative analysis of the number of sibling cells showed variability in clone size and dispersion across astrocyte subpopulations (Fig. 6C). Additionally, the frequency analysis of astrocyte subpopulations indicated that the “strong” type of reactive response (A + B) had the highest proportion of protoplasmic (upper layers) and fibrous astrocytes. In contrast, the “very strong” reactive response exhibited the highest proportion of pial astrocytes (Fig. 6D). Overall, our data provide a framework for detecting morphological similarities and differences among astrocyte subpopulations, suggesting that TBI induces distinct morphological signatures in reactive astrocytes.

Fig. 6

Exploratory analysis of reactive astrocytes responses in animal model of TBI. A. Representative TBI sections, astrocytes distribution and reactive responses. B. Representative images of the clonal analysis in the TBI model. Astrocytes with the same color code composition were assigned as sibling cells (clones). C. Pie chart, and bubble graphs of the frequency and dispersion of sibling cells (clones) among astrocyte subpopulations. Datasets on pie charts in relative values. The size of the bubble corresponds to the relative frequency (%). D. Bubble graph of astrocytes subpopulation frequency among reactive responses. Datasets on pie charts in relative values. The size of the bubble corresponds to the absolute frequency (units). Scale bar A, 250 μm; and B, 100 μm