Lipopolysaccharide-induced neuroinflammation disrupts functional connectivity and community structure in primary cortical microtissues

Culturing platform for longitudinal live imaging

For our in vitro experiments, we utilized a three-dimensional, self-assembled, primary cortical culture previously shown to contain myriad cell types and produce an endogenous extracellular matrix^13,14. To facilitate live imaging, we adapted a spheroid cortical culture model into a “trampoline” shaped microtissue previously described by Schell et al³⁴. Trampoline microtissues are large multicellular cultures with a heterogeneous cell composition (Supplemental Fig. 1) and a physically stabilized center region, approximately 800 µm wide and 300 µm thick. The trampoline shape was implemented to provide a larger viewable surface area compared to spheroids, increasing the number of imageable cells within the z-axis constraints of the microscope.

As high magnification objectives have very limited optical working distances, 3D cultures typically require physical manipulation to bring them closer to the objective by either transferring into a glass bottom imaging dish or removal of media to prevent samples from floating. As neural cultures are sensitive to environmental changes²⁰, we aimed to collect image data without disrupting the culturing environment. To achieve this, we designed a custom agarose mold to place the cells within 350 µm of the bottom of the plate and, importantly, within the working distance of high-power objectives. Our mold is built upon published work utilizing agarose “stamping” for live imaging of spheroids³⁸. We created a custom machined stainless steel injection mold consisting of a hollow tube with an endcap containing 6 machined holes to form the agarose pegs (Fig. 1A). The injection mold is placed into a single well of a multi-well culture plate, suspended by a 3D printed holder (Fig. 1B). Hot liquid agarose can be pushed through the injection mold and into the well with a syringe (Fig. 1C, Supplemental Fig. 2). When the agarose cools, the injection mold can be removed (Fig. 1D), leaving behind the agarose pegs within the microwell (Fig. 1E). After primary tissue collection and dissociation, cells were seeded into the microwell (Supplemental Fig. 3), placing them approximately 350 µm from the bottom of the plate and within the working distance of the objective for live imaging (Fig. 1F).

AAV mediated expression of GCaMP6s reveals spontaneous neural activity

Once microtissues were constructed, we applied a commercially available GECI to visualize neural activity in vitro. The GCaMP6s calcium indicator was expressed through AAV mediated transgene delivery (AAV1-hSyn1-mRuby2-GSG-P2A-GCaMP6s-WPRE-pA, AddGene). Expression of the fluorescent protein packaged in the viral capsid was driven under the human synapsin (hSyn1) promoter, as it produces neuron-specific expression patterns³⁹. The GCaMP6s probe was chosen over GCaMP6f due to its higher signal to noise ratio²⁴, which we expected to compensate for any optical interference from agarose or tissue itself.

After culturing microtissues for 1 day (Fig. 2A), fluorescence expression was achieved via transgene delivery by culturing in virus-containing media for 3 days (Fig. 2B). Initial fluorescence appeared at DIV 8 and progressively increased in brightness through DIV 14 (Fig. 2C). Onset of spontaneous calcium transients were observed and recorded starting at DIV 14 (Fig. 2D). Python code was used to identify cell bodies by applying Laplacian of Gaussian filtering to maximum projections of time-lapse images and non-maxima suppression to detect individual regions of interest (ROIs, Fig. 2E). ROIs were curated to remove cells outside of the microtissue edge using manual interactive correction. Using curated ROI seed positions, ROI masks were created and overlaid onto each image of the time-lapse videos (Fig. 2F) in MATLAB, and calcium transients and events were extracted from the masked videos (Fig. 2G) with the FluoroSNNAP⁴⁰ application.

Microtissues exhibit significant remodeling of functional connectivity during development

We characterized FC of microtissues from onset of spontaneous calcium transients. To accomplish this, we recorded activity in 9 microtissues across three biological replicate litters from DIV14-34. We analyzed network activity from recordings using graph theory analysis (GTA), a mathematical method of describing complex interactions between “nodes” in an interconnected network³¹. Here, we employed GTA to describe in vitro network activity at the microcircuit level, where each node within the graph represents a single cell^32,41. Using a network analysis toolkit⁷, we characterized correlation coefficients, clustering coefficients, and path lengths of microcircuits. To assess trends in neural activity over time, recordings were grouped into weeks (week2 = DIV14-20, week3 = DIV21-27, week4 = DIV28-34), where each data point represents a single recording of a microtissue within the time range.

The correlation coefficient is a metric of FC between pairs of nodes. Correlation was determined by the statistical similarity of activity between each node pair by Pearson cross-correlation. Higher correlation coefficients indicate stronger functional connections between nodes³². We assessed the average correlation of microcircuits, calculated by taking the mean correlation coefficient of all node-pairs. We found a significant (p = 0.0026) reduction in the average correlation from week 2–3 and a significant (p = 0.0151) recovery in week 4 (Fig. 3A).

The clustering coefficient is a measure of FC between triplets of nodes, representing the formation of “small world” network organization⁴². A higher clustering coefficient corresponds to the presence of highly interconnected clusters of nodes, an indication of higher network complexity compared to a random network^31,43. Average clustering coefficients followed the same trend as average correlations over 3 weeks in vitro, with a significant reduction (p = 0.0032) from week 2–3 and a significant (p = 0.016) recovery in week 4 (Fig. 3B).

Path length is a metric of network integration, inversely related to the correlation and representing the ability of nodes to communicate efficiently with limited intermediary nodes^31,44. Node-pairs with short path lengths are thought to be efficient at sharing information. We found a significant increase (p = 0.0054) in the average path lengths from week 2–3 and a significant (p = 0.0263) recovery in week 4, inversely following trends of the correlation and clustering (Fig. 3C). Collectively, the disruption and recovery of correlation, clustering, and path length indicate significant remodeling of FC in microtissues during maturation.

Additionally, we investigated if these changes in FC were associated with changes to the oscillatory activity, comprising synchronous neuronal bursting across the whole microtissue. While synchronous oscillatory activity is a phenomenon related to learning and memory functions²³, it may intrinsically alter correlation coefficients of the network. Thus, we examined the oscillatory behavior over weeks of maturation by calculating the whole-tissue firing rates, measured by the number of tissue-wide synchronous bursts per minute identified by the FluoroSNNAP event detector. While there was a decrease in the average whole-tissue firing rate, the changes were not statistically significant (week2–3, p = 0.5851; week3–4, p = 0.4209; Fig. 3D). This finding suggests that global oscillatory activity is likely not responsible for the changes in functional connectivity at this stage of maturation.

To better understand the changes in connectivity, we examined the longitudinal FC of individual microtissues. While there was no statistically significant change in whole-tissue firing rate from week to week, individual microtissues displayed variability in the firing patterns over time (whole-tissue summed traces; Fig. 3E, Supplemental Fig. 4). Correlograms of a representative microtissue from each week (Fig. 3F) showed that the net reduction in average correlations in week 3 did not reflect a blanket reduction in all node-pairs, marked by a visible increase in correlation for a subset of node-pairs. Correlational connectomes (Fig. 3G), which overlay correlation values above 0.5 onto the physical microtissue, showed this increase in strongly correlated node-pairs (correlation > 0.8) more clearly. This qualitatively supports the idea that microtissues undergo selective pruning of FC during maturation. To investigate the structural composition of the network, we then characterized the correlation of proximal nodes. We found that physical proximity of nodes was not a strong predictor of correlation in individual microtissues (Simple linear regression; DIV14, R² = 0.45; DIV24, R² = 0.030; DIV32, R² = 0.00045, Fig. 3H) indicating that the underlying structural network is likely not a regular network, which is primarily made up of proximal connections. These data further support the presence of significant whole-tissue remodeling over the first 4 weeks in vitro.

Microtissues develop and refine community structure over time

We assessed the strength of microcircuit community structure, characterized by the formation of strongly connected groups of neurons within the functional network. The development of community structure is an important feature of neural network maturation both in vitro and in vivo^19,45,46. Here, we used a compiled network analysis toolkit⁷ to characterize community structure.

Modularity is a metric of community structure maturation defined by the presence of highly connected groups of nodes called “modules”^31,32,41. At the microcircuit level, modules represent small subnetwork ensembles of cells, as opposed to larger regional connections. By examining modules at the microcircuit level, we aim to characterize subtle changes to local network topology, which are more closely related to functional changes at the single-cell, molecular, and synaptic level. Modules can be identified by hierarchical clustering of node-pair correlations^7,47. If modularity increases, network segregation increases, resulting in strong node-pair connections within modules and weak node-pair connections between modules⁴⁷. As such, we would expect to see increased modularity as the network matures through strengthening synapses. Modularity significantly increased (p = 0.046) from week 2–3 (Fig. 4A), despite the decrease in overall correlation, suggesting that functional remodeling coincides with construction of modules. Additionally, microtissue modularity remained high at week 4 (p = 0.0475; Fig. 4A), indicating that the increased correlation did not disrupt the community structure developed in week 3. At this microcircuit scale, modules did not represent physical “regions” within the microtissue. Nodes of different modules and intra-modular connections were interspersed (red, module1; blue, module2; Fig. 4B).

Although the modularity changed over time, the absolute number of identified modules did not (week2–3, p = 0.8449; week3–4, p = 0.6127; Fig. 4C), suggesting that increased modularity may be related to the segregation of established modules rather than further subdivision of modules. To investigate, we compared node-pair correlations within modules (intra-modular correlations) to those between modules (inter-modular correlations). Intra-modular and inter-modular correlations were not significantly different in week 2 (p = 0.0834), but progressively separated in week 3 (p = 0.0045) and 4 (p < 0.0001; Fig. 4D). These data indicating that microtissues at week 2 contained weak community structure in which connectivity within modules was equivalent to those between modules (Fig. 4E). At week 3, the overall FC decrease disproportionately reduced the strength of inter-modular connections, resulting a separation of intra- and inter-modular correlations (Fig. 4E, Supplemental Fig. 5). Further, the overall increase in connectivity in week 4 disproportionately increased intra-modular connections (Fig. 4E, Supplemental Fig. 5), resulting in the emergence of well-defined community structure. Additionally, the number of nodes within strongly correlated modules increased over time (Fig. 4F), with module size showing limited correlation to module connectivity in week 2 (Simple linear regression; R² = 0.01134; slope deviation from zero, p = 0.3320) and week 3 (Simple linear regression; R² = 0.0015, slope deviation from zero, p = 0.7818), but a stronger correlation in week 4 (Simple linear regression; R² = 0.1462; slope deviation from zero, p = 0.0043; Fig. 4F). Collectively these data indicate a progressive strengthening of community structure created through selective pruning of FC over weeks in vitro.

Functional connectivity significantly increases 5–9 days after LPS exposure

We then investigated FC of 3D in vitro neural networks during acute neuroinflammation. We applied the methods of functional imaging and analysis described above to a well-established lipopolysaccharide (LPS) model of acute neuroinflammation. LPS stimulates glial inflammatory cascades, producing changes in cellular morphology¹⁸, gene expression⁴⁸, and secretome³⁵. While LPS-induced neuroinflammation does not model a specific disease pathology, it can add to our understanding of basic neuroinflammatory cellular dynamics. LPS-induced neuroinflammation has significant downstream effects on neural function, including connectivity changes in mesocircuits^49,50,51, the development of epileptiform hyperactivity⁵², and cognitive deficits in rodents and humans^49,53. Here, we examined the effect of LPS on microcircuit FC in microtissues.

As LPS exposure in vivo has effects on the secretome in tissues directly after exposure³⁵, morphology after 2 days¹⁸, and synaptic proteins at one week⁵⁴, we examined FC at three corresponding timeframes (2 h, 1–3 days, and 5–9 days) after exposure. Microtissues were treated with either 10ug/mL of LPS or a PBS control at DIV 25 and imaged at 2 h after treatment (LPS = 9 microtissues; PBS = 8 microtissues, Fig. 5A). Samples were subsequently imaged at 1–3 days (DIV26/27/28) and 5–9 days (DIV30/32/34). Due to the sensitivity of neuronal activity to any environmental changes, microtissues were allowed to recover from media changes for 2 h prior to imaging²⁰.

We found no significant changes in correlation (p = 0.9805; Fig. 5B), clustering (p = 0.9924; Fig. 5C), path length (p = 0.5195; Fig. 5D) or whole-tissue firing rate (p = 0.0636; Fig. 5E) between PBS and LPS treated samples at 2 h. However, firing rates of individual samples before and after treatment showed no change in PBS firing rate (p = 0.4479; Fig. 5F), but a significant reduction in LPS firing rate (p = 0.0249; Fig. 5G). At 1–3 days, we found no significant differences in LPS and PBS correlation (p = 0.1539; Fig. 5H), clustering (p = 0.0769; Fig. 5I), path length (p = 0.0627; Fig. 5J), or firing rate (p = 0.0580; Fig. 5K). Correlational connectomes from representative PBS and LPS treated microtissues show connectivity at 2 days post-treatment (Fig. 5L,M). At 5–9 days, however, we found significant differences in FC between LPS and PBS treated samples. LPS microtissues showed increased correlation (p = 0.0014; Fig. 5N) and clustering coefficients (p = 0.0096; Fig. 5O), and decreased path length (p = 0.0198; Fig. 5P). These changes in LPS FC were not associated with any changes in firing rate (p = 0.2126; Fig. 5Q). Correlational connectomes at 9 days show the drastic increase in FC of LPS samples over PBS controls, representing a trend occurring over days after LPS exposure (Fig. 5R,S, Supplemental Fig. 6).

LPS interferes with community structure development through disruption of selective functional remodeling

Changes in microcircuit community structure is an indicator of neural dysfunction associated with memory deficits^26,28. Here, we investigated the influence of LPS-induced acute neuroinflammation on modularity and the underlying community structure of microcircuits in vitro. Because microcircuit modularity is closely related to single cell dynamics underpinned by synaptic changes, we would expect any LPS-induced dysregulation of subnetworks to be reflected in a reduction of modularity.

At 2 h after treatment, LPS had no significant effect on modularity (p = 0.3074; Fig. 6A), intramodular correlations (p = 0.3056; Fig. 6B), or inter-modular correlations (p = 0.056; Fig. 6C). Schematic representations of module connectivity (Fig. 6D,E), show that at 2 h (DIV25), microtissues have established separation between intra-modular and inter-modular connections in both LPS (p < 0.0001) and PBS (p < 0.0001) samples. At 1–3 days, modularity of LPS remained the same while PBS showed an upward trend (p = 0.0505; Fig. 6F). PBS intra-modular correlations showed a significant increase over LPS samples (p = 0.0070; Fig. 6G), while inter-modular correlations were static (p = 0.2690; Fig. 6H), suggesting that PBS controls continued to strengthen community structure while LPS samples did not. Schematics of connectivity data (Supplemental Fig. 7) show slightly increased intra-modular connections in PBS (intra, p = 0.0622; inter, p = 0.8413; Fig. 6I), while LPS-treated networks remained the same (p = 0.5200; p = 0.8524; Figure J). At 5–9 days, the overall FC increase in LPS samples was associated with a significant reduction in modularity compared to PBS controls (p < 0.0001; Fig. 6K). While we found a significant increase in LPS intra-modular correlation (p = 0.0133; Fig. 6L), there was a disproportionate increase in LPS inter-modular correlation (p < 0.0001; Fig. 6L,M), effectively reducing network segregation. Schematics of connectivity data (Supplemental Fig. 7) show PBS controls maintained well-defined modules at 5–9 days (Fig. 6N), while LPS treated samples reduced the strength and separation of modules (Fig. 6O). The resulting effect of LPS treatment was the disruption of community structure in microcircuits beginning at 1–3 days, and more robustly at 5–9 days after treatment.