A commercially available female hiPSC line (NIBSC-8, UK Stem Cell Bank) was maintained at 37 °C, with 5% CO2 and 5% O2 according to the manufacturer’s recommendations in mTESR Plus medium (STEMdiff™, Catalog #100-0276). It was >90% triple positive for standard iPSC markers (OCT4, SOX2, SSEA4), as evaluated with immunofluorescence. All cell batches were confirmed to be mycoplasma-free and contained no chromosomal aberrations, as evaluated with karyotyping (ThermoFisher Scientific Karyostat+). The cell line was authenticated with short tandem repeat profiling (Johns Hopkins Genomic Resources Core Facility).

Microglial differentiation was performed in three phases. First, hiPSCs (passage 13–17) were differentiated into hematopoietic progenitor cells (HPCs) utilizing the STEMdiff™ Hematopoietic Kit (STEMCELL Technologies™, Catalog #05310), according to the manufacturer’s recommendations. Subsequently, HPCs were differentiated into premature microglia (PMs) with the STEMdiff™ Microglia Differentiation Kit (STEMCELL Technologies™, Catalog #100-0019), with terminal maturation, using the STEMdiff™ Microglia Maturation Kits (STEMCELL Technologies™, Catalog #100-0020). This protocol uses a single medium to differentiate HPCs into PMs over 24 days, with the addition of a final supplement to initiate maturation over an additional 4–12 days. For all integration experiments, premature microglia (PM, day 24 of microglia differentiation) were used, whereas for monoculture experiments, day 24 premature microglia were matured for 8 days prior to assays.

Neural progenitor cells (NPCs) were differentiated as previously described (Romero et al., 2023). Briefly, hiPSCs (passages 7–15) were maintained in mTESR Plus medium on Vitronectin (ThermoFisher Scientific, Catalog# A14700)-coated plates (ThermoFisher Scientific, Catalog #08-772-49). They were then plated at low density on Matrigel (Corning, Catalog# 354277)-coated plates (ThermoFisher Scientific, Catalog #140675) and differentiated into NPCs using Gibco neural induction medium (ThermoFisher Scientific, Catalog #A1647801) for 7 days, following the manufacturer’s instructions. NPCs were then cultured in Neural Expansion Medium, made of Neurobasal/ADVANCED DMEM/F12 media and induction supplement (ThermoFisher Scientific, Catalog # A1647801), as previously described (Romero et al., 2023), for at least five passages with stable and homogeneous morphology prior to all experiments. NPC were differentiated at 37 °C, with 5% CO2 and 5% O2 and then moved to a 20% O2, 5% CO2 incubator after passage 5. Banked NPCs were over 90% double positive for NPC markers (SOX2, Nestin), as confirmed by immunofluorescence (Romero et al., 2023).

To produce a standard brain MPS, we followed our previously published protocol (Romero et al., 2023) and refer to it here as bMPS^6-6, indicating that the MPS was generated and grown in 6-well plates. Briefly, bMPS^6-6 were established by plating 2×10⁶ NPCs (no higher than passage 15) in non-coated 6-well plates (Corning, Catalog # 351146), then maintained under constant gyratory shaking (88 rpm and 19 mm orbit) to enable sphere formation. After 2 days, the Neural Expansion Medium was changed to Differentiation Medium (Neurobasal^® Plus Medium (ThermoFisher Scientific, Catalog # A3653401) supplemented with 1x B-27 Plus (ThermoFisher Scientific, Catalog # A3582801), 1% GlutaMAX (ThermoFisher Scientific, Catalog #35050061), 0.01 μg/mL human recombinant GDNF (GeminiBio, Catalog# 300-121P-010), 0.01 μg/mL human recombinant BDNF (GeminiBio, Catalog #300-104P-1MG)), and 1% pen/strep/glutamine (ThermoFisher Scientific, Catalog#10378016) to support cellular differentiation and maturation. Cultures were maintained at 37◦C, 5% CO2 and 20% O2 for up to 9 weeks, with two-thirds of the medium exchanged every 2 days.

We assayed three different methods for generating microglia-containing MPS (μbMPS) (Supplementary Figure S2). Integration technique 1: bMPS^6-6 was generated as previously described (2×10⁶ single-cell suspension of NPC added to a 6-well plate with gyratory shaking), then grown for 3 days. Subsequently, PMs were added (3×10⁵/well) to the wells. For the first 24 h post-integration, plates were incubated statically at 37 °C with 5% CO₂ with brief shaking every 6 h by an automatic timer to facilitate attachment of the PMs to the bMPS without organoid aggregation. After 24 h, cultures were returned to constant gyratory shaking (88 rpm, 19 mm orbit) for 4 weeks. Integration technique 2: 3×10⁵ PMs and 2×10⁶ NPCs were added to each well of a non-coated 6-well plate. Cultures were maintained with constant gyratory shaking at 88 rpm and a 19 mm orbit for 4 weeks. Integration technique 3 was modified from Xu et al. (2021); 3,000 PMs and 7,000 NPCs per well were added to each well of a non-coated 96-well plate. The following plates were tested: V-bottom plates (CELLSTAR 96-well V-shaped-bottom microplate, Greiner Bio-One, Catalog #GB651204), U-bottom plates (96-well clear round bottom ultra-low attachment microplate, ThermoFisher Scientific #50-211-3787), and Biofloat plates (BIOFLOAT™ 96-well plate #F202003, faCellitate). An additional plate (AggreWell 400 6-well plate, STEMCELL Technologies™, Catalog #34460) was included, as the microwell design could potentially facilitate a higher yield of MPS. AggreWell cultures were plated with a seeding density of 2.4×10⁵/well PMs and 5.6×10⁵/well NPCs per well. All co-cultures were gently pipetted up and down and maintained statically at 37 °C with 5% CO₂ for up to 4 days in Neural Expansion Medium to form μbMPS.

To assess aggregation, 14 individual bMPS were randomly selected on day 4 and imaged with an EVOS XL Core Imaging System microscope (ThermoFisher Scientific). Size conformity was measured automatically using FIJI OrgM, and cellular processes radiating from the bMPS were manually counted by a blinded observer. The optimal plate was further assessed for inter-plate consistency by repeating the above experiment in triplicate.

For the remaining longitudinal and functional studies, U-bottom 96-well plates (ThermoFisher Scientific #50-211-3787) were selected, where PMs and NPCs were seeded in a 50:50 ratio with 1,000 PMs and 1,000 NPCs per well to form μbMPS^96-6. NPCs-only controls containing 2,000 NPCs/well were used for comparison (bMPS^96-6). After 3 days in 96-well plates with Neural Expansion Medium, all 96 μbMPS^96-6 or bMPS^96-6 cultures were transferred to a single well of a non-coated 6-well plate. The medium was then replaced with bMPS Differentiation Medium, and cultures were maintained under gyratory shaking (88 rpm, 19 mm orbit) for 9 weeks. These cultures were designated as μbMPS^96-6, where μ represents microglia inclusion, and 96-6 refers to their initial generation in 96-well plates followed by maintenance in 6-well plates. Unless otherwise noted in the figure caption, all experiments were conducted using a 50:50 ratio of NPCs and PMs.

In summary, we will refer to the following cultural conditions throughout the text:

Organoids—These are used throughout the introduction and discussion sections to refer to published protocols that identify their model as organoids, in accordance with published nomenclature consensus for nervous system organoids (Pașca et al., 2022). bMPS is used in our study to include both neural organoids and microglia-containing assembloids.

bMPS were washed twice with ice-cold PBS, and all supernatant was removed before being snap-frozen in liquid nitrogen. Samples were stored at −80 °C before RNA extraction (Quick-RNA Miniprep Kit, Zymo Research, Catalog # R1054). RNA quantity and purity were determined using a NanoDrop 2000c (Thermo Fisher Scientific, Catalog # ND-2000C). 500 ng of RNA was reverse transcribed using the MLV Promega RT (Promega, Catalog #M1701) according to the manufacturer’s recommendations. Gene expression was evaluated using TaqMan^® Gene Expression Assays (Supplementary Table S1, Life Technologies) on a 7,500 Fast Real-Time system (Applied Biosystems). Fold changes were calculated with the 2^−ΔΔCt method (Schmittgen and Livak, 2001). β-ACTB and 18S were used as housekeeping genes. Data are presented as mean ± SD, normalized to housekeeping genes and age-matched bMPS^6-6.

Samples were washed with ice-cold PBS and fixed with 4% paraformaldehyde in PBS. They were permeabilized with 0.05% Tween-20 and 0.05% Triton X-100 in PBS for 50 min on ice, followed by blocking with BlockAid for 1 h (Invitrogen, Catalog #B10710). Subsequently, samples were immunostained with primary antibodies (Supplementary Table S2) for 24–48 h at 4 °C with rotation. After three washes (1% Triton X-100, 0.5% BSA w/v in PBS), samples were incubated with secondary antibodies (bMPS were incubated overnight at 4 °C with rotation; monocultures were treated for 1 h at room temperature). Following three additional washes and staining with Hoechst33342, samples were transferred to glass microscope slides, mounted with ImmunMount (ThermoFisher Scientific, Catalog # 9990402), and imaged using LSM800 confocal microscope (63x Airyscan), LSM880 confocal microscope (63X and 20x), Olympus FVS3000R confocal microscope (100x and 20x), or Yokogawa C1Q dual-spinning disk High-content Imaging microscope (10X, 20X, and 63X), as indicated in the figure captions. Live-cell images were also obtained during medium changes throughout the culture period using an EVOS brightfield microscope. For 2D monoculture samples, cells were plated and fixed in 24-well glass-bottom plates with high-performance #1.5 coverslips (Cellvis, Catalog # P24-1.5H-N) and imaged using LSM800 (63X with Airyscan).

To perform the live-labeling of premature microglia, a 2.5 mg/mL stock solution of DiD’ oil (ThermoFisher Scientific, Catalog #D307) was prepared in 100% DMSO. At the time of labeling, 1 μL of DiD’ oil was mixed with 1 mL of cell suspension and incubated for 10 min at 37 °C. Following incubation, cells were pelleted at 302 g for 3.5 min, washed with Knockout DMEM (ThermoFisher Scientific, Catalog #10-829-018), and resuspended in Neural Expansion Medium. Labeled PMs were then combined with NPCs as described above. The labeled microglia were visualized live using far-red fluorescence imaging on a Yokogawa C1Q microscope maintained at 37 °C with 5% CO₂.

μbMPS with DiD’-oil-labeled microglia were fully dissociated as previously described (Romero et al., 2023; Morales Pantoja et al., 2024). Briefly, bMPS were washed twice with Hibernate-E (Thermofisher Scientific, Catalog #A1247601), then incubated one per well in a 24-well plate with Papain and DNase1 (Worthington Biochemical Corporation, CAS: 9035-91-1), prepared according to the manufacturer’s instructions, under gyratory shaking. The cells were gently pipetted every 30 min and returned to gyratory shaking until complete disassociation was achieved (approximately 4 h). The contents of each well were transferred to a 15-mL conical tube, centrifuged at 302 g for 4 min, and the supernatant removed. The pellet was resuspended in 1 mL Hibernate-E and incubated with 1 mL of ovomucoid-albumin (Worthington Biochemical, CAS: 9035-91-1), prepared according to the manufacturer’s instructions, to inactivate papain. Nuclei were labeled with Hoechst 33342 (ThermoFisher Scientific, Catalog #62249), and all cells were plated onto a single coverslip. Slides were imaged on a Yokogawa C1Q, and total cell number per organoid, as well as the percentage of DiD’-positive microglia, were quantified across three μbMPS.

Synaptic density was evaluated per region of interest (ROI) (full image taken at 100x), with synapses identified by the colocalization of pre- and post-synaptic markers using the SynapseJ plugin in ImageJ (Moreno Manrique et al., 2021). Briefly, calibrated images were smoothed with a median blur, puncta were identified by thresholding and size exclusion, and ROIs without pre- and post-synaptic marker overlap were eliminated.

Microglia morphology was analyzed in ImageJ using the “Microglia Morphometry” plugin (Martinez et al., 2023). Briefly, calibrated images were smoothed and binarized using user-defined intensity thresholds. Individual microglia were then automatically segmented, and morphology parameters were quantified, including diameter and perimeter measurements, along with skeleton-based features such as average branch length, roundness, circularity versus elongation, and branch complexity.

Calcium imaging was performed as previously described (Alam El Din et al., 2025). Briefly, samples were placed at 37 °C and 5% CO₂ without shaking and incubated with 10 μM Fluo-4 AM (Tocris, Catalog # 6255) and 0.5% Pluronic Acid (Invitrogen, Catalog # P3000MP) for 2 h. Samples were then washed with differentiation medium three times and placed in a glass-bottom dish for live imaging. The samples were imaged using an Olympus FV3000-RS at a speed of 12.5 frames per second for 6 min while maintaining a temperature of 37 °C and 5% CO₂ throughout. The data were processed and quantified as previously described (Alam El Din et al., 2025).

Samples were plated and recorded as previously described (Alam El Din et al., 2025). Briefly, both MaxWell Biosystems 6- and/or 24-well HD-MEA chips were coated for 1 h at 37 °C with 5% CO₂ using a 0.07% Poly(ethyleneimine) solution (Millipore Sigma catalog # 03880) diluted in 1x borate buffer (ThermoFisher Scientific, Catalog # 28341). The HD-MEA chips were then washed three times with water and dried in a sterile hood for 1 h. A solution of 0.04 mg/mL mouse laminin was diluted in differentiation medium and applied to the chips for overnight incubation at 37 °C with 5% CO₂. Afterwards, laminin was removed, and samples were placed on the MEA in differentiation medium, with one bMPS per well. Breathe-EASY membranes (Millipore Sigma, Catalog #Z380059) were used to seal the HD-MEA plates, prevent evaporation, and maintain humidity in the wells. Day on MEA 0 (DOM 0) is defined as the day samples were placed on the chip. The MaxWell Biosystems MaxLab Live Software (version 22.2.6, Switzerland) was used to calculate percent active area, interspike interval within bursts, burst interspike interval, interburst interval coefficient of variation, burst peak firing rate, spikes per burst per electrode, spikes per burst, spikes within bursts, burst frequency, burst duration, and interspike interval outside bursts. Finally, a gain of 512x, a high-pass filter of 300 Hz, and a spike threshold of 5.5 RMS mV were used for all recordings.

To evaluate secreted cytokines, we collected 50 μL of supernatant twice per well at the end of exposures, yielding two technical replicates per biologic replicate, in accordance with the manufacturer’s recommendations. Supernatants were frozen at −80 °C in 96-well plates (Falcon, Catalog #353072) sealed with adhesive aluminum foil (ThermoFisher Scientific, Catalog #AB0626) until measurement. Although the Curiox DropArray miniaturization step requires only 10 μL of supernatant, 50 μL of supernatant was collected and stored at −80 °C to facilitate pipetting and minimize the risk of sample evaporation. On the day of the assay, samples were thawed and mixed by pipetting up and down 8–10 times before the assay. Cytokines were evaluated using the Luminex 25-plex human cytokine panel (ThermoFisher Scientific, LHC0009M) and the Luminex 30-plex human cytokine panel (LHC6003M, Supplementary Table S3), miniaturized with Curiox DropArray technology. Briefly, Curiox DropArray plates were blocked with 1% BSA, followed by three Curiox washes. All washes were performed with 0.5% BSA and 0.05% Tween-20 in PBS. Subsequently, 10 μL of Luminex beads were added to each well, and the plates were kept on magnet arrays throughout all subsequent steps. Standards were prepared using 1:4 serial dilutions of the provided standard solutions. Then, 10 μL of sample supernatant was added to each well, with biological triplicates and technical duplicates, according to the manufacturer’s recommendations. Samples were incubated on a shaker at 4 °C overnight, followed by three Curiox washes. Finally, samples were incubated for 1 h with biotinylated antibody and for 1 h with streptavidin detection antibody, with three washes between each step. Samples were removed from the magnet array, vigorously pipetted with Magpix drive fluid to collect Luminex beads, and transferred to a hard-walled PCR plate (ThermoFisher Scientific, Catalog#AB-0800) for analysis on a Luminex Magpix machine.

Then, 5 pL regression curves were generated from standard curves independently for each cytokine, with standard curves determined per plate to account for inter-plate variability. Extrapolated values beyond the standard ULOQ and LLOQ were accepted. Values below the range were adjusted to be LLOQ or the minimum extrapolated value/SQRT (Matejuk and Ransohoff, 2020). Values above the range were adjusted to the ULOQ or to the maximum extrapolated value*1.5 (Ortega-Villa et al., 2021). Technical replicates were averaged, and cytokine concentrations were normalized to unexposed controls within each group (bMPS^96-6 or μbMPS^96-6).

The traumatic brain injury (TBI) model was adapted from a previously published protocol (González-Cruz et al., 2024). Eight-week μbMPS^96-6 were randomly split into three groups: (1) TBI-challenged (n = 16), (2) no-TBI controls (n = 16), and (3) time 0 controls (n = 4). For TBI induction, μbMPS^96-6 were transferred to microcentrifuge tubes containing 1 mL of medium, centrifuged at 2,862 g-force for 4 min at room temperature, then returned to a 6-well plate on a gyratory shaker and incubated for 4 and 24 h. Batch-matched controls were transferred to a microcentrifuge tube, left static for 4 min at room temperature, and then returned to a 6-well plate on a gyratory shaker and incubated for 4 and 24 h. Time 0 controls were fixed before any manipulations. All cultures were fixed, stained as described above, and imaged using an Olympus FVS3000R microscope.

Two-week-old μbMPS were washed with ice-cold PBS and then snap-frozen in liquid nitrogen. RNA extraction was performed as described above (for qPCR) in triplicate. Samples were transferred to the Johns Hopkins Genomics Core for library preparation and Illumina sequencing. mRNA libraries were constructed by the core facility following the manufacturer’s recommended protocol with the NEBNext^® RNA Library Prep Kit for Illumina sequencing. Samples were then multiplexed and run on a NovaSeq 6,000 instrument at the Johns Hopkins Genomics Core Facility to generate 50 million 150-base pair paired-end reads per sample.

Raw FASTQ data were aligned using the Nextflow nf-core rnaseq pipeline (version 23.10.1) to the GRCh38 genome with ENSEMBL gene annotations (release 111) (Ewels et al., 2020; Martin et al., 2023). Poor-quality sequences and adapters were removed using “Trim Galore!” (Krueger et al., 2023). Transcripts were aligned with STAR and quantified using Salmon (Dobin et al., 2013; Patro et al., 2017). Sorting and indexing of reads were performed with SAMtools, duplicates were identified with Picard, and genome coverage was assessed with BEDtools (Quinlan and Hall, 2010; Li et al., 2009; Picard Toolkit, 2019). RSeQC, Qualimap, dupRadar, and Preseq were used to determine read and mapping quality (Wang et al., 2012; García-Alcalde et al., 2012; Sayols et al., 2016; Computational Genomics Research, 2019). Expression data were normalized using TMM (weighted trimmed mean of M-values) with the tidybulk package and subsequently log2-transformed (Mangiola et al., 2021). Known microglia marker genes were selected from published databases (Miller-Rhodes, 2022), and log2 fold changes were plotted for individual replicates.

Statistical analysis was conducted using GraphPad Prism (Version 10). All data were evaluated for normality, outliers, and homoscedasticity, and non-parametric analyses were utilized when appropriate. Optimizations to select plates included area conformity, the number of processes of μbMPS, and inter-plate consistency, which were assessed by Kruskal-Wallis with post-hoc Dunn’s test. In the longitudinal assessment, qRT-PCR results were analyzed by the Kruskal-Wallis test, and bMPS size over time was evaluated with a two-way ANOVA and repeated Tukey’s multiple comparisons test. Image analysis, when quantified, and cytokine measurements were assessed with ordinary one-way ANOVA with subsequent Dunnett’s multiple comparisons to the control group. A statistically significant difference in synaptic density was evaluated by the Kolmogorov–Smirnov test. An adjusted p-value of <0.05 was considered statistically significant.

A non-scoping literature review was performed using a Google Scholar search with the terms “microglia” AND “organoid” OR “assembloid” OR “microphysiologic system.” Titles were reviewed, and all articles discussing brain or neural organoid systems were selected for further consideration, with a subsequent review of all articles’ abstracts to identify those characterizing a 3D, microglia-inclusive cell culture model (excluding retinal organoids).

Publications were briefly reviewed to exclude documents that were subsequent publications based on already published and described microglia-containing models. Models derived from brain tumor resection or human brain slices were also excluded. Xenotransplantation models were only included if they had an entirely in vitro characterization. Only peer-reviewed publications were included.

Articles were then reviewed in full, with particular attention to the source of microglia (iPSC or other source), whether the organoid media was altered in any way from the standard organoid media used by the authors’ non-microglia model, whether the microglia were present from the formation of organoids, and whether long-term survival was demonstrated in the main figures (long-term considered to be ≥75% of the culture period). Venn diagrams were generated with the publicly available tool jvenn (Bardou et al., 2014).

First, we confirmed the proper differentiation of NIBSC8 iPSCs into mature microglia in monolayer cultures following the STEMCELL Technologies protocols for hematopoietic differentiation, microglia differentiation, and microglia maturation. NIBSC8 hiPSCs differentiated robustly into CD34/43/45-positive hematopoietic precursor cells (Figure 1A) and a highly pure population of microglia cells that expressed mature and specific microglia markers CD11b and Trem2 (Figures 1B,C).

To confirm that the microglia cells were capable of phagocytosis and could be activated, we challenged them with lipopolysaccharide (LPS) and fluorescently labeled E. coli bioparticles at 0, 50, 200, and 500 ng/mL. In vitro exposures to LPS are not an appropriate model for clinical sepsis; however, these doses are within the range used to induce inflammation in other models and are non-cytotoxic (Abreu et al., 2018; Vasilescu et al., 2009; Yücel et al., 2017; Honda and Inagawa, 2023). The pHrodo red E. coli bioparticles contain pH-sensitive fluorescence, thus increasing their fluorescence when internalized by cells into acidic compartments such as the phagolysosome. We demonstrated that microglia cells are capable of phagocytosing bioparticles (Figure 2A). The cells were responsive to LPS in a concentration-dependent manner, showing trends toward ameboid morphology with decreased cell mask area, decreased skeletal branch length, and decreased CD68⁺ (microglial lysosomal or endosomal marker) area, consistent with a trend toward ameboid morphologies (Figure 2B).

To achieve our goal of developing a model that allows for a controllable ratio of microglia within the organoids, which is both consistent and reproducible, we assessed three separate integration methods to identify the most robust technique. Briefly, the three methods are as follows: (1) adding premature microglia to 3-day post-generation bMPS; (2) combining neuronal progenitors and premature microglia directly in a 6-well plate; and (3) adding neuronal progenitors and premature microglia to individual wells of a 96-well plate, allowing aggregation, and then transferring the resulting μbMPS to a 6-well plate (schematized in Supplementary Figure S2A).

To assess integration efficiency, all bMPS from a well were immunostained with the microglia marker IBA1 and the neuronal marker NF200, and the percentage of aggregates containing IBA1⁺ cells was manually counted by a blinded observer. Integration technique 3, which combines PM and NPCs in a 96-well plate before transferring to a 6-well plate on a shaker, was found to be the most robust technique, with 85.7% of bMPS containing microglia. Integration techniques 1 and 2 both resulted in PMs present as single cells in suspension or having formed PM-only spheres, and microglia were visible in only 5.6% or 0% of bMPS analyzed, respectively. Since integration technique 3 allowed precise control of the cell number of integrated microglia with a high level of successful integration, we followed this technique for the subsequent experiments.

To assess the impact of plate type on culture size and conformity, different 96-well plate formats were tested using increased cell densities to facilitate rapid evaluation. 3,000 PMs and 7,000 NPCs were seeded into V-bottom, U-bottom, and Biofloat plates, while a microwell-based AggreWell plate was also evaluated for its potential to enhance bMPS yield and throughput (Figure 3; Supplementary Figure S2B). Across the different plates assessed, μbMPS area and conformity varied significantly (Figure 3A). AggreWell plates failed to enable μbMPS formation of a noticeable size. V-bottom plates generated the largest and most variably sized μbMPS. Only U-bottom plates and Biofloat plates produced μbMPS with reproducible area (p > 0.999); however, there was higher inter-μbMPS consistency with U-bottom plates (Figure 3A). Additionally, some μbMPS were observed to have attached to the biofloat plates or have cellular protrusions (Figure 3B and Supplementary Figure S2B).

Assessment of inter-plate consistency across three independent experiments using U-bottom plates showed no significant differences in μbMPS size between plates (p = 0.83; Supplementary Figure S2C). However, using 10,000 cells per well produced exceptionally large organoids (more than 1 mm in diameter at 2 weeks). To address this, the total seeding number of NPCs and PMs was reduced to 2,000 cells per well, which decreased organoid size (approximately 0.7 mm diameter at 2 weeks) and improved μbMPS yield per batch of premature microglia. The protocol was also refined by reducing the aggregation time in a 96-well plate from 4 to 3 days. This adjustment eliminated the need for a medium change after 48 h and limited organoid size, as NPCs proliferate rapidly in Neural Expansion Medium.

Finally, a 7:3 ratio of NPCs to PMs was initially selected based on Xu et al. (2021). However, due to the rapid proliferation of NPCs, we found that a final ratio of 50:50 was more appropriate for an optimal microglia-to-neuron ratio in the mature μbMPS. With these refinements of the protocol, 100% of μbMPS^96-6 had IBA1⁺ microglia visible when analyzed at 2 weeks post-integration.

Overall, due to the improved uniformity and consistency of μbMPS^96-6 in U-bottom plates—along with minimal attachment or process formation—this protocol was used for all subsequent experiments, with 2,000 cells per well at a 50:50 PM:NPC ratio, 3 days of aggregation in 96-well plates and subsequent culture in a 6-well plate for up to 9 weeks.

We generated bMPS with and without PMs in U-bottom 96-well plates (μbMPS^96-6 and bMPS^96-6, respectively), alongside the standard protocol (bMPS^6-6), to evaluate whether the inclusion of premature microglia impacted bMPS growth and size and to compare it with our standard protocol over time. We quantified the size of these three cultures over 6 weeks of differentiation. bMPS^96-6 were significantly larger than bMPS^6-6 as early as 2 weeks after integration and continued to grow in size throughout the culture period, with no significant differences due to the presence of microglia (Figure 3C).

To confirm that the differentiation produces cells resembling a microglia lineage, we performed bulk RNA sequencing at 2 weeks post-integration. We identified a panel of microglia-specific or microglia-related genes from the literature (Miller-Rhodes, 2022) and evaluated their expression in these samples. Although the relative expression of some markers was low, we observed an enrichment of many microglia-specific genes, such as SAL1 or HEXB (Figure 3D). Although single-cell sequencing would more effectively confirm that we have a homogenous population of microglia, these data are suggestive of microglia, rather than macrophage, fate.

Finally, a representative image of a microglia in an 8-week-old μbMPS demonstrates the highly ramified morphologies typically observed, consistent with a homeostatic state (Figure 3E).

To validate integration, maturation, and survival during bMPS differentiation, we harvested μbMPS^96-6 at 3, 6, and 9 weeks post-integration and immunostained with microglia markers (CD68, IBA1, PU.1, TREM2, and P2RY12) (Figures 3E, 4). We demonstrated that microglia are evenly distributed in the μbMPS^96-6 at 2 and 6 weeks (Figure 4E, Supplementary Figure S3B and Supplementary video 1); we observed persistent survival of microglia in the μbMPS^96-6 for at least 9 weeks, and microglia can adopt ramified morphologies consistent with a non-diseased state and healthy surveillance of the environment (Figures 3E, 4A,B,D). Microglia have also been identified with ameboid (Figure 4C) morphologies, suggestive of an inflammatory response.

We then assessed the neural cellular composition in μbMPS^96-6, bMPS^96-6, and bMPS^6-6 to evaluate whether the culture technique and/or the presence of microglia influence the differentiation efficiency of the main neural lineages. During 8 weeks of differentiation, there were no statistically significant differences in the expression of neuronal (MAP2) and oligodendrocyte (MBP) markers across the conditions, as measured by RT-qPCR. Notably, we observed a higher expression of GFAP in μbMPS^96-6, although it did not reach statistical significance (Figure 5A).

Additionally, we quantified the expression of the synaptic marker (synaptophysin, SYP) in 8-week μbMPS^96-6, bMPS^96-6 vs. bMPS^6-6, and detected no difference between the conditions (Figure 5A). We then analyzed synaptic density in 9-week-old μbMPS^96-6 and bMPS^96-6 by co-staining and colocalization of pre- and post-synaptic markers (Synapsin1 and Homer1). Consistent with qRT-PCR, we identified no statistically significant differences in synaptic density between bMPS^96-6 and μbMPS^96-6 (Figures 5B,C).

In addition, we stained μbMPS^96-6 with neuronal (NF200, β-III-Tubulin), astrocyte (GFAP), and oligodendrocyte (O4) markers at 6 and 8 weeks of differentiation to demonstrate the presence of microglia and all main neural cell types in the mature μbMPS^96-6. The presence of the expected morphologies of astrocytes and oligodendrocytes appeared as early as 6 weeks (Figures 6A–D).

To quantify the proportion of microglia in the μbMPS^96-6, we labeled the microglia with a lipophilic dye before integrating them into the organoids. We dissociated 6-week-old μbMPS^96-6 and counted the total number of nuclei, which averaged 1.9×10⁵ cells per organoid. The total number of microglia per organoid was on average 2.5% (1.34–3.5%) (Supplementary Figure S3A). In a previously published analysis of our bMPS^6-6 model, we quantified the proportions of other cell types (Romero et al., 2023). Therefore, we expect the relative cell proportions to be approximately 35% astrocytes, 20% oligodendrocytes, 50% neurons, and 2.5% microglia at a 6-week timepoint, but single cell RNA sequencing is needed to estimate the percentage of different lineages with higher precision.

Ultimately, recognizing the importance of microglia in neuronal network maturation and electrical activity, we characterized the spontaneous electrical activity of microglia-containing bMPS vs. those without microglia. Historically, bMPS^6-6 exhibit no or very low spontaneous network activity detectable with calcium imaging at 3 weeks of maturation (Alam El Din et al., 2025). bMPS^96-6 similarly does not exhibit spontaneous calcium potentials at week 3; however, μbMPS^96-6 has clear network bursting patterns (Figure 7) with representative frames (Figure 7A) and traces (Figure 7B) shown. The bursts were characterized by average rise times, decay times, peak amplitudes, burst durations, firing rates, and the number of peaks observed in the recording period (Figure 7C). This suggests that microglia might contribute to the maturation of neural networks in vitro, as they do in vivo (Arnoux and Audinat, 2015).

Then, we placed the bMPS^96-6 and μbMPS^96-6 on a High-Density Micro-Electrode Array (HD-MEA). We assessed active area metrics (including the percentage of active area, firing rate, amplitude, and inter-spike intervals), network metrics (characterizing spiking and bursting activity), and axon tracking in both conditions. Consistent with the calcium flux data, μbMPS^96-6 showed an overall trend toward increased spontaneous electrical activity relative to bMPS^96-6 (Figure 8). Active area metrics were significantly higher in μbMPS^96-6 (Figures 8A,B), as were the firing rate (Figure 8B), the percentage of spikes in bursts, burst duration, and burst interspike interval (Figures 8C,D). These changes are also evident in graphical formats, raster plots, and quantifications.

Additionally, at the earlier time point of differentiation, the active axons were longer, as represented by the length from the initiation site, which was significantly increased (Figure 8F). Overall, this data demonstrated that bMPS containing microglia were more active (based on both active area and burst metrics), which is consistent with the calcium flux data. Future studies will help to elucidate the mechanisms through which microglia contribute to earlier network activity and maturity.

To confirm that microglia in μbMPS^96-6 can respond to their environment, we challenged them with a Traumatic Brain Injury (TBI) model. The TBI protocol was modified from a previously published protocol (Ortega-Villa et al., 2021), where it is characterized as a sustained compression injury of moderate severity. Here, the absence of surrounding hydrogel results in a larger force; therefore, we consider this a severe compression injury. A total of 32 μbMPS^96-6 were randomly selected from the same pool; 16 were challenged with TBI via centrifugation, while the other 16 were not centrifuged. Samples were collected before any manipulation and after 4 and 24 h post-centrifugation. Although no significant changes in microglia morphology were observed, TBI-treated μbMPS^96-6 exhibited a statistically significant accumulation of microglia on the surface of the organoid z-stack, indicating that microglia traversed to the outer layers of the bMPS to respond to surface neuronal injury (Figures 9A,B).

Next, to verify whether microglia in μbMPS^96-6 are capable of phagocytosis, we exposed the cultures to 1 mg/mL pHrodo Red E. coli particles for 3 h. Z-stacks were obtained of the whole cell (including projections) to confirm that the bioparticles were internalized within the 3D structure of the cell body, rather than in front of or behind the cell. Bioparticles within the microglia appear smaller than surrounding particles, likely representing partial digestion, and colocalize with CD68 labeling of the lysosome or endosome (Figure 9C). While we would expect E. coli bioparticles to be phagocytosed rather than internalized during endocytosis, CD68 also labels endosomes. To ensure that pHrodo Red bioparticles are truly internalized into lysosomal compartments, it may be beneficial to label cells with Lysotracker or lysosome-specific proteins such as LAMP1/2.

Finally, we measured secreted cytokines and chemokines (Supplementary Table S3) in 8-week-old bMPS^96-6 and μbMPS^96-6 following exposure to lipopolysaccharide, a component of the cell wall of E. coli bacteria (Figure 10A), and TNFα cytokine (Figure 10B). We identified a statistically significant increase in two microglia-specific cytokines, MIP-1α and MIP-1β, only in μbMPS^96-6; the cytokine flux was absent in non-microglia-containing bMPS (bMPS^96-6). Low concentrations measured can reflect the low number of microglia (2.5%) per bMPS and low number of bMPS in a large volume of media (2 mL) to allow for standard maintenance on gyratory shaking in a 6-well plate during exposure. In addition, presence of trace endotoxin in bMPS differentiation medium can attenuate the response to LPS challenges. Subsequent studies will control and eliminate endotoxin from differentiation media to mitigate this effect.