A 3D Human Tri-Culture System Modeling Neurodegeneration and Neuroinflammation in Alzheimer’s Disease

Generation of human AD tri-culture model (Neu+AC+MG AD) composed of neuron, astrocyte and microglia using a microfluidic platform.

The AD brain platform used here consists of a microfluidic device containing two chambers mimicking the in vivo AD environment to observe interactions between neurons/astrocytes and microglia. The central chamber was loaded with the 3D AD neuron/astrocyte differentiated cells while the angular chamber was loaded with adult microglia (Fig. 1a–c). Human neural progenitor cells (hNPCs) that produce high levels of Aβ were developed through overexpression of a variant of the human beta-amyloid precursor protein (APP) containing FAD mutations with both K670N/M671L (Swedish) and V717I (London) FAD mutations (APPSL) as previously published (Supplementary Fig. 1)¹⁵. These transduced cells with control-GFP (neuron+astrocyte or Neu+AC) or APPSL-GFP (AD neuron+astrocyte or Neu+AC AD) lentiviral particles were enriched on the basis of GFP signals through fluorescence-activated cell sorting (FACS). Representative fluorescence microphotographs of differentiated ReN cells showed stable expression for more than 3 weeks (control-GFP: Neu+AC and APPSL-GFP: Neu+AC AD, Supplementary Fig. 1). The central and angular chambers were linked by migration channels (10 × 50 × 500 μm in height x width x length) forming gradients of soluble factors. The long and thin migration channels prevented inadvertent entrance of inactivated microglia through mechanical constraints. Using this platform, we observed the migration of microglial cells, at single cell resolution, in real-time, for extended periods (Fig. 1d). Human neural progenitor cells, differentiated into neuron and astrocyte (green) were successfully tri-cultured with adult microglia (red) (Figure 1d–g). These observations demonstrated that our 3D microfluidic device enabled the formation of physiologically relevant human brain tissue-mimetic 3D structures (Supplementary Fig.2).

To demonstrate hNPCs differentiation in 3D cultures (Supplementary Video 1&2), fixed cells were analyzed by immunofluorescence for the presence of neuronal/astrocyte markers. As previously reported, human hNPCs robustly expressed class III beta-tubulin (Tuj1), microtubule-associated protein 2 (MAP2), glial fibrillary acidic protein (GFAP) and calcium-binding protein (S100, S100A6 and S100β) in week 3 (Fig. 1h, Supplementary Fig.3–4)¹⁵. Activated microglia, as demonstrated by CD68 expression, were also observed in tri-culture conditions in the 3D Neu+AC+MG AD culture model (Fig. 1h). Neuronal (Tuj1, MAP2) and glial (GFAP) markers in undifferentiated and 3-week differentiated AD ReN cells were quantified by Western blot analysis (Fig. 1i–k). Western blot analysis for neuronal markers Tuj1 and MAP2, and the astrocytic marker GFAP demonstrated that neuronal and astrocyte differentiation of NPCs was significantly promoted by 2.1-fold, 2.9-fold and 2.3-fold, respectively, in 3D culture in microfluidic devices compared to 2D cultures. These differences may reflect the ability of 3D environments to promote differentiation and/or maturation of neural progenitor cells providing a high surface area for growth^25,26.

We next characterized action potential-evoked calcium dynamics in individual neurons (Fig. 1l) filled with Cal-590²⁷. The recorded individual neurons were immunoassayed with the neuronal marker Tuj1 to confirm that the calcium dynamics was from neurons. The high-magnification microphotographs of the soma of stained neurons displayed characteristic ring-shape appearance arising from a lower fluorescence level in the nucleus compared with the surrounding cytoplasm (Supplementary Fig. 5). We analyzed the spontaneous activity of these neurons by combining calcium activation, neuronal marker, fast-time lapse imaging (50 frames/time) and found that single-action potential-evoked calcium transients were discernable with a good signal-to-noise ratio (Fig. 1m & Supplementary Video. 3)²⁷. Interestingly, 3D Neu+AC AD showed a significant increase of neuronal activity levels could be observed when compared with week-matched control (3D Neu+AC) in the 3-week-old cells. Furthermore, the fraction of neurons that were hyperactive (>6 transients per 60 seconds) was markedly larger in 9-week-old 3D Neu+AC AD than in 3-week-old 3D Neu+AC AD, while the fraction of silent neurons (0 transients per 30 seconds) was not different between 3D Neu+AC AD at week 3 and week 9 (Supplementary Fig. 6 & Video. 4–5). Taken together, these data clearly demonstrate the generation of functional neuron/astrocyte cell models in our 3D microfluidic culture system.

Increased soluble Aβ, inflammatory cytokines/chemokines and phosphorylated tau (p-tau) in the 3D Neu+AC AD model

The levels of soluble 40-amino acid and 42-amino acid Aβ isoforms (Aβ40 and Aβ42) were determined in conditioned media collected after 3, 6, and 9-weeks of differentiation in either 2D or 3D cultures of control neuron/astrocyte (Neu+AC) or AD neuron/astrocyte (Neu+AC AD) cells. Aβ40 (10.2-fold and 1.7-fold) and Aβ42 (8.2-fold and 1.1-fold) levels were compared to the control 3D Neu+AC and 2D Neu+AC AD (Fig. 2a and b) at week 3. Furthermore, at week 9, 3D Neu+AC AD was associated with significantly higher accumulations of Aβ40; 78-fold and 2.8-fold compared to levels observed in 3D Neu+AC and 2D Neu+AC AD, respectively. Interestingly, the soluble Aβ42_week9/ Aβ42_week0 ratio in 3D Neu+AC AD was significantly decreased (≈ 20%), indicating that the 3D culture conditions may increase the aggregation of Aβ42 (Fig. 2a and b, Supplementary Fig.7&8). This observation was further validated through Western blot analysis of the various forms of aggregated Aβ in whole-lysates (Fig. 2c and 2d) with increased aggregated forms of Aβ in 3D Neu+AC AD²⁸. These observations are consistent with an accelerated Aβ accumulation due to limited diffusion in 3D culture conditions and diffusion of AD neuron-derived Aβ into the media in 2D cultures as previously reported^15,16.

To determine whether our 3D Neu+AC AD secretes CCL2 (MCP-1), a chemokine up-regulated in AD brains, we compared the level of CCL2 in various stages of the AD culture models²⁹. We found that 3D Neu+AC AD (1800 pg/ml) showed 2.9-fold and 1.3-fold more CCL2 secretion in conditioned media compared to 3D Neu+AC (620 pg/ml) and 2D Neu+AC AD (1384 pg/ml), respectively (Fig.2e). To further explore pathology development in 3D cultured AD neurons/astrocyte, chemokine and cytokine secretion by 3D Neu+AC, 2D Neu+AC AD, and 3D Neu+AC AD cells was determined. First, we evaluated TNF-α levels in the culture medium and found TNF-α levels increased by 2.8-fold and by 8.2-fold in week 6 (61 pg/ml) and week 9 (180 pg/ml) in 3D Neu+AC AD, respectively (Fig.2f), while moderate increases were observed in 2D Neu+AC AD. Surprisingly, IFN-γ was exclusively secreted by 3D Neu+AC AD cells (9.2 pg/ml) but not detected in the other conditions tested (Fig.2g).

Regarding tau alterations³⁰, ELISA analysis of soluble phosphorylated tau (pSer231 tau; phosphorylated at Serine 231) in 3D Neu+AC AD cells showed that pSer231 tau levels increased by 2.2-fold and 5.1-fold at weeks 6 (271 pg/ml) and 9 (630 pg/ml), respectively, compared to 3D Neu+AC (Fig.2h). The presence of p-tau phosphorylation at Ser396/Ser404 and Ser202/Thr206 was also confirmed by immunofluorescent staining using p-tau antibody PHF-1 and AT8, respectively, in both neuritic and soma bodies of 3D Neu+AC AD cells (Fig.2i and Supplementary Fig.9).

3D Neu+AC AD model induced microglial migration, phenotype changes and secretion of pro-inflammatory factors

To test the hypothesis that neurons/astrocytes in AD environments induce activation/recruitment of microglial cells, we plated microglial cells in the angular chamber at Day 0 of week 3, week 6, and week 9. To characterize the migration and phenotype changes of microglia, individual microglial cells were monitored for seven days using time-lapse imaging microscopy (Fig. 3a). In the first 24 hours, microglia in the angular chamber showed widely branched filopodia in all directions, a morphology typically associated with ‘resting’ microglia. Furthermore, most of the cells remained stationary during that period. Morphological changes and microglial migration towards the central chamber seeded with 3D Neu+AC AD was observed following cell seeding after 48 hrs (Fig. 3b). As determined by image analysis, microglial cells became elongated with significantly increased cell length and microglial cell body area (Fig. 3b and c). Furthermore, exposure of microglia to 3D Neu+AC AD induced up-regulation of microglial activation markers such as CD11b (Fig. 3d). The migration rate of microglial cells in the 3D Neu+AC+MG AD environment was significantly higher compared with those in the other conditions tested: 3D Neu+AC+MG or 2D Neu+AC+MG AD. The microglial recruitment index (recruitment index, R.I., Supplementary Fig.10) was 8.1 ± 0.78% in 3D Neu+AC+MG AD cells compared with 4.2 ± 2.3% in 2D Neu+AC+MG AD cells and 0.9 ± 1.2% in 3D Neu+AC+MG cells (Fig. 3e). Interestingly, neutralization of CCL2 chemokine signaling using an anti-CCL2 antibody reduced the ability of microglial to migrate in response to Neu+AC AD co-culture conditions (Supplementary Fig. 11). Other factors, such as ATP³¹ (Supplementary Fig. 12) may also contribute to microglial recruitment in our system. Notably, in the 3D Neu+AC+MG AD model, the aged cultures (week 9) led to the recruitment of more microglia cells as compared to early stage cultures. 3D Neu+AC+MG AD and 2D Neu+AC+MG AD culture platforms had comparable effects on maximal microglial migration speed (5.0 ± 0.91 vs. 4. 3 ± 0.81, data not shown). The observed speed of migration was similar with that previously observed in AD environments with soluble and bound Aβ conditions¹⁵

Next, we set out to identify the cytokines and chemokines that regulate the microglial migration/activation in our 3D AD models. For this purpose, we first measured the secreted cytokine/chemokine levels in 3D Neu+AC and 3D Neu+AC AD models without microglial cells. Conditioned cell culture medium were collected from the week 9 models and levels of 36 cytokines were analyzed. Interestingly, protein levels of CCL2, CXCL10 and CX3CL1 were significantly increased in 3D Neu+AC AD compared to the control 3D Neu+AC model (Fig. 3f). In contrast, no changes in CCL3/4, CCL5 and CXCL1 expression were found in either culture model. Next, we assessed how inflammatory signaling was impacted by adding microglial cells to the 3D Neu+AC AD model. AD-related neuroinflammatory responses, such as chemokine release (Fig. 3f) and pro-inflammatory factor secretion (Fig. 3g, Supplementary Fig.13), were generally increased by the addition of microglial cells. Indeed, IL-6, IL-8 and TNF-α concentrations were elevated by 2.2-, 2.7- and 1.3-fold, respectively, compared to 3D Neu+AC AD alone. In addition, activated microglia induced the release of multiple chemokines including CCL2 (2.1-fold), CCL5 (26-fold), CXCL10 (2.6-fold), CXCL12 (1.2-fold) in 3D Neu+AC+MG AD compared to the 3D Neu+AC AD. We also observed the unique production of the leukocyte growth factors GM-CSF and G-CSF, which typically are not produced by neurons. MIF and PAI-1, important biomarkers of tau hyper-phosphorylation in late-stage AD mouse model, were greatly increased in 3D Neu+AC+MG AD (Fig 3g)³¹. Secretion of inflammtory cytokines such as IL-1β and IL-17 and anti-inflammatory markers such as IL-1RA, IL-10, and TGF-ß was very limited or below detectable levels. These data suggest that microglia elicit broad inflammatory signaling in response to our 3D Neu+AC AD model.

Recruited microglia are toxic to AD neurons and astrocytes

Next, we assessed the extent of cellular damage mediated by microglia in our 3D Neu+AC+MG AD culture model. On day 2, after adding microglial cells to 9-week-differentiated 3D Neu+AC AD system, we observed morphological changes of microglial cells in co-localized with 3D Neu+AC AD cells expressing GFP signals (Fig. 4a, Supplementary Video. 6). After day 4, we observed slight decreases in neuron/astrocyte surface area, including cell body and axon, when co-localized with microglia (Fig. 4a). Axonal damages were also observed with co-localized microglial cells (Fig. 4b, Supplementary Fig. 14 & Video.7). Time-lapse imaging also indicated that AD neurons undergo changes in cell morphology, including retraction of the neurites toward the main cell bodies (Fig.4c, Supplementary Video.8). To quantitatively analyzed the cellular damages, we next performed image analysis by converting nucleus stained images into a dimension matrix to determine nucleus density in tri-culture conditions with microglia via heat map contour (Fig.4d, Supplementary Fig. 15). At day 2, cells exhibited relatively moderate density across ROI regions. However, at day 6, a marked decrease of cellular density was observed around nearby microglia cells (Fig. 4e). Indeed, after day 6 of tri-culture with microglial cells, there was a 21% reduction in neuron+astrocyte survival counted by nucleus (Fig. 4e) and a marked reduction of about 37% in neuron+astrocyte surface area (Fig. 4f) when tri-cultured with microglia compared to 9-week-old 3D Neu+AC AD only (Supplementary Fig. 15). To elucidate and quantify how microglia influence the induction of neuronal and/or astrocyte loss, we analyzed 9-week-old 3D Neu+AC+MG AD at day 2 and day 6 post- microglial cell recruitment using immunostaining and immunoblotting to compare and quantify the loss of each cell type (Fig. 4g and h). TUJ1/MAP2 (neurons) and ALDH1L1 (astrocyte) markers were used for the quantification of the neuron and astrocyte loss, respectively. Our data indicated broad losses of neurons (Tuj1: 35.5%, MAP2: 27.9%) and astrocytes (15.7%) across all 3D Neu+AC+MG AD model systems tested, whereas the number of microglia (CD68-positive) remained the same (Fig. 5e). Our results, therefore, suggest that the addition of microglial cells promotes distinct changes in the number and morphology of neurons/astrocytes, which is consistent with the neuropathological findings previously demonstrated in the 5xfAD mouse model³².

Recruited microglia induced neuronal loss through a IFN-γ and TLR4-dependent mechanism

To test whether cytokines and chemokines contribute to neuron and astrocyte loss, we evaluated the contribution of IFN-γ and TLR4 on the production of neurotoxic mediators in our tri-culture system (Fig. 5a). TNF-α and NO concentrations were measured in the presence of either neutralizing or blocking antibodies to IFN-γ or TLR4, respectively³³. Recruited microglia in the 3D Neu+AC+MG AD model exhibited a 2.3-fold increase in TNF-α concentrations compared to the 3D Neu+AC AD model without microglia (Fig. 5b). NO levels were also increased by 9.1-fold in the 3D Neu+AC+MG AD model, while not detectable in the 2D Neu+AC+MG AD model possibly due to lack of IFN-γ as in this system (Fig. 2g and 5c). Addition of either TLR4 or IFN-γ neutralizing antibodies reduced TNF-α and NO concentrations demonstrating that the neuron/astrocyte loss in 3D Neu+AC+MG AD is dependent on microglial activation. We also measured lactate dehydrogenase (LDH) release in conditioned media as a biochemical marker of cell death (Fig. 5d). Notably, incubation with the TLR4 antagonist (LPS-RS) significantly decreased LDH levels in the 3D Neu+AC+MG AD culture model (Fig. 5d). These data suggest that the microglia-induced neuronal loss in the 3D Neu+AC+MG AD model, occurs at least partially via IFN-γ− and TLR4 -dependent mechanisms, in agreement with the other models³³.

We also observed increased GFAP signals and altered astrocyte morphology in the 3D Neu+AC AD and the Neu+AC+MG AD models, suggestive of astrogliosis (Fig. 5e). Both 3D Neu+AC AD and Neu+AC+MG AD induced astrogliosis in the week 9 model, as indicated by increased GFAP expression. Furthermore, astrocyte activation was concomitant with activation of microglia as indicated by increases in CD68 immunoreactivity and thickening of their processes indicative of a shift in state toward a more activated amoeboid morphology (Fig. 5e). Additionally, we observed that some of the neuronal loss was associated with the activation of apoptotic pathways as an observation of fragmented nuclei after 8h in 3D Neu+AC+MG AD (Fig. 5f).

Knocking down TLR4 expression provided protection against neurotoxic microglial activation

Since we observed a reduction in TNF-α and NO concentrations following the addition of TLR4 blocking antibodies, we hypothesized that TLR4 activity is required for microglial activation. Thus, we investigated whether selective knock-down of TLR4 receptor expression can similarly reduce the inflammatory status of microglia in our AD models. To test this, we used lentiviral particles to knock-down TLR4 in WT microglial cells (TLR4 KD) and plated these cells in the angular chamber. TLR4 KD microglial cells exhibited an 79% decrease in TLR4 compared to WT microglial cells (Fig. 5g). Knocking down TLR4 in microglial cells protected neuron and astrocyte loss (2.3-fold) compared to WT microglial cells in the 3D Neu+AC+MG AD systems.

Human iPSC-derived AD neurons/astrocytes replicated microglial recruitment, activation and neuron/astrocyte death in tri-culture model system

To confirm that the proposed mechanism can be replicated in other cultured human neuron/astrocyte, we newly generated human induced pluripotent stem cell (iPSC)-derived neural stem cells (NSC) expressing high levels of pathogenic Aβ species using the similar techniques used for the ReNcell VM-based 3D AD models (Fig. 6a). Human NSCs, stably expressing APPSL/GFP (AD-iPSC) or GFP alone (control iPSC), were plated and differentiated for 3 weeks under 3D culture conditions. In these culture conditions, we also confirmed that cells were differentiated into neurons and astrocytes, respectively, as determined by the Tuj1, MAP2 and GFAP immunofluorescence staining (Fig. 6b and c). To further examine the development of synapses-like structures, dendritic protrusions at week 0 and week 3 were compared (Fig. 6d and e). Especially, we imaged these protrusions making connections in between synapses-like structures in the volume (Fig. 6f). As expected, the 3D Neu+AC+MG AD-iPSC culture produced high levels of Aβ40 (1.75ng/ml) and Aβ42 (0.45ng/ml) (Fig. 6g). We also analyzed the cytokine secretion profile at week 3. Incubation of 3D Neu+AC AD-iPSC for 3 weeks led to increased cytokine/chemokine secretion. Indeed, the release of inflammation-related CCL2, CCL5, IL6, IL8 and MIF were upregulated following the addition of microglial cells (Fig. 6h). Interestingly, IL-1β, IL-6, IL-10, and TNF-α were not detected in the 3D Neu+AC AD-iPSC cultures without microglia (Fig. 6h). Importantly, microglial migration/recruitment was largely elevated in the 3D Neu+AC+MG AD-iPSC as compared to 3D Neu+AC+MG control-iPSC (Fig. 6i–j). We next treated 3D Neu+AC+MG AD-iPSC with IFN-γ and compared the neuron/astrocyte loss between AD and control cells3D Neu+AC+MG AD-iPSC showed significantly higher levels of neuron/astrocyte loss while the IFN-γ neutralizing antibody treatments protected neuron/astrocyte loss (Fig. 6k–l). These data further validated our findings on glial activation and the role of cytokines in 3D tri-culture models with iPSC-derived human neuron/astrocyte cells.

In conclusion, we were able to recapitulate dynamic neural-glial interactions in a novel 3D human tri-culture (Neu+AC+MG) model system. This model was generated by differentiating genetically modified human neural progenitor cells into neurons and astrocytes, expressing AD pathological hallmarks: Aβ aggregation, p-tau formation, and chemokine/cytokines such as CCL2, TNF-α and IFN-γ. We achieved this by co-culturing on a microfluidic platform, and by tri-culturing human adult microglia in separated angular chambers to observe microglia recruitments with AD neuron and astrocytes. The relevance of this tri-culture model lies in the demonstration of physiologically relevant neural-glial interactions: microglial morphogenesis, recruitment, and the release of proinflammatory cytokines and leukocyte-chemokines regulated by AD neurons and astrocytes, and microglial neurotoxic activation contributing to neuron/astrocyte damage. These findings carry important implications for studies aimed at a deeper understanding of the gliosis and pathogenesis of AD and screens aimed at finding new therapeutic molecules for preventing and treating AD.

Media and reagents of neural progenitor cells (NPCs)

ReN cell VM human neural progenitor cells (NPCs) were purchased from EMD Millipore (Billerica, MA, USA). The cells were plated onto BD Matrigel^® (BD Biosciences, San Jose, CA, USA) coated T25 cell culture flasks (BD Biosciences, San Jose, CA, USA) and maintained in DMEM/F12 (Life Technologies, Grand Island, NY, USA) media supplemented with 2 mg heparin (StemCell Technologies, Vancouver, Canada), 2% (v/v) B27 neural supplement (Life Technologies, Grand Island, NY, USA), 20 mg EGF (Sigma-Aldrich, St Louis, MO, USA), 20 mg bFGF (Stemgent, Cambridge, MA, USA) and 1% (v/v) penicillin/streptomycin/amphotericin-B solution (Lonza, Hopkinton, MA, USA) in a CO₂ cell culture incubator. Cell culture media were changed every 3 days until cells were confluent. For 2D neuron/astrocyte differentiation, the cells were plated onto Matrigel^®-coated microfluidic devices with DMEM/F12 differentiation media supplemented with 2 mg heparin, 2% (v/v) B27 neural supplement, and 1% (v/v) penicillin/streptomycin/amphotericin-B solution without growth factors. One half volume of the differentiation media was changed every 3–4 days for 3–9 weeks. LPS-RS was purchased from InvivoGen (San Diego, CA). LPS-RS was resuspended in dimethylsulphoxide (DMSO; 0.1 mg in 10 ml) followed by 1:10 dilution in differentiation culture medium. This stock solution of 22.2 mM was then diluted for cell treatment experiments.

Viral infection of NPCs

To transduce the NPCs with the lentiviral constructs, 50–100 μl viral solution (1 × 10⁶ TU ml⁻¹, MOI=4) were added to 85% confluent proliferating ReN cells in 6-well dishes, incubated for 24 h, and washed three times to stop the infection. The expression of the infected genes was confirmed by GFP expression by fluorescence microscopy and western blot analysis.

3D cell cultures and differentiation of NPCs

For 3D cultures, BD Matrigel^® (BD Biosciences) was mixed with the cells (10 × 10⁶ cells per ml). The final cell concentration for the mixture was approximately 5 × 10⁶ cells per ml (1:1 3D thick culture for confocal and westernblot) and 2 × 10⁶ cells per ml (1:5 3D thin culture for immunostaining and recruitment). Ten microliter of cell mixtures were transferred into the microfluidic device using pre-chilled pipettes. The microfluidic devices were incubated for 1h at 37°C during which 3D gels (100–600 μm) formed and media were changed. The 3D-plated cells were differentiated for 3–9 weeks depending on the experiments; media was changed every 3–4 days. For drug treatments, differentiation media containing either 1mM LPS-RS diluted in DMSO or the concentration of DMSO were added to 3–9-week differentiated 3D AD neuron-microglia cells and the 3D cultures maintained for an additional 1 week.

Expansion, viral infection and differentiation of hiPSC-NPCs

Human iPSC derived neural progenitor cells (hiPSC-NPCs) were purchased from EMD Millipore (SCR131). The cells were plated onto BD Matrigel® (BD Biosciences, San Jose, CA, USA) coated 6 well plate (BD Biosciences, San Jose, CA, USA) and expanded in expansion medium (EMD Millipore, SCM004) supplemented with FGF-2 (20ng/ml) and L-Glutamine (2 mM) in a CO2 cell culture incubator. To transduce the hiPSCs with the lentiviral constructs, 50–100 μl viral solution (1 × 10⁶ TU ml⁻¹, MOI=4) were added to 85% confluent proliferating hiPSCs in 6-well dishes, incubated for 48 h, and washed three times to stop the infection. The expression of the infected genes was confirmed by GFP expression by fluorescence microscopy and followed by FACS sorting. To differentiate, hiPSC-NPCs were prepared approximately 10 × 10⁶ cells per ml of Matrigel^®. Ten microliter of cell mixtures were transferred into the microfluidic device using pre-chilled pipettes. The microfluidic devices were incubated for 1h at 37°C during which 3D gels (100–600 μm) formed and media were changed in differentiation media (EMD Millipore, SCM017). The 3D-plated cells were differentiated for 3 weeks; media was changed every 3–4 days.

FACS enrichment of the transduced NPCs and iPSC-NPCs

The infected ReN cells were washed with DPBS and then incubated with Accutase (Millipore) for 5 min. The cell pellets were resuspended in PBS supplemented with 2% serum replacement solution (Life Technologies) and 2% B27, and then passed through a cell strainer filter (70 mm Nylon, BD Biosciences). The cell concentrations were adjusted to 2 × 10⁶ cells per ml and then enriched by using FACSAria cell sorter (Flow Cytometry Core Facility, Charlestown Navy Yard, MGH)). GFP channels were used to detect the expression of the transduced genes in the individual cells. The sorted/enriched cells were maintained in normal proliferation media.

Microfluidic Device Fabrication

Negative photoresists, SU-8 50 and SU-8 100 (MicroChem, Newton, MA, USA), were sequentially patterned using standard lithography on a 4” silicon wafer to create a mold for cell migration channels of 50 μm in height and chemokine compartments of 100 μm in height. A mixture of a base and a curing agent with a 10:1 weight ratio (SYLGARD 184 A/B, Dow corning, Midland, MI, USA) was poured onto the SU-8 mold and cured for one hour at room temperature under vacuum and, subsequently, cured for more than 3 hours in an oven at 80 °C. The cured poly dimethyl-siloxane (PDMS) replica was peeled off from the mold and holes were punched for fluid reservoirs. Arrayed holes were also laser-cut (Zing 24, Epilog Laser, Golden, CO, USA) into a thin PDMS membrane of 250 μm in thickness (HT 6240, Bisco Silicones, Elk Grove, IL, USA) and an acrylic plate of 6 mm in thickness. The machined membrane and the plate were glued together using uncured PDMS and incubated at 80 °C overnight. This assembly was irreversibly bonded first to the PDMS replica using oxygen plasma at 50 mW, 5 cm, for 30 seconds (PX-250, March Plasma Systems, Petersburg, FL, USA), and later to a glass-bottomed Uni-Well plate (MGB001-1-2-LG, Matrical Bioscience, Spokane, WA, USA). Immediately after the bonding, 10 μL of poly (l-lysine) solution (PLL, M.W. 70,000–150,000, 1.0 mg/mL, Sigma-Aldrich Co. LLC, St. Louis, MO, USA) was injected into each platform and incubated for 2 hours at a room temperature to promote cellular adhesion. PLL-treated surface was rinsed with autoclaved and 0.2 μm filtered water (AM9920, Life Technologies, Grand Island, NY, USA).

Calcium Imaging

To assess neural activity, cellular calcium dynamic was monitored using Rhod-2 (Life Technologies), a Ca2+ indicator. Neuron-cultured samples were incubated for 30 min at 37°C in artificial buffered cerebrospinal fluid (i.e., 25 mM NaHCO₃, 25 mM D-glucose, 125 mM NaCl, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 1 mM MgCl₂•6H₂O, and 2 mM CaCl₂•2H₂O) supplemented with 2 M of OGB-1 diluted in pluronic F-127 20% solution/DMSO (Sigma). After incubation, the samples were washed twice with fresh artificial buffered cerebrospinal fluid and incubated for 30 min at 37°C. Fluorescence intensity dynamics was measured using time-lapse imaging (Texas red filter cube; 20X objective; Nikon microscope). at a 30ms frame rates. Raw intensity values were extracted using the Nikon software suite, and relative changes in fluorescence intensity to baseline ((F-F0)/F0) were regarded as neuronal calcium signals.

Microglia preparation

The immortalized human microglia-SV40 cell line derived from primary human microglia was purchased from Applied Biological Materials, Inc. (ABM, Inc.) and cultured in Prigrow III medium supplemented with 10% (vol/vol) FBS and 1% penicillin/streptomycin in type I collagen-coated T25-flasks (ABM, Inc.). Before the experiment, cells were washed using medium without serum and the cell membrane was labeled with red fluorescent dye (PKH26PCL, Sigma-Aldrich). Briefly, after centrifugation (400 g for 5 minutes), the cells were re-suspended in 1 mL of Diluent C (G8278, Sigma-Aldrich) and immediately mixed with 4 μL of dye solution (PKH26PCL, Sigma-Aldrich). The cell/dye mixture was incubated at room temperature for 4 minutes and periodically mixed by pipetting to achieve a bright, uniform, and reproducible labeling. After the incubation, the staining was stopped by adding an equal volume (1 mL) of 1% BSA in PBS and incubating for 1 minute to remove excess dye. Unbound dye was washed by centrifugation and suspending cells in culture medium (10⁶ cells/ml). Ten μL of the cell suspension was injected into each platform and 100 μL of a culturing medium was added into side and central extra wells. The loaded 3D micro-devices were then incubated at 37°C supplied with 5 % CO₂.

Time-lapse imaging

After microglia loading, cells were recorded using time-lapse imaging using a fully automated Nikon TiE microscope with a heated incubator to 37°C and 5 % CO₂ (10x magnification; Micro Device Instruments, Avon, MA, USA). To achieve accurate cell tracking, the maximum time resolution of acquisition was 1 frame per second.

ELISA

Aβ40 and Aβ42 levels were mainly measured by Human/Rat Aβ ELISA Kit from Wako (Osaka, Japan). The conditioned media from un- differentiated or differentiated ReN cells were collected and diluted by 1:10 or 1:100 with a dilution buffer provided by the manufacturer. A Synergy 2 ELISA plate reader (BioTek, Winooski, VT, USA) was used to quantify Aβ40 and Aβ42 ELISA signals.

IFN-γ (DIF50) secretion were quantified in supernatants by Human Quantikine ELISA Kits (R&D Systems, Oxon, UK) according to manufactures instructions.

Human cytokine array

The Human Cytokine Array (ARY005) kits were purchased and used to simultaneously detect relative expression levels of 36 human cytokines (R&D Systems, Oxon, UK). Conditioned medias from each sample, 0.5 ml respectively, were mixed with a cocktail of biotinylated detection antibodies. The mixture was then incubated with the Human Cytokine Array Panel A membrane. A wash step was needed to remove unbound material, and then Streptavidin-HRP and chemiluminescent detection reagents were added. Light was produced at each spot in proportion to the amount of analyte bound. The positive signals on developed films were identified by placing the transparency overlay on the ChemiDoc imaging system (Bio-Rad) and aligning it with the reference spots in three corners of each array. Pixel density was analyzed with ImageJ (Wayne Rasband, NIH).

Aβ preparation.

Soluble Aβ monomer was prepared by dissolving a synthesized clear human Aβ, 1–42 (4349-v, Peptide Institute, Inc., Japan) and a FITC-labeled human Aβ,1–42 (A-1119, rPeptide, Bogart, GA, USA) in DMSO at 1 mg/mL for a monomer form. Although Aβ pretreatment has generally been conducted in HFIP, because of recent reports of HFIP toxicity Aβ was directly dissolved in DMSO according to manufacturer’s recommendations (Peptide Institute, Inc., Japan). Indeed, the slow evaporation of HFIP may trigger the formation of Aβ oligomers, which could hasten fibril formation in combination with the humidity in air. The dissolved Aβ monomer solution was further diluted in PBS at 0.1 mg/mL and stored up to a week at 4°C or incubated in a shaker at 37°C to prepare oligomeric forms.

Immunostaining

For immunofluorescent stains, we rinsed the cells and 3D cultures twice with PBS (phosphate buffered saline). Cells were then fixed through a RT 15–30 min incubation in fresh 4% paraformaldehyde aqueous solution (157–4, ElectronMicroscopy Sciences) followed by rinsing twice with PBS. Cells were permeabilized through incubation in 0.1% Triton X-100 in PBST (phosphate buffered saline with 0.1% tween®20) for 15 minutes at RT. Cell on-specific binding was blocked through overnight incubation in 3% human serum albumin in PBST at 4 °C. After a 24-hours incubation with the primary antibody solutions at 4°C, the cells were washed five times. The following antibodies (and dilutions) were used: anti-p-tau antibody (1:40, AT8, Thermo Scientific, MN1020); anti-PHF (1:1000, A gift from P. Davies, Albert Einstein College of Medicine); anti-GFAP antibody (1:500, Neuromap, N206A/8), anti-MAP2 antibody (1:200, Cell Signaling Technology, 4542); anti-CD68 (1:100, Cell Signaling, 76437); anti-cd11b (1:100, Life Technologies, NB110–89474); anti-S100 (1:400, Abcam, ab868); anti-S100A6 (1:200, Cell Signaling, D3H3W); anti-S100B (1:500, Abcam, Ab218515); anti-ALDH1L1 (1:200, EMD Millipore, MABN495).

Western blot analysis

Protein lysates (15–75 μg) were resolved on 12% Bis-Tris or 4–12% gradient Bis/Tris gels (Life Technologies) and the proteins were transferred to nitrocellulose membranes (Bio-Rad). For amyloid-ß western blot analysis, the membranes were cross-linked with 0.5% glutaraldehyde solution before blocking. Western blot results were visualized using enhanced chemiluminescence (ECL). Signals were captured using ChemiDoc imaging system (Bio-Rad) and quantified using ImageJ software (NIH). The following primary antibodies were used (dilutions): anti- Aβ (1:100, 6e10; Signet); anti-Tuj1 (1:200, Abcam, ab24629); anti-MAP2 (1:200, Cell Signaling, 4542); anti-GFAP (1:100, Neuromap, N206A/8); anti-p-tau AT8 (1:30, Thermo Scientific, MN1020), p-tau PHF-1 (1:200, Abcam, ab66275), CD68 (1:200, BD Bioscience, 556059), CD11b (1:200, EMD Millipore, MM_NF-MABF515), anti-ALDH1L1 (1:200, EMD Millipore), Cy5 anti-mouse secondary antibody (1:400, Jackson Immunoresearch, 715-175-150).

Statistical analysis

All statistical analysis were performed by using Graphpad prism software. Either two-tailed Student’s t-test when comparing two groups/conditions or one-way ANOVA followed by post hoc test when comparing 3 or more groups/conditions were used. Data are expressed as means ± SEM, and a P-value ≤ 0.05 was considered significant. Special symbols (**, ***, #, †, #* and †*) to denote significant differences between two-group or three group analysis. Statistical analysis are detailed in figure legends. No statistical methods were used to predetermine sample sizes, and experiments were repeated to adequately reduce confidence intervals and avoid errors in statistical testing. Data collection and analysis were not performed blind to the conditions of the experiments. Randomization was not performed, and exclusions were not made.