In the brain, complex interactions between different cell types can drive inflammation and disease. This cross talk is hard to study in vitro, because culture systems are often too simple to recapitulate the intricacies of the brain, or the mix of cells is too poorly defined to give reliable results. In the February 8 Nature Neuroscience, researchers led by Lorenz Studer at the Sloan Kettering Institute for Cancer Research in New York City describe a triculture system of neurons, microglia, and astrocytes that combines pure populations of each cell type in a defined ratio. Generated from human pluripotent stem cells, each cell type can be genetically manipulated. Because the cells can be frozen in large quantities, the cultures can be readily scaled up as well.

  • A new triculture system combines pure populations of neurons, astrocytes, and microglia.
  • The model revealed inflammatory cross talk between microglia and astrocytes.
  • Mutant APP exacerbates this cross talk, implicating it in AD.

Studer noted that the iPSC field has struggled with reliability, with the same protocol often producing different data from one experiment to the next due to impurities. He believes his methodology offers a better option. “The system combines flexibility and reliability with sufficient complexity,” Studer told Alzforum. “The precision is unprecedented.”

Other researchers agreed. “This methodology is a huge step forward for the field,” Cynthia Lemere at Brigham and Women’s Hospital, Boston, told Alzforum. Doo Yeon Kim, Joseph Park, Matthias Hebisch, and Rudolph Tanzi at Massachusetts General Hospital, Boston, called the approach elegant. “This full hPSC-derived triculture model will provide a useful tool for studying physiological and pathological neural-glial interactions in human cells, like their previous groundbreaking human cellular models,” they wrote (full comment below).

The authors used this system to investigate inflammation. They found that microglia and astrocytes both had to be present to generate a strong immune response to an inflammatory challenge. Microglia were the instigators, pumping out inflammatory factors that triggered astrocytes to release complement protein C3. C3 then fed back on microglia, spurring them to pour out C3 as well.

The same inflammatory loop engaged when neurons carried the APP Swedish familial Alzheimer’s mutation, with both microglia and astrocytes participating. “To understand the neuroinflammatory response in AD, we think it is essential to have these three cell types together,” Studer told Alzforum. That is not always done in cell culture experiments.

Inflammatory Cross Talk. Under inflammatory conditions, microglia release factors that stimulate astrocytes to produce C3, which feeds back on microglia and triggers them to generate their own supply. [Courtesy of Guttikonda et al., Nature Neuroscience, 2021.]

Studer’s group previously developed protocols for producing pure neuronal and astrocyte populations from embryonic or induced pluripotent stem cells (Chambers et al., 2009; Qi et al., 2017; Tchieu et al., 2019). They wanted a similar method for microglia, which originate, together with macrophages, from a hematopoietic, not neural precursor lineage. The researchers wanted to mimic embryonic hematopoietic development, without embryoid bodies or cumbersome cell-sorting steps, which might contaminate the culture with other cells.

Starting with human pluripotent stem cells (PSCs), first author Sudha Guttikonda pharmacologically activated a burst of WNT signaling for exactly 18 hours before quenching it. This was followed by two days of Nodal signaling, which normally occurs early in embryonic development, to produce hemangioblasts, the precursors to all blood and vascular cells. A week of treatment with hematopoietic growth factors VEGF and FGF2 coaxed these cells into becoming erythromyeloid progenitors, which gave rise to premacrophages. Culturing premacrophages for 10 days with IL-34 and M-CSF generated a nearly pure population of macrophage-like cells. Lorenz noted that macrophages and microglia have the same developmental lineage; macrophages that migrate to the brain become microglia, suggesting that interactions with neurons may complete their patterning.

Making Microglia. In the presence of neurons (green), macrophage-like cells (red) act like microglia, putting out long processes that survey their surroundings. Nuclei are blue. [Courtesy of Guttikonda et al., Nature Neuroscience, 2021.]

To make tricultures, the authors combined these macrophage-like cells with human PSC-derived astrocytes and neurons in a ratio of 2:1:8. Astrocytes proliferated, while some macrophages failed to attach; this led to a final ratio of 1:11:20, similar to the proportions of these cells in the human brain. As expected, once mixed with neurons, macrophages extended cellular processes and behaved like microglia, probing their environments and engulfing synapses (see image above). Their gene-expression profile matched those of microglia generated by other differentiation protocols, and those isolated from human brain (Muffat et al., 2016; Douvaras et al., 2017; Galatro et al., 2017). 

This complement protein rises in inflammatory conditions such as aging and AD. When only neurons and astrocytes were present in the culture, C3 was almost undetectable. Neuron and microglia co-cultures made only a little more. With all three cell types, however, C3 shot up sevenfold. The same pattern occurred when cultures were challenged with lipopolysaccharide (LPS) to produce a vigorous immune response—both glial cell types were required for robust C3 secretion. The amount of C3 doubled under LPS.

How did each glia contribute? To dissect this, the authors used CRISPR to knock down C3 in the human PSCs before differentiating them. When they made tricultures with C3 knockout microglia but wild-type astrocytes and neurons, the media contained almost no C3. This was notable because in the brain, astrocytes, not microglia, are the main source of the protein. The data suggested microglial signaling was necessary to trigger astrocyte C3 production. Confirming this, media from wild-type, but not C3-knockout, microglial cultures stimulated C3 secretion from astrocytes.

The authors profiled C3 knockout microglia, and found they did not produce the pro-inflammatory cytokine TNF-α. Adding TNF-α to astrocyte-neuron cultures triggered C3 release, confirming this cytokine as a mediator of glial cross talk. As with wild-type tricultures, the authors saw the same results with and without LPS challenge, with the sole difference that maximum C3 levels were somewhat higher in the LPS condition.

Meanwhile, tricultures made with C3 knockout astrocytes but wild-type microglia produced about half as much C3 as did control tricultures. This showed that microglia also make C3. Adding C3 to microglia-neuron cultures stimulated additional C3 secretion, indicating the presence of an autocrine feed-forward loop in microglia. Conversely, adding C3 to astroglia-neuron cultures had no effect on astrocyte C3 production.

Altogether, the data suggested that microglia containing C3 kick off an inflammatory loop by releasing TNF-α. This provokes astrocytes to gush C3, which stimulates microglia to produce even more of the complement protein (see model above). “If you take away C3, nearly the whole neuroinflammatory system crashes,” Studer told Alzforum.

What happens in Alzheimer’s disease? When the authors used human neurons carrying the APP Swedish mutation in their tricultures, C3 levels were 64 percent higher than in isogenic wild-type tricultures without adding LPS. As before, both microglia and astrocytes fueled the cycle. It is unclear how mutant APP triggers this glial response. Overproduction of Aβ by itself has no effect on C3, Studer told Alzforum. He is trying to identify the neuronal factor responsible by way of transcriptomic and proteomic profiling.

Mutant APP tricultures also made more of another complement factor, C1Q, than did wild-type cultures, again without adding LPS. C1Q interacts with C3 to stimulate inflammation and synaptic pruning. It is made by microglia and acts on astrocytes. Curiously, when tricultures contained C3 knockout astrocytes or microglia, they accumulated more C1Q than did their wild-type counterparts, hinting that C3 may help trigger phagocytosis and cleanup of C1Q.

Lemere noted that these cell-culture findings resemble data from C3 conditional knockout mice made in her lab. In those otherwise wild-type animals, C1Q rises in the brain after C3 is shut off by inactivating the gene. “This [triculture] model seems to reflect processes going on in human disease,” Lemere noted. “It’s a great system and will be able to quickly define cellular pathways.” However, she cautioned that the findings need to be validated in vivo, because the triculture does not capture age- and time-dependent effects. For example, André Batista in Lemere’s group found that in mice, the increase in C1Q lasts about three months after C3 knockdown before falling back to baseline, perhaps due to compensatory mechanisms kicking in.

Similarly, Dominik Paquet and Julien Klimmt at the University of Munich noted a spatial limitation to these tricultures, in that they are two-dimensional. They believe three-dimensional culture systems will be necessary to fully model processes such as protein aggregation and microglial-synapse interactions. Nonetheless, they said the tricultures could be a powerful tool. “Their simple and modular platform may offer potential for genetic and drug-screening approaches,” they wrote (full comment below). Florent Ginhoux at the Agency for Science, Technology and Research in Singapore agreed the cultures could be useful for identifying therapeutic targets and testing anti-inflammatory interventions (comment below).

This triculture approach is not limited to AD. By replacing cortical neurons with dopaminergic or motor neurons, Studer already uses the same system to study Parkinson’s disease and amyotrophic lateral sclerosis.—Madolyn Bowman Rogers


  1. This interesting study from the Studer lab establishes a defined and modular triculture model of hPSC-derived neurons, astrocytes, and microglia that is useful to study cell-cell interactions and cell-type-specific disease contributions in the context of Alzheimer’s disease. Using APPSwe/Swe knock-in stem-cell line and isogenic control (Paquet et al., 2016), they show that triculture of the three cell types is required to recapitulate inflammatory phenotypes, including increased C3 secretion and C1q deposition.

    Guttikonda et al. started out by developing and carefully characterizing a novel protocol to differentiate microglia from human pluripotent stem cells (hPSCs). The protocol yields highly pure populations of cells that are similar to fetal human microglia on a transcriptomic level and show typical functions such as surveillance of the environment and phagocytosis of synaptic material. The new protocol is based on primitive hematopoiesis followed by patterning of precursors to mature microglia, which is conceptually similar to existing protocols (Pandya et al., 2017; Douvaras et al., 2017; Abud et al., 2017), and it is encouraging to see that the gene-expression signatures of the resulting human microglia-like cells also largely overlap with microglia obtained from those protocols, as well as primary human microglia.

    They next established a triculture system of their microglia with hPSC-derived neurons and astrocytes using protocols published by their lab (Qi et al., 2017; Tchieu et al., 2019), and investigated neuroinflammatory “cross talk” between the different cell types. After challenging the cultures with LPS, they found increased levels of secreted C3, a known marker for neuroinflammation that is also increased in AD brains. Remarkably, the effect was strongest in the tricultures of neurons, astrocytes, and microglia, also compared to neuron-microglia cultures, indicating a cross talk of microglia and astrocytes that potentiates C3 expression. Indeed, further analyses suggested a reciprocal signaling between the two cell types, where microglia activate astrocytes, which in turn (further) activate microglia, leading to synergistic increases in C3 secretion.

    Previous studies had already implicated C3 in AD pathogenesis (Lian et al., 2015; Shi et al., 2017; Wu et al., 2019), but the underlying mechanisms are poorly understood and have not been studied in fully human co-culture models. The authors therefore applied their triculture platform to study neuroinflammation in an AD state using homozygous APPSwe knock-in iPSCs. Interestingly, C3 secretion was increased in tricultures with AD neurons compared to tricultures with isogenic WT neurons (astrocytes and microglia were WT in all cases), while it was not increased in cultures lacking astrocytes and/or microglia. This again corroborated the requirement for all three cell types. Finally, they analyzed C1q, an upstream regulator of C3 that is also known to accumulate in AD, and found increased C1q secretion and accumulation in APPSwe tricultures, with microglia being the main source.

    Taken together their data imply that increased Aβ secretion by neurons causes C1q and C3 secretion in microglia, which in turn stimulates a cycle of increased C3 secretion in astrocytes and microglia that may trigger neuroinflammation. Pro-inflammatory cytokines such as TNFα may also be involved.

    Their findings emphasize the importance of studying not only iPSC-derived human neurons in vitro, but more physiological combinations of brain-cell types to increase the relevance and translatability of the models. The experiments also nicely illustrate how the modularity of the triculture approach can be exploited to get mechanistic insights. Down the line, their simple and modular platform may as well offer potential for genetic and drug-screening approaches to find factors and compounds that break this vicious cycle of reciprocal activation.

    As with every intriguing study, important questions remain that may be studied with the established model:

    Is the activation of microglia directly mediated by increased Aβ production or could other APP-processing products such a β-CFTs play a role (Kwart et al., 2019)? In this context, it would be interesting to directly block Aβ production with inhibitors. How is Aβ detected by microglia and what are the players downstream? Is C1q directly involved as an inductive factor and how does this relate to C3 activation by NFκB (Lian et al., 2015)? Future studies could also investigate if glial activation and subsequent C3 and C1q deposition lead to downstream pathologies in the model, such as synaptic degeneration, for example by aberrant phagocytosis of synapses by microglia for which C1q and C3 are important factors (Hong et al., 2016), and eventually neurodegeneration, if the cultures are kept long-term.

    Finally, while defined and modular tricultures can be very powerful to study cellular crosstalk, as nicely demonstrated by Guttikonda et al., they may not be sufficient to study more complex, three-dimensional physiological cellular interactions, such as myelin involvement and microglia-synapse cross talk, or pathologic effects such as protein aggregation. This would require brain tissue models obtained by three-dimensional co-culture or organoids containing all relevant cell types, in which both Aβ and complement factors cannot easily diffuse away into the media but accumulate and act locally on surrounding cell types (discussed in Klimmt et al., 2020). 


    . Efficient introduction of specific homozygous and heterozygous mutations using CRISPR/Cas9. Nature. 2016 May 5;533(7601):125-9. Epub 2016 Apr 27 PubMed.

    . Differentiation of human and murine induced pluripotent stem cells to microglia-like cells. Nat Neurosci. 2017 May;20(5):753-759. Epub 2017 Mar 2 PubMed.

    . Directed Differentiation of Human Pluripotent Stem Cells to Microglia. Stem Cell Reports. 2017 Jun 6;8(6):1516-1524. Epub 2017 May 18 PubMed.

    . iPSC-Derived Human Microglia-like Cells to Study Neurological Diseases. Neuron. 2017 Apr 19;94(2):278-293.e9. PubMed.

    . Combined small-molecule inhibition accelerates the derivation of functional cortical neurons from human pluripotent stem cells. Nat Biotechnol. 2017 Feb;35(2):154-163. Epub 2017 Jan 23 PubMed.

    . NFIA is a gliogenic switch enabling rapid derivation of functional human astrocytes from pluripotent stem cells. Nat Biotechnol. 2019 Mar;37(3):267-275. Epub 2019 Feb 25 PubMed.

    . NFκB-activated astroglial release of complement C3 compromises neuronal morphology and function associated with Alzheimer's disease. Neuron. 2015 Jan 7;85(1):101-15. Epub 2014 Dec 18 PubMed.

    . Complement C3 deficiency protects against neurodegeneration in aged plaque-rich APP/PS1 mice. Sci Transl Med. 2017 May 31;9(392) PubMed.

    . Complement C3 Is Activated in Human AD Brain and Is Required for Neurodegeneration in Mouse Models of Amyloidosis and Tauopathy. Cell Rep. 2019 Aug 20;28(8):2111-2123.e6. PubMed.

    . A Large Panel of Isogenic APP and PSEN1 Mutant Human iPSC Neurons Reveals Shared Endosomal Abnormalities Mediated by APP β-CTFs, Not Aβ. Neuron. 2019 Oct 23;104(2):256-270.e5. Epub 2019 Aug 12 PubMed.

    . Complement and microglia mediate early synapse loss in Alzheimer mouse models. Science. 2016 May 6;352(6286):712-6. Epub 2016 Mar 31 PubMed.

    . Neurodegeneration in a dish: advancing human stem-cell-based models of Alzheimer's disease. Curr Opin Neurobiol. 2020 Apr;61:96-104. Epub 2020 Feb 26 PubMed.

  2. Emerging evidence strongly supports that neural-glial interactions play a pivotal role in Alzheimer's disease pathogenesis. However, there are few human cellular models for studying these interactions. Lorenz Struder's team described a fully defined human neuron-astrocyte-microglia triculture model that recapitulates the elevated C3 complement cascade in AD.

    Previously, our lab, in collaboration with Hansang Cho, developed a three-dimensional human triculture model of AD, which recapitulates robust neurodegeneration and neuroinflammation, as well as Aβ and phospho-tau accumulation/aggregation (Choi et al., 2014; Park et al., 2018). We used immortalized transgenic human neural progenitor cells to generate AD neurons and astrocytes while using SV40-immortalized human microglia, as well as the microfluidic devices to facilitate robust Aβ accumulation, Aβ-induced tau pathology, and microglial migration.

    In this study, Guttikonda and colleagues used an elegant approach to generate a human triculture model using human pluripotent stem cell (hPSC)-derived neurons, astrocytes, and microglia. This full hPSC-derived triculture model will provide a useful tool for studying physiological and pathological neural-glial interactions in human cells, like their previous groundbreaking human cellular models. We congratulate Struder's team for developing a triculture model of AD with more physiologically relevant cell components.

    However, we would like to caution that the hPSC-derived triculture model might be limited in recapitulating full-blown AD pathology without robust Aβ and Aβ-driven tau hallmarks of AD pathology. We previously found that it is not easy to induce sufficient extracellular Aβ accumulation/aggregation and Aβ-driven tau using the conventional two-dimensional cell-culture conditions without the transgenic approaches (Choi et al., 2014). This iPSC-derived triculture can still provide a valuable model to study neural-glial interactions at the early stage of AD pathogenesis. However, without robust Aβ and tau pathology, the current model might be limited in recapitulating full-blown neuroinflammation and neuronal death, as well as microglial regulation of Aβ clearance in AD brains.

    We wonder if synaptic pruning (via C3 complement pathway activation) and neuronal death (by C1q activating A1 cytotoxic astrocytes) are also elevated in this hPSC-derived triculture model with the APP Swedish mutation. It is also exciting to see the selective elevation of the C3 complement cascade in this model. Still, it has not yet been fully characterized if this is through Aβ-dependent or independent pathways. 

    Despite these concerns, we believe that the improved triculture system described in this study holds great promise for understanding neural-glial cross-talks in AD. We hope this study triggers more sophisticated and physiologically relevant human brain cellular models for basic mechanistic studies and drug testing.  


    . A three-dimensional human neural cell culture model of Alzheimer's disease. Nature. 2014 Nov 13;515(7526):274-8. Epub 2014 Oct 12 PubMed.

    . A 3D human triculture system modeling neurodegeneration and neuroinflammation in Alzheimer's disease. Nat Neurosci. 2018 Jul;21(7):941-951. Epub 2018 Jun 27 PubMed.

  3. This is a great study and another proof of concept that human IPSC-derived culture/organoid systems, although not perfect since not recapitulating fully what is happening in vivo, are providing better models of human physiology and will be useful to gain new insights in the pathophysiology of AD. This study really exemplifies the importance of studying cell type interaction to identify circuits of regulation/reciprocal signaling pathways that could be now targeted on both cell types to potentiate the effect of an anti-inflammatory intervention.

Make a Comment

To make a comment you must login or register.


Paper Citations

  1. . Highly efficient neural conversion of human ES and iPS cells by dual inhibition of SMAD signaling. Nat Biotechnol. 2009 Mar;27(3):275-80. PubMed.
  2. . Combined small-molecule inhibition accelerates the derivation of functional cortical neurons from human pluripotent stem cells. Nat Biotechnol. 2017 Feb;35(2):154-163. Epub 2017 Jan 23 PubMed.
  3. . NFIA is a gliogenic switch enabling rapid derivation of functional human astrocytes from pluripotent stem cells. Nat Biotechnol. 2019 Mar;37(3):267-275. Epub 2019 Feb 25 PubMed.
  4. . Efficient derivation of microglia-like cells from human pluripotent stem cells. Nat Med. 2016 Nov;22(11):1358-1367. Epub 2016 Sep 26 PubMed.
  5. . Directed Differentiation of Human Pluripotent Stem Cells to Microglia. Stem Cell Reports. 2017 Jun 6;8(6):1516-1524. Epub 2017 May 18 PubMed.
  6. . Transcriptomic analysis of purified human cortical microglia reveals age-associated changes. Nat Neurosci. 2017 Aug;20(8):1162-1171. Epub 2017 Jul 3 PubMed.

Further Reading

Primary Papers

  1. . Fully defined human pluripotent stem cell-derived microglia and tri-culture system model C3 production in Alzheimer's disease. Nat Neurosci. 2021 Mar;24(3):343-354. Epub 2021 Feb 8 PubMed.