Microglia spend their entire lives cloistered in the brain, but that doesn’t stop them from experiencing the outside world vicariously. According to a study published April 11 in Nature, microglial “memories” of systemic inflammatory events determine the way they respond to subsequent threats on their home turf. Researchers led by Jonas Neher at the German Center for Neurodegenerative Diseases in Tübingen reported that repeat intraperitoneal injections of the bacterial toxin lipopolysaccharide transformed microglia into inflammatory cells that failed to clear Aβ plaques, or into calmer cells that quietly mopped up the detritus. The difference was the number of injections. Memories of the exposures were enshrined in the epigenomes of the microglia, which ultimately dictated their response to Aβ pathology.

  • Depending on how many times a mouse’s immune system was challenged, its microglia became either “trained” or “tolerized.”
  • Tolerized microglia cleared plaques; trained microglia exacerbated them.
  • Epigenetic modifications dictated these microglial responses.

“Systemic immune challenges have long been assumed to affect immune function in the brain,” commented Marco Colonna and Wilbur Song of Washington University in St. Louis. “This interesting and thorough study formally demonstrates that such challenges can have lasting effects on microglia identity and function and, importantly, correlates these effects to lasting epigenetic changes.”

Though immunological memory is classically associated with the T and B cells that make up the adaptive immune system, studies have reported that innate immune responders also have the capacity to remember (e.g., Netea et al., 2016). Circulating monocytes or cultured macrophages pump out inflammatory cytokines when first challenged with lipopolysaccharide (LPS), but become “tolerized,” in immunology parlance, to the bacterial endotoxin after repeated exposure (Collins and Carmody, 2015). Multiple studies have extended this phenomenon to immune cells in the brain, reporting that the responsiveness of microglia is strongly influenced not just by adjacent neuropathology, but also by past bouts of systemic inflammation (Holmes et al., 2009; Mar 2015 conference news; reviewed in Perry et al., 2007, and Cunningham, 2012). Recent studies have started to chip away at the genetic basis for such microglial phenotypes, reporting that microglia take on unique gene-expression signatures when exposed to Aβ pathology or other threats within the brain (Jun 2017 news; Sep 2017 news). 

Co-first authors Ann-Christin Wendeln and Karoline Degenhardt set out to test how microglial responses were shaped by changes outside the brain. For four consecutive days, they injected mice daily in the abdomen with low doses of the lipopolysaccharide, which triggered sickness and temporary weight loss, but not sepsis. As expected, a cadre of inflammatory cytokines, including IL-1β, TNF-α, and IL-6, shot up in the blood shortly after the first injection. However, the response was diminished after the second and subsequent injections, suggesting peripheral tolerance.

In the brain, the opposite was true: The first systemic LPS injection invoked a meager inflammatory response, but the second triggered an inflammatory cytokine explosion. This suggested the microglia had somehow been “trained” by the response to that first LPS injection. However, by the fourth injection of LPS, brain cytokines were barely detectable, suggesting the microglia had been tolerized. Notably, in line with its role as an anti-inflammatory cytokine, IL-10 levels rose in response to successive LPS injections. The researchers proposed that cytokines released by circulating immune cells somehow skewed microglia toward training or toward tolerization.

How would training or tolerization affect microglial response to Aβ pathology? The researchers injected three-month-old wild-type or APP23 mice with one or four doses of LPS, or saline as a control. They measured amyloid plaques six months later. Strikingly, they found plaque burden and total Aβ levels were highest in mice that received a single LPS injection, and lowest in those that received four. This suggested that tolerized microglia aided in plaque clearance, while trained microglia hindered it. Interestingly, microglia surrounded plaques equally in all treatment conditions, but tolerized microglia appeared to gobble up more Aβ. They contained nearly twice the amount as those isolated from animals that had received no LPS or only a single dose.


Microglia (dark cells) surrounded plaques (red) in APP23 mice, regardless of their past exposure to LPS or saline (PBS). [Courtesy of Wendeln et al., Nature, 2018.]

To investigate the genetic basis of microglial training and tolerization, Wendeln and colleagues asked if the different LPS regimens altered the cells’ epigenomes. Six months after treating mice with LPS, they isolated microglia and measured histone modifications that mark active enhancers, grouping tagged genes into biological pathways. The epigenetic profiles were influenced by the prior LPS regimen. For example, microglia from wild-type or APP23 animals that received a single shot of LPS had more active enhancers near genes involved in the HIF1α signaling pathway. In monocytes, this pathway responds to inflammatory stimuli, promoting aerobic glycolysis, and mediating a similar training effect as observed in the microglia (Cheng et al., 2014). Microglia from mice treated with four doses of LPS did not activate HIF1α pathway enhancers, but did activate enhancers near genes involved in the Rap1 signaling pathway, which is involved in phagocytosis. Comparing the epigenetic landscape in microglia from trained and tolerized animals, the former tended to activate enhancers near inflammatory genes.

Aβ pathology appeared to have its own effect on the microglial epigenome. Compared with microglia from wild-type mice, those from APP23 mice activated some of the same enhancers that turned on in animals that received a single LPS dose, including one for HIF1α. Together, the findings suggested that microglial “training,” by either a single dose of LPS or Aβ, modifies the epigenome to promote expression of inflammatory genes, while tolerized microglia have more of a phagocytic epigenetic profile.

Colm Cunningham of Trinity College Dublin drew parallels between the concept of microglial training and the previously reported microglial priming. “The data describing ‘innate immune training’ are highly reminiscent of the long-standing and now well-established description of ‘microglial priming,’” said Cunningham, whose prior work demonstrated that microglial exposure to inflammation steered the outcome of neurodegenerative disease (Cunningham et al., 2005). “Although the idea of ‘training’ gives a sense of doing something beneficial, the term ‘training’ itself carries no information about the consequences: In the end the key issues will be to understand what microglia have been trained by, when they were trained, and what, if anything, are they being trained to do?” Cunningham wrote.

Oleg Butovsky of Brigham and Women’s Hospital added that the concept of microglial memory is not new, pointing to reports that the cells are even shaped by infections that occur during early development (see, e.g., Williamson et al., 2011). 

To determine if the epigenetic effects altered gene expression, the researchers correlated activity of 772 enhancers with expression of their nearest genes. The effects turned out to be modest. Uptick/downtick in both enhancer and gene matched 58 percent of the time, just a tad more than expected by chance. Weak concordance between epigenomic and gene-expression profiles is common in such studies, Neher pointed out, and likely stems from the complexity of epigenetic regulation. Stronger parallels emerged in pathway analysis. Differentially expressed genes and those marked by activated enhancers occurred more often in the same pathway than they would by chance.

To look for broader patterns in the genetic chaos, the researchers ran a weighted gene correlation network analysis on the gene-expression data for all the LPS-treated and APP23 mice. This unbiased statistical approach groups genes into modules based on their level of co-expression and arranges genes within each module into biological pathways. Because the modules have no obvious function or ontology, the researchers labelled them with color names, just for descriptive purposes. Modules red and green contained genes that were upregulated in training conditions; their expression rose in APP23 mice and in response to a single LPS shot, but was suppressed by four shots. The red module included HIF1α, ApoE, and other AD risk factors, such as CD33 and Inpp5d, while the green module contained genes involved in glycolysis, which is promoted by HIF1α signaling.

Butovsky was excited to see HIF1α and ApoE upregulated under these conditions, as this fit with his previous work implicating ApoE signaling as a suppressor of microglial function in neurodegenerative settings (Krasemann et al., 2017). 

Microglia Modules. In wild-type and APP23 mice, expression of modules of co-regulated genes (left) increased or decreased in response to previous LPS exposures, and/or Aβ pathology. Correlation coefficients measured the extent to which gene expression in each module was influenced by genotype or treatment conditions. [Courtesy of Wendeln et al., Nature 2018.]

In keeping with this data, microglia from APP23 mice that had received a single shot of LPS six months prior had elevated mitochondrial membrane potentials and contained lactate—a mark of glycolysis. This suggested that in these mice, microglial training promoted an HIF1α signaling cascade. Because aerobic glycolysis produces ATP less efficiently than respiration, Neher speculated that training saps microglial energy, which could hinder the cells from energy-demanding effector functions, including phagocytosis.

Richard Ransohoff of Third Rock Ventures in Boston agrees that the trained microglia are dysfunctional, but suspects that loss of phagocytosis may not be the main reason they seem to exacerbate plaques. Instead, he wondered if enhanced production of Aβ was to blame. He pointed out that the trained microglia rounded up and retracted their processes, which could prevent them from dampening neuronal activity. This could promote synaptic firing and thus boost Aβ release, he said. Ransohoff was intrigued by the identification of the HIF1α/glycolysis pathway, pointing out that ATP-starved microglia would be even less adept at mingling with neurons. However, he added that while the researchers did a thorough job of investigating the genetic pathways involved in microglial responses to these defined stimuli, how microglia might be shaped by a lifetime of complex inflammatory exposures is still a mystery.

Another group of genes—the gray module—had an opposite pattern of regulation. Gray-module genes were expressed highest in wild-type, downregulated in APP23 mice, and a single shot of LPS stymied their expression further. They include members of the Rap1 signaling pathway that promotes phagocytosis, and also TGF-β, which promotes microglial homeostasis.

Not all homeostatic genes were co-regulated. Butovsky noted that several genes he had previously identified as part of a homeostatic microglial signature appeared to be upregulated in APP23 animals in Neher’s study. “This is not in line with our findings,” he said. He added that it will be crucial to understand how real-life exposures, including chronic stress and disease, affect the function of microglia.

In all, the researchers identified 10 modules of co-regulated genes altered by LPS or APP23. Some commentators emphasized that although LPS treatments influenced gene expression, the presence or absence of Aβ pathology had a more profound effect, but Cunningham countered that given the six-month lag between LPS and gene-expression analysis, it is unsurprising that the APP transgene had a stronger effect.

Echoing reactions from other commentators, Michael Heneka of the German Center for Neurodegenerative Diseases in Bonn wondered how the microglial training and priming paradigms would apply to real-life situations. “In the future, it will be interesting to see which other factors in and outside of the CNS can induce trained immunity, how this is different from microglia priming, and if different factors lead to differential epigenetic reprogramming,” he wrote. “This is of particular importance, since several risk factors, including, but not restricted to, systemic inflammation, midlife obesity, brain trauma, and sedentary lifestyle, share an increased innate immune activity,” he said.—Jessica Shugart


  1. Essentially, systemic immune challenges have long been assumed to affect immune function in the brain. This interesting and thorough study formally demonstrates that such challenges can have lasting effects on microglia identity and function and, importantly, correlates these effects to lasting epigenetic changes.

    One striking feature of the study is that subtly different challenges appear to cause opposite effects on certain readouts and similar effects on other readouts. For some readouts, a shorter-duration challenge leads to deviations from baseline that disappear with longer-duration challenge. Notably, injecting mice once with LPS appears to increase amyloid deposition, while injecting them four times appears to decrease amyloid deposition. This complex relationship makes extrapolating the results difficult. For example, what would be the effect of six injections or 10 injections? What if a different immune stimulus was applied rather than LPS? Would microglia enter a distinct trained state with each successive injection? The answers to such questions will be important for understanding the role of trained immunity in human neurological disease, as the systemic immune stimuli experienced by humans are diverse and innumerable.

    If the microglial-trained immune response in humans is indeed so sensitive to different degrees of stimulation, it may prove hard to target therapeutically.

  2. Wendeln and colleagues present an intriguing study that provides evidence for an innate immune memory in the brain. The expansion of the trained-immunity concept from the periphery to long-lived innate immune cells in the brain is a very promising step forward in our understanding of microglia biology in health and disease. In the future, it will be interesting to see which other factors in and outside of the CNS can induce trained immunity, how this is different from microglia priming, and if different factors lead to differential epigenetic reprogramming. This is of particular importance, since several risk factors, including, but not restricted to, systemic inflammation, midlife obesity, brain trauma, sedentary lifestyle, etc., share in increased innate immune activity (Heneka et al., 2015). 

    While this study identified histone marks in the complete microglia population in diseased brains of, e.g., APP23 mice, it will be interesting to see if trained immunity is a possible mechanism in the recently identified disease-associated microglia (Keren-Shaul et al., 2017). Also, it will be very important to expand this concept of innate immune memory to human brains, which could be a promising lead toward new therapeutic approaches for patients.


    . A Unique Microglia Type Associated with Restricting Development of Alzheimer's Disease. Cell. 2017 Jun 15;169(7):1276-1290.e17. Epub 2017 Jun 8 PubMed.

    . Innate immunity in Alzheimer's disease. Nat Immunol. 2015 Mar;16(3):229-36. PubMed.

  3. The paper from Wendeln and colleagues highlights the complexity of innate immune responses in the context of AD and demonstrates the ability of microglia to learn and modify their behavior in response to immunological stimuli. The authors’ work represents a technical tour de force, relying on numerous unbiased global approaches to amplify our knowledge of the landscape of genes that are modulated after distinct immune challenges. As is typical with intriguing science, I found this work to be thought-provoking. I am left wondering whether the pathways that the authors uncovered are necessary or sufficient for immune training vs. tolerance. One way to answer this type of question would be to apply their tolerance/trained immunity approach to APP transgenic mice serially deficient in candidate genes.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Systemic Inflammation: A Driver of Neurodegenerative Disease?
  2. Hot DAM: Specific Microglia Engulf Plaques
  3. ApoE and Trem2 Flip a Microglial Switch in Neurodegenerative Disease

Research Models Citations

  1. APP23

Paper Citations

  1. . Trained immunity: A program of innate immune memory in health and disease. Science. 2016 Apr 22;352(6284):aaf1098. Epub 2016 Apr 21 PubMed.
  2. . The Regulation of Endotoxin Tolerance and its Impact on Macrophage Activation. Crit Rev Immunol. 2015;35(4):293-323. PubMed.
  3. . Systemic inflammation and disease progression in Alzheimer disease. Neurology. 2009 Sep 8;73(10):768-74. PubMed.
  4. . Systemic infections and inflammation affect chronic neurodegeneration. Nat Rev Immunol. 2007 Feb;7(2):161-7. PubMed.
  5. . Microglia and neurodegeneration: The role of systemic inflammation. Glia. 2012 Jun 6; PubMed.
  6. . mTOR- and HIF-1α-mediated aerobic glycolysis as metabolic basis for trained immunity. Science. 2014 Sep 26;345(6204):1250684. PubMed.
  7. . Central and systemic endotoxin challenges exacerbate the local inflammatory response and increase neuronal death during chronic neurodegeneration. J Neurosci. 2005 Oct 5;25(40):9275-84. PubMed.
  8. . Microglia and memory: modulation by early-life infection. J Neurosci. 2011 Oct 26;31(43):15511-21. PubMed.
  9. . The TREM2-APOE Pathway Drives the Transcriptional Phenotype of Dysfunctional Microglia in Neurodegenerative Diseases. Immunity. 2017 Sep 19;47(3):566-581.e9. PubMed.

Further Reading


  1. . Epigenetic programming of monocyte-to-macrophage differentiation and trained innate immunity. Science. 2014 Sep 26;345(6204):1251086. PubMed.

Primary Papers

  1. . Innate immune memory in the brain shapes neurological disease hallmarks. Nature. 2018 Apr;556(7701):332-338. Epub 2018 Apr 11 PubMed.