Above: Scanning electron micrograph of microglia (white, indicated by black arrows) surrounding neuronal processes. From [Couturier 2011, Figure 1] via OPENi.

Microglia, as the name suggests, are considered glial cells. Glia are a broad category, basically just defined as non-neuronal cells of the nervous system. In the CNS this includes astrocytes and oligodendrocytes, each of which originate from the same multipotent precursors that give rise to neurons [Koblar 1998]. Among glia, microglia are the oddball: they come from something more like a hematopoeitic stem cell lineage, they are considered macrophages, and their closest developmental relatives are other bone marrow-derived macrophages circulating in the periphery, rather than any other cell type in the CNS [Saijo & Glass 2011]. I said “something more like a hematopoeitic stem cell lineage” because it appears that most microglia in the adult mouse brain (and presumably human, though we have less data) appear to have gotten there during embryogenesis, before hematopoeisis even started. Only if mice are whole body irradiated to kill pre-existing microglia do any substantial number of peripheral stem cells migrate to the CNS to become new microglia [Ajami 2007, Saijo & Glass 2011].

Microglia become activated upon brain injury or neurodegeneration, and are postulated to have important roles in both neurodegenerative and psychiatric illnesses. Perhaps the most striking demonstration of this comes from Hoxb8 knockout mice. Hoxb8 is, as the name would suggest, a homeobox gene. Its family members are best known for their roles in laying out a body pattern during embryonic development - see for instance the “antennapedia” mutation in Drosophila in genetics lecture 16. Hoxb8 disruption in mice results in a phenotype of excessive grooming - the mice obsessively clean themselves to the point of tearing out their hair, and they even excessively groom their cagemates, suggesting the origin of the phenotype is psychiatric, rather than being due to, say, an itch owing to a dermatological or peripheral nerve problem [Greer & Capecchi 2002]. If these mice are irradiated and transplanted with wild-type bone marrow, the phenotype is greatly ameliorated. Conversely, if wild-type mice are irradiated and transplanted with Hoxb8 knockout bone marrow, they acquire the excessive grooming phenotype [Chen 2010]. Using a Cre system to delete Hoxb8 only in cells of hematopoeitic origin also recapitulates the excessive grooming phenotype [Chen 2010]. Since microglia are the only cells in the CNS of that lineage, they must therefore mediate the Hoxb8 phenotype. There are precious few animal models of psychiatric disorders [Hyman 2010], so the similarity between excessive grooming in Hoxb8 knockouts and OCD (specifically, trichotillomania) in humans is particularly remarkable. There is also a mouse model of the neurodevelopmental disorder Rett syndrome, caused by mutations in MECP2, and in those mice too, irradiation followed by transplantation with wild-type microglia can rescue the phenotype [Derecki 2012].

A new study that made waves earlier this year demonstrated that microglia may also play a role in schizophrenia. It has been known for years that the single strongest genome-wide association signal with schizophrenia is at the cluster of immune genes on chromosome 6 known as the major histocompatibility complex (MHC) [PGC 2011], and even outside of the MHC, schizophrenia-associated loci are enriched for immune enhancers [PGC 2014]. But no one ever knew why this was the case, or how exactly the immune system might be involved in schizophrenia. The new study [Sekar 2016] showed that the genetic association at MHC traces to complex variation in complement component 4, with higher C4A gene copy number conferring greater schizophrenia risk. Complement components have roles in peripheral immunity, but also in the tagging and elimination of synapses in the brain. Evidence from mice has shown that that microglia react to complement cascade signals to prune synaptic connections in the developing brain [reviewed in Bilimoria & Stevens 2014]. Complement C1q, for instance, is found localized on synapses in the developing mouse brain, and in C1q knockout mice, synapse elimination is impaired and extra synpatic connections are retained into adulthood [Stevens 2007]. This elimination appears to be mediated by microglia, as microglia have been observed physically engulfing presynaptic axonal termini [Schafer 2012]. When complement receptor 3 (CR3) is knocked out, this engulfment activity is reduced, and again, inappropriate synaptic connections are retained, leading to an overall increase in synapse density compared to wild-type mice [Schafer 2012]. Together these findings have suggested the existence of a system where synapses get tagged with complement to mark them for later elimination by microglia. And now it turns out that mice lacking complement component 4 also exhibit inappropriate retention of synapses into adulthood [Sekar 2016]. The authors reasoned, therefore, that the higher C4A copy number could confer schizophrenia risk in humans by causing too many synapses, or the wrong synapses, to get pruned by microglia. They noted that this would make sense with what is known from neuropathology: the brains of people with schizophrenia have reduced gray matter volume but no neuronal loss, consistent with neurons losing their processes but not their bodies [Cannon 2002, Cannon 2015].

The discovery that the dysregulation of synaptic pruning by microglia plays a role in psychiatric disease heightens interest in a question that people have been asking for years: what role do microglia play in neurodegeneration? Microglia eat synapses, and prion disease involves both microglial proliferation and synapse loss, so it is easy to point a finger at them. On top of that, they also move around, so mightn’t they spread prions as they do so? But on the flipside, microglia also clean up apoptotic bodies and may engulf and degrade protein aggregates. So which is it — are they good or bad [Aguzzi 2013]?

For prion disease, at least, it turns out that a large majority of the evidence points to microglia being protective rather than harmful. And earlier this year, new data from the Aguzzi lab have confirmed and extended earlier findings pointing to this protective role [Zhu & Herrmann 2016].

To figure out whether microglia are good or bad, one would like a counterfactual, some way of studying what would happen without microglia. Much of what we know comes from clever mechanistic studies have been made possible by the development of CD11b-HSVTK mice [Heppner 2005], which express herpes thymidine kinase (HSVTK), which confers sensitivity to gancyclovir, under the CD11b promoter, active in microglia and other myeloid cells. Treating these mice with gancyclovir either kills or “paralyzes” their microglia. It also causes a lethal anemia, but this can be rescued/avoided in a few different ways. In [Heppner 2005] they irradiated the mice and then gave them a wild-type bone marrow transplant so that they would have wild-type peripheral macrophages. You might say, doesn’t that mean you just end up with wild-type microglia in the brain, like you had with the Hoxb8 mice that received wild-type transplants? Yes, but evidently the process of microglia migration to the brain is sufficiently slow that it gives you a time window to observe what happens to the brain in the near-total absence of microglial activity. The other way is to administer the gancyclovir via direct infusion into the brain, which is tolerated for about four weeks before it kills the mice [Grathwohl 2009]. And the third option is just to study cultured primary brain tissue ex vivo.

There are a few other tricks out there in the literature too. Knockout of Spi1, encoding the PU.1 transcription factor, results in mice lacking microglia along with many other myeloid cell types) [McKercher 1996, Beers 2006]. Csf1r knockout likewise results in mice without microglia [Dai 2002, Ginhoux 2010], although these mice also have a reproductive phenotype, so presumably the mouse colony has to be maintained in the form of heterozygotes. Il34 knockout mice may be the most tractable of all these options [Wang 2012]. They aren’t totally devoid of microglia — they have <20% as many microglia as wild-type mice — but, conveniently, most other cell types are intact, and their phenotype is relatively mild.

And finally, if your goal is not to study mice without microglia, but to study mice whose bodies have one genotype and whose microglia have a different genotype, you also have the option of whole body irradiating the mice to kill off their bone marrow, and then reconstituting them with bone marrow from other mice. This was the trick used in some of the Hoxb8 and Mecp2 studies mentioned up top.

All in all, there exists a rich and growing toolkit for studying the roles of microglia in vivo, and Aguzzi and others have been using it for years to chip away at the question of what microglia are doing in prion disease.

For one, they did the experiment of irradiating either wild-type or PrP knockout mice to kill off microglia, and then reconstituting the mice with microglia from either wild-type or PrP knockout mice (all four combinations), and then infecting them with prions [Priller 2006]. Reconstitution of wild-type mice with PrP knockout microglia had no effect on disease course compared to wild-type mice with wild-type microglia, and PrP knockout mice reconstituted with wild-type microglia couldn’t replicate prions at all in their brains (though they did exhibit prion replication in their spleens). Conclusion: microglia don’t seem to play much of a role in replicating or spreading prions.

Fun fact: early on in Sonia’s and my scientific quest, one scientist suggested to us that perhaps if all else failed, perhaps total body irradiation followed by bone marrow reconstitution with autologous PrP knockout stem cells could be a cure. This option never sounded particularly appealing, and now, thanks to Priller et al, we know it wouldn’t have worked anyway. Whew.

A couple years later, the Aguzzi lab developed the cerebellar organotypic cultured slice (COCS) assay [Falsig 2008], in which 350 μm thick slices of mouse cerebellum are kept alive in primary culture for weeks and function as a sort of brain in a dish, with all the different cell types of the CNS. Combining this trick with the CD11b-HSVTK genetic manipulation [Heppner 2005], it became possible to see what would happen to prion accumulation in culture without microglia [Falsig 2008]. When Tga20 slices (which express ~10x wild-type levels of PrP) were infected with RML prions, they produced about 5.5 log10 ID50s of prions within 35 days. Killing off microglia with gancyclovir increased this figure to about 6.5 log10 ID50s, and resulted in an obviously visible increase in PK-resistant PrPSc (Fig 5A). Adding purified macrophages back into these microglia-depleted slices reversed the increase in PrPSc accumulation, and, consistent with the findings of [Priller 2006], it did so equally regardless of whether the exogenous macrophages were from Tga20 or PrP knockout mice (Fig 5C). So in the COCS system, at least, microglia didn’t seem to do much to replicate or spread prions — to the contrary, they seemed to be suppressing prion replication, perhaps by digesting prions produced by neurons and/or astrocytes.

A study of prion disease in Mfge8 knockout mice published around the same time provided some insight into how exactly microglia might be helping out. Mfge8 encodes a secreted protein called milk fat globule epidermal growth factor 8, or Mfge8. Understanding how the protein Mfge8 works requires going all the way back to cell biology week 2. Of the phosphoplipids that make up the plasma membrane, phosphatidylserine (PS) is found almost exclusively in the cytoplasmic, as opposed to exoplasmic, leaflet. In apoptotic bodies, PS can become exposed on the outside of the vesicle. Mfge8 binds to PS, decorating these vesicles and serving as a signal for engulfment and degradation. In the periphery, Mfge8 is expressed by follicular dendritic cells, but once the protein is secreted and attaches to apoptotic bodies, it is macrophages that recognize the Mfge8 and engulf the tagged bodies [Kranich 2008]. Analogously, in the CNS, Mfge8 is exclusively expressed in astrocytes, while its receptors are exclusively expressed by microglia, such that Mfge8 secretion induces microglia to digest apoptotic bodies and other extracellular vesicles [Kranich 2010]. Mfge8 knockout accelerates the time to terminal illness in prion-infected mice by about 20% (~160 dpi vs. 200 dpi), with hemizygotes neatly about halfway in between. At terminal illness, the brains of Mfge8 knockouts have increased levels of PK-resistant PrPSc and increased TUNEL staining (a marker of apoptosis). Together these findings suggest that Mfge8 serves to tag prion-infected apoptotic bodies for microglial engulfment, slightly hindering the spread of prions [Kranich 2010].

In 2008, when all this was going on, COCS couldn’t be kept alive long enough in culture to observe neurodegeneration, so it was impossible to test whether depletion of microglia from COCS would help or hurt neuronal survival. After improving its protocols over time, however, the Aguzzi lab was eventually able to observe neuronal death due to prion infection ~50 days post-infection in Tga20 slices [Falsig 2012].

In the new paper [Zhu & Herrmann 2016], they took advantage of this innovation to go back and revisit some of the experiments from [Falsig 2008], now keeping the COCS alive long enough to see how microglial depletion affected neuronal survival. And once again, consistent with everything above, they found that treating prion-infected, CD11b-HSVTK slices with gancyclovir to kill off microglia accelerated neuronal death. One worry was that this might just be collateral toxicity from the gancyclovir, but they did a series of co-culture experiments to effectively rule this out.

They also did the analogous experiment in vivo. This is a hard experiment to do, because as noted earlier, treating CD11b-HSVTK mice with gancyclovir kills off some important population of peripheral myeloid cells, resulting in lethality. In [Heppner 2005] they got around this problem very laboriously, by irradiating the mice and then reconstituting them with wild-type bone marrow. More recently, another group had demonstrated that an alternative (though also laborious) solution is to deliver gancyclovir directly to the brain by intracerebroventricular (icv) pumps [Grathwohl 2009], which apparently kills off only microglia and does not leak out into the periphery in any appreciable quantity. So in the new study, they used the latter approach: Tga20/CD11b-HSVTK mice were infected with prions, implanted with icv pumps, and then either treated with no compound, or with gancyclovir, so that they would have (respectively) normal numbers of microglia, or no microglia, left. And: the mice with microglia got sick at 57 days, while the mice without microglia got sick at 50 days. It’s a narrow time difference, but incubation times are very precise in Tga20 mice, and it was highly significant (P < 0.0001). So now, at last, we have the in vivo answer: prion disease is even worse without microglia.

To validate this finding, they turned to another in vivo model as well — the Il34 knockout mice mentioned above, which have <20% of the normal numbers of microglia [Wang 2012]. Here, too, they found that the microglia-deficient mice had an accelerated disease course (162 vs. 176 dpi for females, 170 vs. 191 dpi for males, so an 8-10% acceleration of disease) [Zhu & Herrmann 2016].

All of these results indicate that microglia play a mostly beneficial role, digesting PrPSc and (albeit unsuccessfully in the long run) working to keep prion replication in check.

That’s not to say that we have it all figured out, or that there is total consensus. One study has reported that GW2580, a small molecule inhibitor of CSF1R (a tyrosine kinase involved in microglial proliferation) extended survival in prion-infected mice, and improved performance on some behavioral tests in late-stage disease, implying that microglial proliferation is harmful [Gomez-Nicola 2013]. Caihong Zhu says this may not necessarily be at odds with the Aguzzi lab’s data — perhaps microglia are overall beneficial, but if over-activated they can do some harm. Other work from Aguzzi’s own group has examined a potentially harmful role for microglia in expressing Nox2, which encodes NADPH oxidase, the primary source of reactive oxygen species in the brain [Sorce 2014].

So there may yet prove to be some degree of dual role, whereby microglia both help and harm at the same time, and accordingly, the authors phrase their conclusions tentatively [Zhu & Herrmann 2016]. Nonetheless, after reading all of this literature, I am impressed by the battery of different experiments that have been done. In science, one never expects perfect unanimity, but the data discussed in this post are in fairly strong agreement that in prion disease, microglia seem to be helpful more than they are harmful.