Selective hippocampal neuronal vulnerability in excitotoxicity: Involvement of microglia
Here, we observed that neurons from the hippocampal CA1, CA3 and DG regions showed distinct and selective neuronal vulnerability towards NMDA-induced excitotoxicity with CA1 neurons being most susceptible to NMDA followed by CA3 and DG neurons, respectively. Similar patterns towards excitotoxicity or (hypoxic-) ischemic insults in the hippocampus have been observed previously both in vivo [34–38] and in vitro in organotypic slice cultures [32, 39–44], corroborating our findings. Interestingly, selectivity towards NMDA has been shown to be independent of an intact hippocampal neuronal circuitry as isolated CA3, CA1 and DG slice cultures still respond with a selective vulnerability towards NMDA, with the CA1 and CA3 regions being more susceptible to NMDA than the DG region . The reasons for these distinct regional differences in neuronal vulnerability, however, are not well understood. It has been shown that CA1 neurons express relatively high levels of AMPA- and NMDA-receptor (-subtypes), while neurons in the CA3 region express relatively high levels of kainate-receptors [46, 47]. Accordingly, it has been demonstrated that CA1 neurons are most vulnerable to glutamate- and NMDA-induced insults, whereas CA3 (and DG) neurons are most sensitive to the excitotoxin kainic acid [32, 37, 43, 48]. Thus, variability in glutamate receptor (-subtype) expression and/or endogenous properties of the distinct neuronal populations in the CA1, CA3 and DG regions [41, 49–51] could (in part) explain their selective vulnerability towards excitotoxicity.
Here, we provide evidence that selective vulnerability is not solely based on endogenous neuronal properties. The differences in neuronal sensitivity to NMDA between the three hippocampal regions disappeared in the absence of microglia. Without microglia, neurons from both the CA3 and CA1 region were equally affected upon treatment with 15-25 μM NMDA and treatment of microglia-free slice cultures with 50 μM NMDA even fully abrogated the selective vulnerability as all three hippocampal regions (CA1, CA3 and DG) were equally affected in terms of neuronal cell death. Since the depletion of microglia was achieved under two different conditions (clodronate treatment in C57BL/6 J slice cultures and ganciclovir application in CD11b-HSVTK slice cultures), we assume that our results are not due to a potential influence of the microglia depletion technique itself. Moreover, we did not find morphological differences in neurons and astrocytes or changes in NMDA receptor subunit mRNA expression nor did we observe differences in MK-801 effects when comparing slice cultures with and without endogenous microglia. Clodronate liposomes are used to target the myeloid cell compartment in brain tissue (in slices or in vivo) for more than 20 years; however, a direct effect in neurons has never been described. The CD11b-HSVTK mouse is now used to target microglia for several years. Although there are fewer publications compared to clodronate, so far no direct effect of ganciclovir treatment on neurons has been found. Even when ganciclovir was administered in vivo intraventrically for up to 2 weeks, no signs of neuronal death were observed as we and others showed [48, 52]. Thus, not the ablation technique but the absence of microglia enhanced the neuronal sensitivity, which is in agreement with earlier findings by us and others [13, 14, 33, 53–55].
We provide here the first evidence that replenishment of microglia-free slice cultures with cultured primary microglia is possible. In a surprisingly straight forward manner, added microglia invade the slice cultures throughout the hippocampal layers, acquire a regular distribution and regain a ramified morphology, which is very similar to endogenous microglia in non-depleted slice cultures. Several important clues/conclusions can be drawn from these observations: 1- The fact that the introduced primary microglia infiltrate the depleted-slice cultures and ramify argues for non-pathological "tissue homeostasis" of these slice cultures. Microglia are active sensors for cellular stress and it is anticipated that their ramification would not have occurred in the presence of damaged or stressed cells. 2- Our findings show that primary microglia, despite the well-known fact that these cells have a high activation status due to culture conditions, keep their capacity to acquire a ramified morphology when brought into a homeostatic neural environment. It is interesting to note here that microglia cell lines, such as BV-2 cells, do not show this behavior. Instead, microglia cell lines remained at the surface of the slice cultures and proliferated until the entire surface of the slices is covered (data not shown); 3- Most importantly, the replenishment with ramified microglia restored the original region specific neuronal sensitivity towards NMDA-induced neurotoxicity indicating that microglia not only acquire a ramified morphology, but also regain their protective function.
These results put a question mark behind numerous studies describing prominent neurotoxic properties of cultured microglia. Clearly, cultured microglia have the ability to damage neurons. However, this prominent neurotoxic phenotype of cultured shake off microglia may rather reflect their special activation status in vitro.
Ramified microglia are not "resting" but protective upon excitotoxicity
Morphological activation of microglia was restricted to sites of neuronal cell death in our slice culture model and thus strictly coincided with the selective neuronal vulnerability to the NMDA-challenge. This morphological activation, induced by neuronal stress or cell death signals, is a well-known feature of microglia that has already been reported for by del Rio-Hortega about a century ago  and ever since numerous times both in vivo  and in vitro in slice cultures . At 10 μM NMDA we did not find neuronal loss in the CA3 and DG regions and therefore no morphological microglia activation was observed in these regions. From the classical point of view (looking at morphology only), one could assume that microglia are not active here. However, neuronal loss was profound in these regions in the absence of microglia, clearly indicating that also ramified (morphologically non-activated) microglia have the capacity to support neurons during an insult. Moreover, these findings suggest that a morphological activation of microglia is not a prerequisite for their neuroprotective function. It is now clear that ramified microglia in vivo continuously scan their environment for homeostatic irregularities [6–8]. The data presented here show that ramified microglia contribute to the protection of neurons and that in conclusion, one should not regard ramified microglia as solely monitoring cells, but as a crucial component that protects neurons from excitotoxicity.
How microglia exert their neuroprotective function remains an open question. Recent studies have uncovered potential mechanisms with which microglia could protect neurons under excitotoxicity. For instance, it was shown that CXC3CL1 expression on neurons leads to secretion of adenosine by microglia, which in turn leads to neuronal increase of adenosine A1 receptors and neuroprotection . Exposure of hippocampal slice cultures to GDNF has been shown to activate microglia, leading to increased neuronal survival . There is also evidence for the involvement of cannabinoid receptor 2 activation in microglia in neuroprotection against excitotoxicity in Huntington's disease . Moreover, microglia in the hippocampus of rats subjected to stroke were found to specifically express neuroprotective TNFα . Furthermore, it was recently described that an intravenous injection of the human microglial cell line HMO6 in ischemic rats leads to reduced infarct size and improved behavioral outcome, suggesting a neuroprotective function of these cells by the upregulation of several inflammatory mediators and neurotrophic factors . In line with these results, it has been shown that application of the microglia cell line BV-2 to slice cultures reduced oxygen-glucose deprivation-induced neuronal damage . These studies, however, related protective function to morphologically activated microglia that improved an ongoing pathology. We show here that in the absence of ramified microglia, brain areas (DG, CA3) are affected by a given insult (NMDA) that would not be damaged in the presence of these cells. Thus, only the ablation and replenishment of ramified microglia as demonstrated in the present study unraveled their protective function.
The CD11b-HSVTK model specifically ablates microglia that undergo activation or proliferation , which is why this model is ideal to deplete microglia in slice cultures. Moreover this model has been used in vivo in several publications. Depletion of microglia in vivo in CD11b-HSVTK mice leads to reduced stroke size  or reduced inflammation-dependent pre-conditioning in pilocarpine-induced seizure activity . Thus, in some instances the lack of microglia appears to be beneficial. However, neuronal death in response to pilocarpine was not assessed in the latter report. In two mouse models of Alzheimer's disease (AD) no change in amyloid-beta plaque load was seen in the absence of microglia . Although the available AD mice do not provide an ideal model for neurodegeneration, these data argue against substantial microglia-driven neuritic damage in AD, as amyloid-beta-driven neural dystrophy appeared to be unaltered in the absence of microglia in the AD mouse models .
In kainate-induced neurotoxicity it recently was found that the ablation of morphologically activated microglia was correlated to reduced neuronal loss, showing that morphologically activated microglia can also promote the death of neurons .
However, none of the reports mentioned above contradicts or supports the findings reported in this study, since a reliable way to deplete and to replenish ramified microglia in vivo has yet not been identified. Thus, although slice cultures are a well accepted model as they represent many in vivo properties, the question whether ramified microglia have a protective function in excitotoxicity also in vivo remains unanswered.
At the moment it is not yet understood how ramified microglia offer protection against NMDA-induced excitotoxicity. Since we did not find significant migration of microglia between the different neuronal regions (CA1, CA3 and DG) in response to NMDA-induced excitotoxicity (data not shown), a local communication between NMDA-treated neurons and surrounding microglia can be envisaged. It was recently described that amoeboid microglia in the developing white matter of rats express functional NMDA receptors . Our data, however, do not support an expression of NMDA receptors in ramified microglia, since depletion of microglia did not change the NMDA receptor expression and function in the slice cultures. Moreover, we just published a mRNA expression analysis of ramified white matter microglia from the adult mouse and did not find significant expression of NMDA receptor mRNA in these cells . It therefore is suggested that the NMDA-receptor expressing amoeboid microglia in the developing white matter recently described by Murugan and colleagues  is a specialized phenotype of microglia and these data add up to the concept that there are subtypes of microglia cells with different functions .
Although insensitive to NMDA, microglia most likely respond to NMDA-challenged neurons given the numerous signals that are released from these cells . Whether or not ramified microglia in response release some of their neurotrophic factors [68, 69] should, however, be further investigated.