Microglia are instrumental for anti-viral immunity in the brain
To study whether microglia respond to local cues and are recruited to virus-infected neurons, we made use of the precisely controlled, retrograde transsynaptic spread of the PRV derivative, Bartha-DupGreen (BDG) to central autonomic nuclei [5, 6] from peripheral targets (Fig. 1a). In the hypothalamic paraventricular nucleus (PVN), microglial numbers increased threefold in response to infection and infected neurons were surrounded by numerous Iba1-positive cells (Fig. 1b, c), 6 days after intraperitoneal (i.p.) virus injection. To investigate whether microglia are involved in the control of neurotropic virus infection, we performed selective depletion of microglia, by feeding mice the CSF1R antagonist PLX5622, as demonstrated earlier [20, 56]. After 3 weeks of depletion, 96% of microglia were eliminated from the brain as evidenced by the lack of the microglial markers Iba1 and P2Y12 (Fig. 1d, e). Selective elimination of microglia resulted in a marked increase in the number of virus-infected neurons in the brain (Fig. 1f–g). This phenomenon was not dependent on the route of virus administration, since after i.p. virus injection or injection of the virus into the epididymal white adipose tissue (an organ receiving predominantly sympathetic innervation ), the number of virus-infected neurons was much higher in the brain in the absence of microglia [Fig. 1h; Suppl. Fig. 1 (Online Resource 1)]. In microglia-depleted mice, numerous infected cells were also present in the cerebral cortex already on day 5, and retrograde infection reaching the cortex was more widespread, affecting several areas normally not infected when microglia were present. In addition, more than a threefold increase in the number of disintegrated neurons containing viral proteins was found in the brain parenchyma in microglia-depleted mice (Fig. 1i), indicating the lack of effective elimination of the infected neurons by microglia. To investigate this phenomenon further, we visualized viral proteins using super-resolution microscopy together with the microglial phagosome/lysosome marker CD68 (Fig. 1j, k). In control mice, microglial processes tightly surrounded the cell bodies of infected neurons with viral proteins appearing in microglial phagosomes (Fig. 1j). Importantly, the absence of microglia resulted in a massive increase in extracellular viral proteins and PRV-immunopositive cell debris (Fig. 1k). Confirming these observations, electron microscopy revealed a direct contact between microglial processes and the cell membrane of the infected neurons as well as the uptake of infected neurons by microglia (Fig. 1l, m). In contrast, disintegrated neuronal membranes and extracellular immunogold-labelled viral proteins were observed in microglia-depleted animals (Fig. 1n–p). BDG products, including GFP signal [5, 6], viral capsids and PRV-immunopositive profiles were not observed in the nucleus or the cytoplasm of microglia [Suppl. Fig. 2 (Online Resource 1)], indicating that productive infection does not develop in these cells, in line with earlier reports [6, 50]. The absence of microglia was also associated with the development of diverse neurological symptoms in infected mice starting on the 5th day of infection, when infected neurons were numerous in the brain stem, the hypothalamus and the autonomic-associated nuclei in the limbic system. These symptoms included heavy breathing, muscle spasms and seizure-like episodes, which were absent in control mice (Fig. 1q).
Microglia recruitment is initiated rapidly to virus-infected neurons in the brain
Having confirmed the instrumental role of microglia in controlling neurotropic virus infection, we aimed to investigate whether the recruitment of microglia occurs early enough to allow the isolation of infected neurons prior to the breakdown of neuronal cell membranes. To this end, we made use of the immediate-early marker, GFP, expressed in infected neurons several hours prior to the production of viral structural proteins, which allows time-mapping the different phases of infection at a single neuron level [5, 6, 15]. GFP-positive neurons expressing low levels of viral structural proteins (Phase II cells) already appeared more surrounded by microglia than the majority of neurons expressing GFP only (Phase I, Fig. 2a, b), suggesting that recruitment of microglia is induced within a few hours of infection, by the time viral structural proteins are produced [5, 6, 15]. The number of microglia increased further around neurons expressing high levels of viral proteins (Phase III), resulting in 1–3 microglial cells contacting the cell body of a single, infected neuron (Fig. 2a, b).
To investigate the processes of microglia recruitment in vivo in real-time, we used another virus strain, PRV-Bartha-DupDSRed (BDR), enabling early phases of infection to be identified based on the production of the red fluorescent protein, DSRed . Mice with functional microglia were allowed a longer, 7 days survival after virus injection, resulting in the spread of infection to the upper layers of the cerebral cortex (Fig. 2c). In vivo two-photon imaging in Cx3Cr1+/GFP (microglia reporter) mice revealed the recruitment of microglia within 3 h of the increases observed in neuronal DSRed signal [Suppl. Video 1. (Online Resource 2)]. We used an optimized cranial window preparation for these studies as developed earlier, to avoid any disturbance of microglia . 3D reconstruction from 2P Z-stack revealed that microglial processes formed a barrier-like structure, with several contact points around the cell body of the infected neuron (Fig. 2d). Microglia recruited to infected cells showed increased process velocity compared to microglia distant from sites of virus infection (Fig. 2e), indicating that microglia may respond to local signals induced by infected neurons. To further explore whether microglial contacts with the cell membranes of infected neurons can be formed in the early phases of virus infection, we visualized microglia–neuron contacts with confocal microscopy in Cx3Cr1+/GFP mice, followed by the investigation of selected neurons with correlated electron microscopy and electron tomography. 3D reconstruction from confocal Z-stack revealed the formation of microglial contacts around the cell body and the main dendrites of infected neurons [Fig. 2f; Suppl. Video 2. (Online Resource 3)] prior to the appearance of mature virions in the neuronal cytoplasm (Fig. 2g, h). At this stage of infection, neuronal cell membranes were intact with normal chromatin structure seen in the nucleus [Suppl. Fig. 2b (Online Resource 1)]. Microglial processes surrounding infected neurons showed CD68-immunopositivity, indicating the phagocytic activity of microglia (Fig. 2f, g). In addition, electron tomography revealed the formation of specific membrane interactions between infected neurons and microglia suggesting the recognition and contact of the intact cell membranes by recruited microglial processes (Fig. 2i).
Virus-infected cells are recognised and engulfed by microglia in vitro
To study the mechanisms of microglia recruitment to sites of infection, we first established co-cultures of neurons and GFP-positive microglia from Cx3Cr1+/GFP mice and performed time-lapse imaging over a 48 h period. Microglia contacted the cell body and the main processes of uninfected neurons without causing injury or showing phagocytic activity [Fig. 3a, upper and mid panel and Suppl. Videos 3, 4. (Online Resources 4, 5)]. In contrast, microglia added to virus-infected neurons were recruited to and phagocytosed infected cells [Fig. 3a, bottom panel and Suppl. Video 5. (Online Resource 6)]. Next, we aimed to study the behaviour of microglia with more advanced statistical approaches [24, 43, 44], which required co-cultures of microglia with very sparsely distributed cells. Since this was not feasible with neuronal cultures, we took advantage that astrocytes are also infected with PRV under in vitro conditions . In fact, in sparse astrocyte cultures microglial cells migrated much longer distances on average to reach infected cells. Statistical analysis of longer cell trajectories thus enabled us to more effectively separate random migration from targeted migration of microglial cells to infected cells, followed by localized scanning activity and phagocytosis [Fig. 3b, upper panel and Suppl. Video 6. (Online Resource 7)].
As seen in neuronal cultures, microglia recognised and phagocytosed infected astrocytes, which was confirmed by immunofluorescent detection of the engulfed cells after imaging [Fig. 3b, bottom panel and insert, Suppl. Video 7. (Online Resource 8)]. Statistical analysis revealed the recruitment of microglia to PRV-infected cells and the formation of prolonged cell-to-cell contacts. This was evidenced by microglia trajectories showing characteristic localized pattern as cells tend to remain at virus-infected cells once they met them (Fig. 3c), which is in sharp contrast with microglia trajectories in uninfected cultures showing a random walk behaviour pattern. This phenomenon was associated with a reduction of cell velocities in infected cultures (16.6 ± 3.2 µm/h in control and 10.3 ± 2.6 µm/h in infected cultures, Fig. 3d–g) indicating that signals from infected cells direct microglial migration, scanning behaviour and subsequent phagocytic activity. Similar microglial responses were seen in neuronal/microglial co-cultures [Suppl. Fig. 3 (Online Resource 1)]. Importantly, the development of productive infection was never observed in microglia in vivo or in vitro even after the direct exposure of the cells to high viral titres or following extensive phagocytic activity, as evidenced by the absence of the immediate-early GFP signal and PRV proteins from microglia outside phagosomes [Suppl. Fig. 4 (Online Resource 1)].
Nucleotides released from infected cells trigger microglia recruitment and phagocytosis via microglial P2Y12
To investigate the production of inflammatory mediators induced by neurotropic virus infection, we measured several inflammatory cytokines and chemokines that are commonly upregulated in response to virus infection  in cultured neurons and astrocytes. Bacterial lipopolysaccharide (LPS), a widely used proinflammatory stimulus, induced a robust increase in TNFα, IL-6, CXCL1, CCL5 (RANTES), G-CSF and MCP-1 levels in astrocytes and CXCL1, CCL5 and MCP-1 levels in neurons. In contrast, virus infection increased only CCL5 levels in both cell types at mRNA and peptide levels 24 h after infection [Suppl. Fig. 5 (Online Resource 1)], which did not explain the rapid recruitment of microglia to infected cells.
Since synthesis and release of chemokines could last for several hours and our in vivo data suggested rapid microglia recruitment to sites of virus infection, we checked whether purine nucleotides such as ATP that are chemotactic for microglia at a short time scale  could be released from compromised cells. We found that cultured neurons released ATP after virus infection, which was associated with reduced ATP, ADP, AMP and adenosine levels in cell lysates (Fig. 4a, b), within hours upon the expression of the immediate-early marker, GFP, which precedes the expression of viral structural proteins required for productive infection [5, 6, 15]. The changes in purinergic metabolites were associated with increased ecto-ATPase levels in infected cells (Fig. 4c), but were not due to apoptosis or necrosis, since at the early stages of infection neurons expressing high levels of GFP showed no annexin V binding or uptake of propidium iodide [Suppl. Fig. 6 (Online Resource 1)]. In addition, increased ecto-ATPase levels and NTDPase1 expression were found in microglia at sites of virus infection in the brain (Fig. 4d–f), indicating that microglia respond to changes in the levels of purine nucleotides . To investigate the mechanisms mediating microglial responses to purine nucleotides released from infected cells, we assessed microglial responses in co-cultures of P2X7−/− or P2Y12−/− microglia and wild type astrocytes. Similarly to that seen in wild type microglia (Fig. 3c–g), motility of P2X7−/− cells decreased when exposed to infected cells (Fig. 4g–j) and trajectories showed characteristic localized pattern due to frequent scanning activity [Fig. 4k, Suppl. Video 8. (Online Resource 9)], indicating that P2X7 deficiency does not prevent the recognition of virus-infected cells by microglia. In contrast, virus-exposed P2Y12-deficient microglia showed increased motility (Fig. 4l–o) with trajectories characteristic of random walk behaviour and lacking the localized pattern [Fig. 4p, Suppl. Video 9. (Online Resource 10)], suggesting that these cells are unable to display targeted recruitment in response to infection. Furthermore, wild type and P2X7−/− microglia showed a markedly increased phagocytic activity in infected cultures, which was fully abolished in P2Y12-deficient microglia [Fig. 4q–s and Suppl. Videos 10–11 (Online Resources 11–12)]. Thus, P2Y12 is a key contributor to recognition of compromised cells by microglia and to microglial phagocytosis of virus-infected cells in vitro.
Recruitment of microglia and elimination of virus-infected neurons are mediated by microglial P2Y12 in vivo
Next, we investigated whether nucleotides released from compromised neurons are involved in the recruitment of microglia in vivo. We found that virtually all microglia surrounding the cell body and the processes of infected neurons in either C57BL/6 or Cx3Cr1+/GFP mice expressed P2Y12 receptors [Fig. 5a, Suppl. Fig. 7 (Online Resource 1)]. STORM super-resolution microscopy, which allowed visualization of P2Y12 receptors at 20 nm lateral resolution showed that microglial P2Y12 receptor numbers increased over twofold in response to infection and P2Y12 clusters in microglial processes contacting infected neurons were localized around the membrane of the infected cell (Fig. 5b, c). To investigate the contribution of purinergic signalling to antiviral immunity in vivo, we induced virus infection in mice lacking P2X7 or P2Y12 receptors (Fig. 5d). Both receptors are abundant in microglia, whereas P2Y12 is a microglia-specific marker in the brain  [see also Supp. Fig. 7 (Online Resource 1)]. We found that an absence of P2Y12 resulted in > 50% reduction in the numbers of microglia recruited to infected neurons in the PVN (Fig. 5f, h; p < 0.05), whereas a non-significant trend to reduction (by 35%) was seen in P2X7−/− mice (Fig. 5e, g). Similarly to that seen after selective elimination of microglia (Fig. 1g), the number of infected neurons containing viral structural proteins increased over threefold in P2Y12−/− mice (Fig. 5f, h), but no changes were seen in P2X7−/− mice (Fig. 5e, g). Interestingly, clusters of microglia observed in the brain at sites of virus infection in P2Y12−/− mice were located in the close vicinity of degenerated, PRV-immunopositive neurons, suggesting that P2Y12 deficiency markedly impairs microglial responses to signals released from infected neurons and compromises phagocytic responses, but does not fully block microglial migration to already disintegrated cells.
Despite the markedly increased number of infected neurons in P2Y12−/− mice, no neurological symptoms have been observed (Fig. 5i), suggesting that the absence of microglia (Fig. 1q), but not of microglial P2Y12 alone, can cause the adverse neurological outcome in this model. To confirm this and to test for possible mechanisms underlying this difference, a new study was performed enabling a direct comparison of control, P2Y12−/− and microglia-depleted mice after infection. In P2Y12−/− mice there was deficient recruitment of microglia to infected neurons. As earlier, both the absence of P2Y12 and microglia depletion caused marked elevations in the numbers of infected and disintegrated neurons, but both these measures were elevated significantly more in the microglia-depleted animals (Fig. 6a–c). Histological analysis on cresyl violet-stained brain sections also demonstrated significantly increased neuronal injury/loss in both P2Y12−/− and microglia-depleted animals with highest levels seen after microglia depletion [Suppl Fig. 8 (Online Resource 1)]. However, the neurological symptoms only emerged with microglia depletion (Fig. 6d). In contrast, levels of extracellular virus proteins were identical in P2Y12−/− and microglia-depleted mice, while markedly increased compared to control mice (Fig. 6e, f). P2Y12−/− microglia did show significantly lower levels of CD68-positive phagolysosomes compared to that seen in control animals (Fig. 6g, h), indicating the lack of normal phagocytic activity in the absence of P2Y12.
Microglia recruit leukocytes into the brain in response to virus infection independently of P2Y12-mediated signalling
Since previous studies have shown that blood-borne cells are recruited to the brain after virus infection [15, 48], we wondered whether microglia and P2Y12-mediated actions are involved in neuroinflammatory and neurobehavioral changes in this model. As expected, numerous round-shaped or elongated leukocytes with high CD45 immunopositivity were recruited to sites of virus infection (Fig. 7a). These were clearly discriminated from microglia based on their higher CD45 expression, morphology, the absence of the microglia/macrophage marker Iba1 from the majority of the cells and the complete absence of P2Y12, which is a microglia-specific marker in the brain [Fig. 7b, Suppl Fig. 7 (Online Resource 1)]. P2Y12 is known to be expressed at high levels by microglia compared to monocytes or monocyte-derived macrophages . Surprisingly, selective elimination of microglia resulted in a profound reduction in CD45-positive leukocytes at sites of virus infection (Fig. 7a, c), despite the increased number of infected neurons compared to that seen in control mice (Fig. 1g). This was not due to changes in peripheral leukocyte populations in response to PLX5622, since elimination of microglia by PLX5622 did not cause a significant reduction in circulating or splenic myeloid cell populations including monocytes, granulocytes and macrophages, and did not affect numbers of T cells and B cells [Suppl. Figs. 9 and 10 (Online Resource 1)], confirming our earlier results obtained by another CSF1R antagonist, PLX3397 . Infiltrating leukocyte populations have also been characterized by flow cytometry. The main population of CD45-positive cells recruited in response to virus infection were monocytes (CD45high, Cx3Cr1+, CD11b+, Ly6Chigh, Ly6G− cells), which population was markedly reduced in the absence of functional microglia (Fig. 7d–g). A non-significant trend to increased CD8 T cells in the brain in response to infection was also observed, which was not influenced by microglia depletion [Suppl. Fig. 11 (Online Resource 1)]. We also found that microglia exposed to PRV in vitro produced (CCL5) RANTES and MCP-1, whereas CCL5 and IL-1α were significantly reduced in hypothalamus homogenates of infected mice after microglia depletion [Suppl. Fig. 13 (Online Resource 1)]. Importantly, exaggerated virus infection in the brain was associated with an increase in circulating granulocytes in microglia depleted mice, suggesting that peripheral myeloid populations were capable of responding to virus infection, but their recruitment into the brain was inhibited by the absence of microglia [Suppl. Fig. 9 (Online Resource 1)]. Next, we investigated whether purinergic signalling through P2Y12 in microglia could contribute to leukocyte recruitment into the brain in response to virus infection. Importantly, no changes in the numbers of CD45-positive, blood-borne leukocytes were seen in P2Y12−/− mice after infection and monocyte infiltration was not impaired [Fig. 7g, h; Suppl. Fig. 12a, c, (Online Resource 1)]. Thus, microglia appear to be key inducers of monocyte recruitment into the brain, but these processes are largely independent of microglial P2Y12-mediated mechanisms that play a key role in controlling the spread of virus infection.
Recruitment of P2Y12-positive microglia and leukocytes at sites of infection in the human brain during herpes simplex encephalitis
To investigate microglia recruitment and neuroinflammatory changes in the human brain, we analysed herpes simplex type 1 (HSV-1) encephalitis temporal lobe samples in which infection had been confirmed both by PCR and immunohistochemistry as reported earlier . Processes of P2Y12-positive human microglial cells were extended to HSV-1-positive cells, and infected neurons were surrounded by activated microglial cells [Fig. 8a, Suppl. Table 1 (Online Resource 1)]. This has been confirmed with another specific microglial marker, Tmem119 (Fig. 8b). Infected cells were contacted by 1–3 microglia (on average 1.5 microglia/HSV1 + cell, Fig. 8c). Groups of recruited, amoeboid cells showing CD68-immunopositivity indicating phagocytic activity were also observed at sites of virus infection (Fig. 8d). Microglial cells were negative to HSV antigens suggesting that productive virus infection does not develop in these cells. Leukocytes identified by CD45-immunohistochemistry and Giemsa staining, whilst being P2Y12- and Tmem119-negative have been found at sites of virus infection in close vicinity of HSV-1-positive cells and microglia (Fig. 8e–g). A strong positive correlation between leukocyte numbers and HSV-1-positive cells was also observed (Fig. 8h). Populations of CD15-positive myeloid cells, and in lower amount, scattered CD3-positive lymphocytes and CD20-positive B cells were identified at areas of HSV1 infection (Fig. 8i). In the absence of HSV1 infection, the vast majority of Iba1-positive microglia was found to be P2Y12 positive (96%) and numbers of Tmem119-positive and P2Y12-positive microglia were similar in the brain parenchyma [Suppl. Fig. 14 (Online Resource 1)]. We found that moderate HSV1 infection (less than 50 HSV1-positive cells/mm2) was mostly associated with the activation of local microglia which were Tmem119- and P2Y12-positive. CD68-positive cells with either ramified or amoeboid morphology were also observed in these areas. At areas of advanced HSV1 infection (50–500 HSV1-positive cells/mm2), numerous CD45-positive cells were observed in the brain parenchyma, which was associated with markedly increased numbers of CD68-positive-macrophages (likely to be of both microglial and blood-borne origin). In line with this, the number of ramified microglia, and the total number of P2Y12-positive or Tmem119-positive cells was reduced [Fig. 8j, k and Suppl. Fig. 15. (Online Resource 1)]. Interestingly, similar reduction in microglial numbers was seen in mice at areas showing heavy virus load at the advanced stages of virus infection in the brain [Suppl. Fig. 16 (Online resource 1)].