Trem-2 Promotes Emergence of Restorative Macrophages and Endothelial Cells During Recovery From Hepatic Tissue Damage

Macrophages are pivotal in mounting liver inflammatory and tissue repair responses upon hepatic injury, showing remarkable functional plasticity. The molecular mechanisms determining macrophage transition from inflammatory to restorative phenotypes in the damaged liver remain unclear. Using mouse models of acute (APAP) and chronic (CCl4) drug-induced hepatotoxic injury we show that the immune receptor Trem-2 controls phenotypic shifts of liver macrophages and impacts endothelial cell differentiation during tissue recovery. Trem-2 gene ablation led to a delayed re-population of Kupffer cells correlating with deterred resolution of hepatic damage following acute and chronic injury. During tissue recovery, we found that macrophages transitioning to Kupffer cells expressed high levels of Trem-2. Acquisition of the transition phenotype was associated with a unique transcriptomic profile denoting strong responsiveness to oxidative stress and downmodulation of the pro-inflammatory phenotype, which was not observed in absence of Trem-2. During tissue recovery, lack of Trem-2 favored accumulation of a liver-damage associated endothelial cell population (LDECs), whose transcriptional program was compatible with endothelial de-differentiation. Accordingly, LDECs precursor potential is supported by the downregulation of surface endothelial cell markers and by striking in vitro morphological changes towards typical endothelial cells. In conclusion, we found that the dynamics of liver macrophages in response to liver injury are critically controlled by Trem-2 and this regulation is interlinked with the de-differentiation of endothelial cells and heightened liver pathology. We propose that Trem-2 promotes the transition from pro-inflammatory to tissue repair phase by driving the acquisition of restorative properties in phagocytic macrophages.


INTRODUCTION
Hepatotoxic insults elicit a multilayered response involving damaged tissue clearance, scar formation and tissue regeneration. Macrophages play decisive roles in inflammatory and tissue repair responses during acute and chronic liver injury (1)(2)(3) as well as in liver damage due to metabolic disorders such as NAFLD (4,5), type 2 diabetes and obesity (6).
In damaged hepatic tissue, macrophages with different surface phenotypes and activation status show sharp population dynamics (1,3) suggesting that distinct macrophage populations perform specific activities that determine the course of response to tissue damage. Macrophage involvement in response to severe damage is often initiated by influx of hematopoietic-derived monocytes that home the liver as Ly6c + cells and dominate the liver macrophage populations at this stage (2,3,7). These Ly6c + cells present a high-inflammatory phenotype including the expression of TNF-a, IL-1b and TGF-b signals that amplify tissue pathology and also, in chronic tissue injury promote transdifferentiation of stellate cells into collagen-producing myofibroblasts, a hallmark of liver fibrosis (1).
Conversely, tissue repair and fibrosis resolution are associated with the emergence of macrophages that phenotypically resemble liver resident macrophage cells (1,3). These proresolution macrophages show phagocytic ability and express high levels of metalloproteinases and anti-inflammatory mediators (e.g. MMP12 and Arg1) that trigger myofibroblasts apoptosis and actively participate in extracellular matrix degradation (1). Still, however, it is unclear what regulates the dynamics of different liver macrophage populations during response to damage. Macrophages show remarkable phenotypic and functional plasticity and are equipped to undergo functional transitions, depending on contextual cues (8,9). Interestingly, it has been shown that pro-inflammatory macrophages can acquire anti-inflammatory and pro-repair phenotypes (1,3,10) but the triggers for this phenotype switch in liver macrophages remain largely unknown (1,9).
Triggering receptor expressed on myeloid cells-2 (Trem-2) is a transmembrane immune receptor typically expressed in the monocyte/macrophage lineage (11). Upon ligand binding Trem-2 signals through the adaptor DAP12, thereby modulating activation of macrophage effector functions (12). Trem-2 has been intensively studied in the context of neurodegenerative diseases, revealing its concurrent role in the engulfment of abamyloid plaques during Alzheimer's disease (13) and in phagocytosis of apoptotic neurons (14). In addition, Trem-2 signaling was shown to limit tissue destruction and to facilitate both repair and cellular debris clearance in a model of Experimental Autoimmune Encephalomyelitis (EAE) (15). Trem-2 ligands leading to macrophage activation in situ have not been identified, however various studies have reported on a binding to phospholipids such as phosphatidylserine (14,16) and a range of acidic and zwitterionic lipids (13), which may accumulate upon cell damage.
Furthermore, Trem-2 has been shown to modulate microglia survival through Wnt/b-catenin signaling (17,18) and also to promote inhibitory signals that restrain pro-inflammatory macrophage activation (19,20). Recent studies have also illustrated that Trem-2 is required for the activation of a specific transcriptional gene program which controls phagocytosis and lipid metabolism of microglial cells in Alzheimer's disease (21) and of lipid associated macrophages (LAM) in metabolic disorders (22).
The impacts of Trem-2 in liver macrophages have been less explored. Previous findings from our laboratory uncovered that Trem-2 is expressed on Kupffer cells (KCs) determining their activation profile upon contact with malaria parasite (23). Recently published work (24) revealed that Trem-2 is involved in liver damage and proposed that Trem-2 expression in nonparenchymal cells acts as a brake of the inflammatory response during hepatotoxic injury. Although this established a link between Trem-2 and liver inflammation, specific effects on macrophages, the critical players in these processes, remain unsettled.
Here we uncovered that upon experimental induction of hepatic injury Trem-2 controls dynamics of liver macrophage populations favoring replenishment of Kupffer cells and consequently promoting tissue damage resolution and regeneration of the hepatic tissue, including the endothelial cell lineage.

Mice and Experimental Models
All procedures involving laboratory mice were in accordance with national (Portaria 1005/92) and European regulations (European Directive 86/609/CEE) on animal experimentation and were approved by the Instituto Gulbenkian de Ciencia Ethics Committee and the Direccão-Geral de Veterinaŕia (the Official National Entity for regulation of laboratory animals usage). Trem2-deficient mice in a C57BL/6 background (19) were kindly provided by Marco Colonna, Washington University School of Medicine, St. Louis, MO. C57BL/6 mice, Trem-2 KO and B6.Actin-GFP mice were bred and housed under a 12-hr light/dark cycle in specific pathogen free housing facilities at the Instituto Gulbenkian de Ciencia.
In the model of chronic liver fibrosis and fibrosis regression, C57BL/6 and Trem-2 KO males with 7-8 weeks of age received PBS or 20%v/v carbon tetrachloride (CCl4, Sigma, St. Louis, MO, USA) in olive oil, administered at 0,4mL/Kg, twice a week during 4 weeks by intra-peritoneal injections (1). Liver and blood were collected at day 1 (fibrosis) or day 3 post-injection (fibrosis regression).
For the in vivo phagocytosis experiments, mice were given a retro-orbital injection of 50x10 6

Non-Parenchymal Cells Isolation, Flow Cytometry and Cell Sorting
Non-parenchymal cells (NPCs) were isolated from liver lobes by perfusion with Collagenase H (Sigma, St. Louis, MO, USA) followed by density centrifugation as previously described (25,26). Non-parenchymal cells (NPCs) were immuno-labeled with fluorochrome-conjugated antibodies (eBiosciences and BioLegend) followed by flow cytometry analysis (LSR Fortessa X20 ™ , BD) or cell sorting (FACSAria, BD For immunofluorescence, PFA fixed cells were stained overnight at 4°C with rat anti-mouse F4/80 diluted 1:50 and rabbit anti-mouse caspase-3, diluted 1:500. On the following day sections were washed and incubated with respective secondary antibodies for 1h at room temperature. Images were acquired using a 5x5 tile-scan protocol on a Nikon Ti microscope using a 20x 0.75 NA objective, coupled with an Andor Zyla 4.2 sCMOS camera (Andor, Oxford Instruments) and controlled through Nikon NIS Elements (Nikon). DAPI, F4/80 and caspase-3 were acquired using a DAPI, Cy5 and TRITC filtersets, respectively. Detailed methods for analysis can be found in Supplementary Methods.

AST/ALT
Serum levels of Aspartate aminotransferase (AST) and Alanine aminotransferase (ALT) were determined by a colorimetric enzymatic assay using the GOT-GPT kit (Spinreact S.A., Spain) according to manufactures' instructions.

RNA Isolation and Gene Expression Analysis
NPCs were collected in lysis buffer (RNeasy MiniKit-Qiagen) and total RNA was obtained using RNeasy MiniKit (Qiagen) and converted to cDNA (Transcriptor First Strand cDNA Synthesis Kit, Roche). Cells-to-ct kit (Applied Biosystems) was used to amplify cDNA from sorted cells. Taqman gene expression assays (TNFa Mm00443258_m1, trem2 Mm00451744_m1, Applied Biosystems) and endogenous control GAPDH were used in multiplex Real-Time PCR reactions (ABI QuantStudio-384, ThermoFischer). Results represent relative quantification calculated using the 2-DDCT method and normalized to GAPDH.

Cell Sorting and RNA Sequencing Analysis
Macrophage populations and CD45 neg SSC hi population were sorted directly into Qiagen RLT lysis buffer. Each sample represents a pooling of 4 mice. Biological replicates were used for each population except for KCs control. Sequencing was performed at the Genomics Unit, Instituto Gulbenkian de Ciencia (IGC, Portugal) following a previously established protocol (27). Briefly, RNA was separated from gDNA using a modified oligo-dT bead-based strategy and DNA libraries were prepared using Pico Nextera protocol. Sequencing was performed using NextSeq500-High Output Kit v2 (75 cycles), single-ended, 20 million reads per sample. Detailed methods for analysis can be found in Supplementary Methods.

Statistics
To analyze differences across genotypes in treated animals we calculated that a sample size of 5 animals per group would be necessary to have 80% power to detect a mean phenotypic difference higher than 9% at 0.05 significance level. Initial experiments detecting statistical differences, were independently replicated, namely at day 3 post treatments, and their cumulative results are shown. We avoid replicating experiments with initial negative results to prevent unnecessary use of animals. Chi-square test was used to analyze proportions of categorical variables while one-way ANOVA was used to compare means within genotypes and twoway ANOVA to compare means between genotypes. Pearson's correlation was used to test for correlation between two variables and Man-Whitney test was use to compare mean intensity fluorescence ranks. Statistics were calculated with GraphPad Prism version 6.
In Supplementary Methods section we provide additional details on Materials and Methods.

Trem-2 Ablation Deters Tissue Repair Upon Acute Liver Injury
To study the role of Trem-2 in responses to acute liver injury we used a well-established experimental model (3). Mice received a single dose of acetaminophen (APAP) and were analyzed after 1 day (D1) or 3 days (D3), corresponding to the times of hepatic damage and tissue repair responses, respectively ( Figure 1A). During acute injury (D1) wild-type and Trem-2 KO mice showed regions of massive necrosis in the liver ( Figures 1B,  C). At D3 wild-type had almost completely cleared necrosis while Trem-2 KO mice retained marked liver pathology with wide coalescent necrotic areas ( Figures 1B, C). Serum levels of hepatic enzymes AST and ALT at day 1 and day 3 and histological analysis at day 1 indicated that the intensity of APAP hepatotoxicity and the extent of tissue damage was not affected by Trem-2 expression ( Figures 1D, E). Nevertheless, the persistence of larger necrotic areas at day 3 indicate that although wild-type and Trem-2 KO mice were similarly affected by acute liver injury, resolution of liver damage in Trem-2 KO mice was impaired. These data show that although wild-type and Trem-2 KO mice were similarly affected by acute liver injury, resolution of liver damage in Trem-2 KO mice was impaired.

Trem-2 Impacts on Non-Parenchymal Cells Dynamics During Recovery From Acute Liver Damage
Given the role of Trem-2 in macrophage functional activation (19) we isolated non-parenchymal cells (NPCs) from APAPtreated mice at the time points of liver injury (D1) and tissue repair (D3) and performed a detailed analysis of the macrophage lineage cells populations ( Figure 2A). Recruited hepatic macrophages (RHM) (CD45 + Ly6c + F4/80 int CD11b high ) known to promote tissue inflammation (7,28) were predominant at D1 but declined during the tissue repair phase (D3). RHM were found in similar proportions in wild-type and Trem-2 KO mice, suggesting that macrophage recruitment to the liver was not affected in absence of Trem-2 (Figures 2A, B). As expected, Kupffer cells (KCs) (CD45 + Ly6c -F4/80 hi CD11b int ) were highly represented in untreated mice and were severely reduced during injury (D1) in wild-type and Trem-2 KO mice (Figures 2A, B). However, we noted that in the tissue repair phase (D3) the recovery of KCs was slower in Trem-2 KO mice ( Figures 2A, B and Supplementary Figure 1). We have recently shown that levels of CD26 enzymatic activity, from unidentified cellular sources, is a serum biomarker that mirrors severe reductions in KCs population (26). Quantification of CD26  activity in the serum showed that at D3 Trem-2 KO mice reach slightly higher levels of CD26 activity (Supplementary Figure  2A) corroborating the delayed KCs replenishment in Trem-2 KO mice. Interestingly, our analysis revealed two related liver Ly6cmacrophage populations with distinctive surface phenotypes: a CD45 + Ly6c -F4/80 low CD11b + population enriched at D1 and resembling the typical recruited Ly6c + counterpart, herein named as recruited-like macrophages (RLM) ( Figure 2A) and a CD45 + Ly6c -F4/80 + CD11b hi population showing a surface phenotype close to KCs and named as transition macrophages that was strikingly enriched during tissue repair (D3) (Figures  2A, B). In contrast to KCs these populations accumulated in Trem-2 KO mice at D3 ( Figure 2B) indicating that in absence of Trem-2 dynamics of macrophage hepatic repopulation was altered during tissue repair response.
Using GFP-labeled monocyte transfers we show that after acute liver injury recruited bone-marrow monocytes give rise to RLM, transition macrophages and KCs (Supplementary Figure  3). Together, these results suggest that in response to acute liver damage Trem-2 is a determinant of macrophage population dynamics promoting the replenishment of the KCs niche from recruited monocytes.
In addition, flow cytometry analysis uncovered a previously unnoticed non-hematopoietic CD45 neg SSC hi population that emerges at D3 and accumulates in Trem-2 KO mice ( Figures  2A, C) paralleling the accumulation of transition macrophages. Strikingly, we noted that accumulation of non-hematopoietic CD45 neg SSC hi cells inversely correlate with the KCs proportion in the livers of wild-type and Trem-2 KO mice ( Figure 2D) and directly correlated with the persistence of liver necrosis ( Figure  2E). These findings suggest that in absence of Trem-2, imbalanced replenishment of liver macrophages associated with accumulation of transition macrophages results in overrepresentation of a non-parenchymal cell population, which correlates with impaired resolution of liver necrosis.

Trem-2 Ablation Delays Tissue Repair and Alters Non-Parenchymal Cell Dynamics in Chronic Liver Damage
To extend our observations to chronic liver damage wild-type and Trem-2 KO mice were exposed to carbon tetrachloride (CCl4) treatment for 4 weeks and analyzed on day 1 (D1) and day 3 (D3) after treatment corresponding to established liver fibrosis and fibrosis regression time points, respectively ( Figure  3A). At D1 Trem-2 KO mice showed a stronger fibrotic phenotype, with increased hepatocyte necrosis and fiber deposition ( Figures 3B-D). Strikingly, fibrosis resolution response was compromised in Trem-2 KO mice by D3 as indicated by persistence of necrosis and collagen deposition that contrasted with nearly complete necrosis clearance and significant fibrosis regression in wild-type mice ( Figures 3B-D).
Accordingly, analysis of macrophage dynamics revealed that recruitment of macrophages at D1 was not affected in Trem-2 KO mice but at D3 KCs replenishment was impaired and non-hematopoietic CD45 neg SSC hi cells were notoriously overrepresented ( Figures 3E, F). Additionally, serum levels of CD26 activity tend to be higher at D3 in Trem-2 KO mice (Supplementary Figure 2B), strengthening the notion that KCs recovery was delayed in these mice. Similar to the acute model, CD45 neg SSC hi cells accumulation at D3 was inversely correlated with KC proportions ( Figure 3G) and associated with the persistence of liver necrosis ( Figure 3H). Furthermore, genetic expression of TNF-a in non-parenchymal cells (NPCs) correlated with the proportions of CD45 neg SSC hi cell population (Supplementary Figure 4). These data reinforce that ablation of Trem-2 impacts the dynamics of macrophage populations favoring a pro-inflammatory milieu and accumulation of CD45 neg SSC hi cells that are in turn associated with delayed resolution of tissue damage.

Trem-2 Is Upregulated in Transition Macrophages Promoting Acquisition of Resident-Like Phenotype
Trem-2 RNA expression was quantified in sort-purified macrophage populations and non-hematopoietic CD45 neg SSC hi cells from wild-type mice 3 days after APAP treatment (D3) (Supplementary Figure 5A). Remarkably, Trem-2 expression was upregulated in transition macrophages and almost undetectable in the other macrophage populations ( Figure 4A). This strongly suggests that expression of Trem-2 in the transition macrophage population promotes adequate dynamics of KCs replenishment. Furthermore, Trem-2 was not expressed in CD45 neg SSC hi cells ( Figure 4A) suggesting that their accumulation in Trem-2 KO mice may result from abnormal macrophage responses.
Analysis of F4/80 surface expression revealed that Ly6cmacrophages from Trem-2 KO mice present decreased surface expression of F4/80 molecule upon acute ( Figure 4B) and chronic ( Figure 4C) injury. An in vivo phagocytosis functional assay with fluorescent beads at D3 after APAP treatment show that Trem-2 KO macrophages have decreased ability to phagocytose compared to wild-type ( Figure 4D). Furthermore, we found that Trem-2 KO transition macrophages in culture maintain a rounder shape as assessed by the total F4/80 area ( Figure 4E), while wild-type macrophages acquire a typical KClike morphology. This suggests that Trem-2 is involved in the phenotype switch from transition to resident-like macrophage. Quantification of caspase-3 expression in culture and ki67 ex vivo staining in Trem-2 KO transition macrophages showed increased apoptosis and decreased proliferation (Figures 4E, F). These results indicate that ablation of Trem-2 impairs KCs replenishment possibly by controlling cell survival and therefore allowing the transition into KCs.

Trem-2 Increases Resilience to Oxidative Stress and Swift Shutdown of Pro-Inflammatory Program in Transition Macrophages
To better discern the functional role of Trem-2 in macrophage phenotypic shifts we performed transcriptomic analysis in sortpurified macrophage populations of wild-type and Trem-2 KO mice (Supplementary Figure 4A) at D3 after APAP treatment. Hierarchical algorithms and principal component analysis (PCA) clustered the different macrophage populations and clearly showed that the transcriptional profiles of RHM are in the vicinity of RLM and that transitional macrophages are closer to KCs ( Figure 5A and Supplementary Figure 5A). This is in agreement with accepted notions that recruited macrophages (RHM) home the liver as inflammatory macrophages and loose Ly6c expression subsequently giving rise to transition macrophages and KCs (1,29).
On the other hand, wild-type and Trem-2 samples were clustered within each macrophage population ( Figure 5A), suggesting that Trem-2 does not have a major impact in the global macrophage transcriptional programs. Given that Trem-2 affects the dynamical switching of transitional macrophage populations we performed a detailed analysis of differentially  expressed (DE) genes comparing RLM and transition macrophages. We found that the transcriptional shift was more prominent in wild-type mice than in Trem-2 KO mice ( Figures  5B, C). Interestingly, this shift encompassed the upregulation of genes associated to oxidation-reduction processes and downregulation of genes associated with inflammatory responses ( Figure 5D), a pattern that was not observed in Trem-2 KO cells. We measured by qPCR the expression of two of these upregulated genes (Hmox1 and Fth1) in sorted transition macrophages confirming that activation of oxidative stress response mechanisms are blunted in Trem-2 KO transition macrophages (Supplementary Figure 6A). This analysis suggests that the transition macrophage transcriptional program was not fully acquired in Trem-2 KO cells. Furthermore, comparison of transition macrophages and replenished KCs again showed that the transcriptional program switch is considerably less prominent in Trem-2 KO mice ( Figure 5E). Gene ontology analysis revealed that genes specifically upregulated in transition macrophages from Trem-2 KO mice are associated to interferon-beta response, suggesting  that these cells sustained a pro-inflammatory profile ( Figure 5F).
In addition, we analyzed the transcriptional switch between recruited macrophage populations comparing RHM to RLM and found a similar change extent in wild-type and Trem-2 KO mice (Supplementary Figure 6B). Moreover, gene ontology analysis showed that irrespective of Trem-2 expression, RLM have a transcriptional profile of increased proliferative capacity (Supplementary Figure 6C). Taken together these results show that Trem-2 expression in transition macrophages is key to shutdown the pro-inflammatory transcriptional program and increase resilience to oxidative stress during acquisition of resident macrophage functions.   Figures 3G, H). The transcriptomic profiles of sorted CD45 neg SSC hi cells in APAP-treated (D3) and untreated mice (Supplementary Figure 4B) were closely related when using KCs as a reference population ( Figure 6A and   Figure 5B). Gene ontology analysis performed for the DEgenes common to control and APAP-D3 CD45 neg SSC hi cells showed striking enrichment in terms related to endothelial cell identity and function ( Figure 6B). In addition, DE-genes upregulated in CD45 neg SSC hi cells of APAP-D3 treated mice revealed enrichment in pathways related to epithelial to mesenchymal transition and to regulation of Wnt signaling pathway ( Figure 6C). On the other hand, DE-genes upregulated in CD45 neg SSC hi cells from control mice revealed enrichment in functional pathways involved in blood coagulation, hemostasis and fibrinolysis, typical of endothelial cells ( Figure 6D). These transcriptomic data clearly identified the CD45 neg SSC hi population accumulating in the damaged liver as belonging to the endothelial cell lineage leading us to operationally name these cells as Liver Damageassociated Endothelial Cells (LDECs). In addition, the ontology analysis suggests that LDECs are undergoing endothelial de-differentiation. We used in silico analysis to infer interactions between transition macrophages and LDECs during liver tissue repair. Niche Net (30), an algorithm that predicts ligand-receptor interactions by combining transcriptome data of interacting cells revealed that TNF-a, ADAM17, Copa, Ebi3 and Il1b were the top 5 transition macrophages ligands connected with LDECs transcriptional profile, irrespective of the mice genotype ( Figure  6E), This supports that a sustained pro-inflammatory profile during macrophage phenotypic transitions, as observed in Trem-2 KO mice, promotes LDECs accumulation.

LDECs Proliferate and Differentiate In Vitro
We characterized LDECs by flow cytometry using liver endothelial cell surface makers, namely CD31 and CD26. Strikingly, LDECs express lower levels of these markers as compared to endothelial cells from control mice ( Figure 6F). This is line with previous reports showing that endothelial cells undergoing endothelial to mesenchymal transition downregulate endothelial specific markers such as CD31 (31). We next explored the ability of sort-purified LDECs to differentiate in vitro in presence of macrophage-colony stimulating factor (M-CSF), a growth factor able to promote angiogenesis (32). LEDCs were very small and granulous until day 8 when larger sized, round-shaped cells emerged and expanded in the culture, eventually developing typical endothelial morphology by 21 days (Figure 6G). These results reveal that LDECs are at a particular activation state characterized by down-modulation of endothelial cell markers and ability to differentiate into morphologically distinct cells, indicative of their precursor potential.

DISCUSSION
This work revealed that Trem-2 controls the replenishment of liver macrophage populations after acute and chronic hepatotoxic damage and is a critical determinant of swift tissue repair responses conditioning the emergence of endothelial lineage cells during regeneration. Recent reports proposed that lack of Trem-2 expression in non-parenchymal cells contributes to increased fibrosis in Trem-2 KO mice submitted to CCl4 treatment, to increased liver inflammation after acute APAP treatment (24) and to increased susceptibility to hepatocarcinogenesis (33). We noted that Trem-2 KO mice did not show significantly higher intensity of liver damage during acute injury (APAP-D1) but showed sustained tissue damage throughout the tissue repair phase (APAP-D3). Likewise, Trem-2 KO mice presented slightly heightened necrosis during the inflammatory phase of chronic liver injury (CCl4-D1) and a marked delay in subsequent resolution of tissue necrosis and fibrosis (CCl4-D3). While Trem-2 may influence the initial hepatic inflammatory and fibrotic reactions (24), here we focused on its impact in liver macrophages responses during recovery from drug-induced damage.
Macrophage phenotypic plasticity is well illustrated in the responses to liver tissue damage (8,9). In particular, we identified a transition macrophage population that expresses Trem-2 in high levels and predominates during the recovery phase from acute and chronic damage. These cells are derived from circulating monocytes and both transcriptional and phenotypic profiling placed them as the immediate source of resident-like macrophages (Kupffer cells) in the recovered liver. These observations strongly suggest that Trem-2 signaling plays a key role in the dynamics and efficiency of this phenotypic transition, thus explaining the delayed replenishment of the KCs compartment in Trem-2 KO mice. Noteworthy, recent reports have shown that replenishment of the Kupffer cell niche occurs upon engraftment of circulating monocytes into the perisinusoidal space, which is dependent of coordinate interactions between hepatocytes, stellate cells and endothelial cells in the liver (34). These interactions induce a particular transcriptional program and epigenetic changes, responsible for the induction and maintenance of Kupffer cell identity (35). Also, a scar-associated Trem2+CD9+ subpopulation of macrophages was identified in cirrhotic human livers (36). Interestingly, a recent report has shown that in a mouse model of NASH, Kupffer cells are lost mostly likely by apoptosis which induced Trem-2 expression and replenishment of the KC compartment by monocyte-derived macrophages (37).
In addition, a Trem-2 dependent transcriptional program has been associated with emergence of restorative macrophage populations in the context of brain tissue degeneration and adipose tissue inflammation (21,22). Similarly, liver transition macrophages show upregulation of genes associated with this Trem-2 dependent transcriptional signature (Supplementary Figure 7).
Taken together, our results contribute to the identification of Trem-2 as a critical determinant of macrophage plasticity instrumental for a swift recovery from tissue damage.
Acquisition of the transition phenotype in wild-type macrophages includes down-modulation of pro-inflammatory genes and up-regulation of genes involved in oxidative stress responses, not observed in Trem-2 KO macrophages. Redox regulation is critical to cellular stress control mechanisms (38). Accordingly, transition macrophages from Trem-2 KO mice showed decreased survival and proliferative capacities. These observations are in line with reports of reduced survival of Trem-2 KO microglia cells during neurodegenerative processes (17,39). Furthermore, reactive oxygen species (ROS) were increased in Trem-2 KO bone-marrow derived macrophages and hepatic lipid peroxides were increased during liver damage in Trem-2 KO mice (24). Together, these findings indicate that transition macrophages may play a relevant role in ROS clearance in damaged liver and that Trem-2 expression is likely fundamental for transition macrophage survival and proliferation, therefore allowing their acquisition of KCs phenotype.
Remarkably, ablation of Trem-2 leads to increased accumulation of a non-hematopoietic population that we identified as Liver Damage Endothelial Cells (LDECs). This population appeared during tissue repair phases and its accumulation correlated with the severity of tissue damage. It has been recently proposed that vascular endothelial stem cells residing in the liver are activated upon acute liver injury and act as angiogenesis-initiating cells showing remarkable vascular regenerative capacity (40). On the other hand, a specific subset of liver sinusoidal endothelial cells was found to sustain liver regeneration after hepatectomy by releasing angiocrine trophogens (41). Nevertheless, the development and rules of engagement of endothelial progenitors in liver angiogenic repair responses remain unclear.
LDECs transcriptional profile denotes a differential functional activation status as compared to endothelial cells from untreated livers. Gene ontology analysis revealed that LDECs are involved in biological processes related to epithelial to mesenchymal transition ( Figure 6C). A similar process designated endothelial to mesenchymal transition (EndMT) has been reported to play important roles in pathogenesis of many diseases (42) as well as in regenerative processes (31). EndMT promotes cell de-differentiation, consequently giving rise to mesenchymal stem cells with the ability to differentiate into new cell types (43). These cells were shown to differentiate into endothelial cells that contribute to neovascularization (42). Interestingly, M-CSF, an essential regulator of macrophage development, induced marked LDECs morphological changes in vitro, highlighting their de-differentiated state and intrinsic proliferative and differentiation potential.
In silico analysis of predicted ligand-receptor suggests an interaction between transition macrophages and LDECs, where inflammatory mediators secreted by macrophages activate a transcriptional profile on endothelial cells. Of note, ligands such as TNF-a and Il1b showed established correlation with transcription of Sox9, Selp and Ccr7, which are expressed by LDECs. It is well accepted that TNF-a and Il1b are able to activate endothelial cells and promote angiogenesis (44,45) as well as potentiate leukocyte transmigration into inflamed tissues (46). Noteworthy, a recent report used in silico analysis to highlight liver macrophage-endothelial cell interactions in the context of cirrhosis. The authors describe a particular scar associated macrophage subpopulation expressing Trem2 as well as endothelial populations restricted to the fibrotic niche (36).
These observations corroborate our findings that, in damaged liver, macrophages and endothelial cell populations are interlinked and may represent a hallmark of liver regenerative responses. Nevertheless, the precise mechanism mediating this interaction warrants further investigation.
In sum, this work describes Trem-2 as a promotor of macrophage phenotypic switching during tissue repair, tuning down the recruited macrophage inflammatory profile, enhancing oxidation-reduction responses and allowing KCs replenishment. In parallel, we identified an endothelial cell population (LDECs) that emerges during tissue repair with a distinct transcriptional profile and phenotypic features of endothelial de-differentiation.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

ETHICS STATEMENT
The animal study was reviewed and approved by Comissão de Ética do Instituto Gulbenkian de Ciencia.

AUTHOR CONTRIBUTIONS
IC designed and performed experiments, and drafted the paper. ND designed and performed experiments, and drafted the paper. AB analyzed the RNASeq data. MPM conceived the project and discussed results. CP-G conceived the project, supervised the work, and drafted the paper. All authors contributed to the article and approved the submitted version.

FUNDING
This work was developed with the support of the research infrastructure Congento, project LISBOA-01-0145-FEDER-022170, co-financed by Lisboa Regional Operational Programme (Lisboa 2020), under the Portugal 2020 Partnership Agreement, through the European Regional Development Fund (ERDF), and FCT -"Fundacão para a Ciencia e a Tecnologia" (Portugal). This work was partially supported by ONEIDA project (LISBOA-01-0145-FEDER-016417) co-funded by FEEI -"Fundos Europeus Estruturais e de Investimento" from "Programa Operacional Regional

ACKNOWLEDGMENTS
The authors acknowledge the histology, flow cytometry, genomics, antibody production and bioinformatics units at IGC, in particular Dr. Rui Pedro Faıśca for excellent technical assistance and Alexander Marta for analysis of immunofluorescence images. This manuscript has been released as a pre-print at bioRxiv, doi: https://doi.org/10.1101/823773 (47).

SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2020. 616044/full#supplementary-material . Volcano plots representing differential expressed (DE) genes (q<0.05) between RLM and RHM in wild-type and Trem-2 KO mice. Red dots represent upregulated genes with LogFC>0, while blue dots represent downregulated genes with LogFC<0 significant for q<0.05 (B). Venn diagram representing DE-genes between RLM and RHM which are common to wild-type and Trem-2 KO (middle), exclusive for wild-type (left) or Trem-2 KO (right). Gene Ontology (GO) enrichment analysis in the 'Biological Process' category for DEgenes in wild-type mice and Trem-2 KO mice (C). Heatmap illustrating genes regulated by Trem-2 in transition macrophages. Log fold change (logFC) of genes previously associated to Trem-2 transcriptional signature (21,22). Heatmap represents DE-genes upregulated in transition versus RLM at APAP-D3 in wild-type and Trem-2 KO mice