Splitting the “Unsplittable”: Dissecting Resident and Infiltrating Macrophages in Experimental Autoimmune Encephalomyelitis

Macrophages predominate the inflammatory landscape within multiple sclerosis (MS) lesions, not only regarding cellularity but also with respect to the diverse functions this cell fraction provides during disease progression and remission. Researchers have been well aware of the fact that the macrophage pool during central nervous system (CNS) autoimmunity consists of a mixture of myeloid cells. Yet, separating these populations to define their unique contribution to disease pathology has long been challenging due to their similar marker expression. Sophisticated lineage tracing approaches as well as comprehensive transcriptome analysis have elevated our insight into macrophage biology to a new level enabling scientists to dissect the roles of resident (microglia and non-parenchymal macrophages) and infiltrating macrophages with unprecedented precision. To do so in an accurate way, researchers have to know their toolbox, which has been filled with diverse, discriminating approaches from decades of studying neuroinflammation in animal models. Every method has its own strengths and weaknesses, which will be addressed in this review. The focus will be on tools to manipulate and/or identify different macrophage subgroups within the injured murine CNS.


Introduction
Multiple sclerosis (MS) is a chronic inflammatory and neurodegenerative disease of the central nervous system (CNS) characterized by focal lesions of inflammation, demyelination, gliosis and axonal loss [1]. It is widely acknowledged that the autoimmune pathologies of MS and its most commonly used animal model, experimental autoimmune encephalomyelitis (EAE), are largely driven by self-reactive T cells [2]. However, many other immune and non-immune cells such as B cells, natural killer cells, neutrophils, macrophages, and astrocytes are significantly involved in the complex and highly variable processes that lead to demyelination and finally neurodegeneration [2,3].
Macrophages are the predominant inflammatory cells in active and chronic MS and EAE lesions [4][5][6][7]. They participate in disease progression and remission by providing a plethora of effector functions: macrophages present antigens to auto-reactive T cells and secrete cytokines, chemokines, free radicals, proteases as well as other mediators of tissue injury [8]. At the same time, they can facilitate remyelination by clearing myelin debris and promoting oligodendrocyte progenitor activity [9,10].
Depending on their origin, macrophages in CNS lesions can be categorized into at least two different groups: (1) blood-borne macrophages (moMΦ) that differentiate from monocytes plexus, where macrophages (cpMΦ) are slowly replaced by monocytes [33]. Therefore, when trying to understand the diverse contributions of macrophages in MS and EAE, one has to consider not only moMΦ and microglia, but also pvMΦ, mMΦ and cpMΦ as potential individual players in autoimmunity.

Strategies to Distinguish Infiltrating and Resident Macrophages in the Inflamed CNS
There are several methods and strategies that can be employed to discern infiltrating from resident macrophages in the inflamed CNS. They are described in detail in the following section and summarized in Tables 1 and 2.

Discrimination Based on Anatomical Location
Before resident and monocyte-derived macrophages merge into a mixed population in inflammatory CNS lesions, these cells are easily distinguishable as they occupy different anatomical locations in the healthy organism. CNS myeloid cells can be targeted via injection of lentiviral particles into the mouse or rat brain. This approach can be used to tag them with fluorescent reporters and/or to alter expression of specific genes [34,35]. Apart from this route of administration, specificity for CNS myeloid cells has been accomplished by placing the transduced genes under regulation of e.g., the CD11b promoter [35] or microRNA-9 (miR-9) [34]. This is necessary as lentiviral vectors unselectively integrate into all the different CNS cell types. While the CD11b promoter is active specifically in myeloid cells, the latter are the only CNS cells lacking miR-9, which abolishes expression of transgenic mRNA carrying miR-9 target sequences.
Ding et al. showed that intracerebroventricular virus injection targeted more than 80% of microglia and pvMΦ of the brain and spinal cord [35], whereas the intracranial route only led to transduction of microglia around the injection site [34]. As CNS lesions appear at random locations in EAE, the latter approach does not seem favorable in this setting. Moreover, while the CD11b promoter drives stable and strong expression in CNS macrophages, it is unclear whether miR-9-based targeting is able to permit transgene expression in transduced microglia under EAE conditions. Åkerblom et al. reported the expression of delivered green fluorescent protein (GFP) irrespective of the activation status. Yet miR-9 has been shown to participate in microglial activation [36] and might therefore also block transgene expression following certain activating stimuli.
In vivo transduction bears the advantage that targeting of various modifications can be achieved without the need for genetic mouse models. However, application of lentiviral vectors into the CNS is technically challenging, highly invasive, and immunogenic [37]. It might therefore even influence the course of subsequent EAE. Besides, retro-/lentiviral transduction comes along with the disadvantages of variation due to copy-number effects as well as random integration. The latter will affect non-macrophage cells as well, even if they do not express the transduced gene.

Discrimination Based on Cell Morphology and Ultrastructure
Microglia show a unique ramified morphology in the healthy adult CNS. They present themselves with slim branching processes stretching from a compact cell body [38]. In the case of an inflammatory insult however, activated microglia retract their processes and shift to an amoeboid morphology, while migrating towards the site of inflammation [39,40]. Here, monocytes may enter the CNS through the leaky BBB and differentiate into macrophages, which share a similar amoeboid morphology. It is therefore acknowledged that cell shape cannot be used as a criterion to distinguish these two populations in CNS inflammation.
While this might be true for conventional light microscopy applications, successful discrimination between microglia and monocyte-derived macrophages has been reported in EAE using serial block face scanning electron microscopy (SBF-EM) [41]. Microglia were shown to have larger cell volume and a higher number of primary processes than monocyte-derived macrophages, while the latter were also characterized by shorter, thicker mitochondria, bi-lobulated or irregular nuclei, and they frequently contained osmiophilic granules and microvilli [41]. It is noteworthy though that these results were validated using Ccr2 rfp/+ ::Cx3cr1 gfp/+ mice. This system itself has certain limitations regarding the identification of monocyte-derived macrophages, which will be discussed in the following section.

Discrimination Based on Marker Expression
Elucidating the different roles of monocyte-derived and resident macrophages in EAE as well as other CNS diseases has been hampered, because these cells both express frequently used microglia/macrophage markers, including fractalkine receptor (CX 3 CR1), CD11b, F4/80, ionized calcium-binding adapter molecule 1 (iba-1) and CD68 [7]. Still several marker-based strategies have been, and are still used to discriminate between these populations, which will be described in the following section. Recently, comprehensive transcriptome analysis of various macrophage populations identified numerous new markers that hold the potential to better define the CNS myeloid compartment in health and disease. The most promising of them will be discussed below.
Many researchers have therefore used the Ccr2 RFP/+ mouse model, which marks CCR2 + monocytes via red fluorescent protein (RFP) expression [44], arguing that the RFP label would identify monocyte-derived macrophages within the macrophage pool [41,45]. Similar claims were made for Ly6C [46]. However, during differentiation into macrophages, classical monocytes downregulate CCR2 and Ly6C to the levels found on CNS myeloid cells [7,47]. CCR2 or Ly6C in combination with e.g., CX 3 CR1 or iba-1 thus only identify cells that are currently in a transitional state between these two cell types. This phase seems to be exceptionally short in the CNS compared to other tissues [48,49] limiting the usefulness of CCR2 and Ly6C as moMΦ markers in EAE even further. In the injured CNS of e.g., Ccr2 rfp/+ ::Cx3cr1 gfp/+ mice, infiltrating classical monocytes as well as differentiating moMΦ can indeed be identified as RFP + GFP lo . However, as soon as these cells have reached the mature CCR2 − CX 3 CR1 + state, they have also shifted reporter expression and again become indistinguishable from resident CNS macrophages, which are RFP − GFP + .
Ly6C-or CCR2-based lineage tracing systems, such as the Ccr2 CreERT2 line [50] crossed to an inducible reporter strain may solve the problem of marker loss during differentiation. Here, tamoxifen-induced Cre recombinase could stably activate the expression of a reporter gene in CCR2 + cells to also label all CCR2macrophage progeny. This system can further be used to introduce genetic modifications specifically into monocytes and moMΦ [50]. Yet, these modifications would also affect NK cells and certain T lymphocytes, which express CCR2 [44,51]. Moreover, the induced recombination only targets CCR2 + monocytes that are circulating in the bloodstream or differentiating into macrophages at a given time point. As monocytes have a short half-life (~20 h) in the circulation [52], tamoxifen treatment has to be applied carefully to ensure optimal targeting of the moMΦ compartment according to the desired experimental setup. At the same time, this allows relatively precise, temporary control of the genetic manipulation within the monocyte pool, in contrast to e.g., regular BM transfer models.

The Microglia CD45 lo/int Phenotype
A popular approach used to identify microglia is based on their uniquely low expression of the tyrosine phosphatase CD45 [53]. Following dissociation of CNS tissue, single cell suspensions can be analyzed via flow cytometry. Here, microglia present themselves as CD11b + CD45 int and can be clearly separated from other macrophages, including mMΦ, pvMΦ and cpMΦ, which are CD11b + CD45 hi . On top of this approach pvMΦ, mMΦ and cpMΦ can further be identified via the expression of the endocytic pattern-recognition receptor CD206 (also known as mannose receptor C type 1 or MRC1) [33,54]. Moreover, mainly pvMΦ and some mMΦ and cpMΦ have been shown to express the scavenger receptor CD163 [55][56][57][58].
While this simple method works well under healthy conditions, its specificity is lost during CNS inflammation. Undisputedly, monocyte-derived macrophages are found within the CD11b + CD45 hi population. However, based on these two markers alone, they cannot be distinguished from their monocyte precursors (CD11b + CD45 hi Ly6C + CX 3 CR1 lo ) as well as from neutrophils (CD11b + CD45 hi Ly6G + ) and monocyte-derived dendritic cells, which also gain entry into the CNS after breach of the BBB [7]. It has also been mostly ignored that with this strategy, monocyte-derived macrophages in the inflamed CNS cannot be separated from non-parenchymal CNS macrophages, which at least in part seem to be related to microglia despite their CD11b + CD45 hi profile [33].
It was suggested that CD206 could be a useful marker to differentiate pvMΦ, mMΦ and cpMΦ from CD11b + CD45 hi moMΦ in the course of CNS inflammation [54]. However, this molecule has long been regarded as a marker for the so-called "M2" phenotype of macrophages [59,60]. Thus, it is conceivable that other macrophage subsets upregulate CD206 under certain circumstances. The same applies to CD163 [60], which has already been demonstrated on activated microglia [61][62][63][64] as well as monocytes/macrophages [65].
Apart from the obvious flaws of the CD45-based separation strategy, there is also the notion that the clear line between CD45 hi macrophages and CD45 int microglia might start to blur during inflammation, as activated microglia in different species have been reported to upregulate CD45 [66][67][68][69][70][71]. Recently, O'Koren and colleagues have disproven this concern using an elegant Cx3cr1 CreER -based lineage tracing approach, which will be discussed in a following section [49]. They show that the expression of CD45, as well as of other markers that have been thought to assimilate their expression to the level of moMΦ, is only marginally altered on activated microglia, therefore allowing a distinction between microglia and moMΦ even in the setting of inflammation. They characterized microglia as CD11b + Ly6C − Ly6G − CD64 + CD45 lo CD11c lo F4/80 lo I-A/I-E − (commonly known as major histocompatibility complex class II), while moMΦ were CD11b + Ly6C − Ly6G − CD64 + CD45 + CD11c + F4/80 + I-A/I-E + .
This strategy is clearly more preferable than the basic two-marker approach, as it excludes non-macrophage cell types from the analysis. It is hitherto questionable whether it is able to distinguish between moMΦ and CNS-resident CD45 hi macrophages, which are for example also F4/80 + and I-A/I-E + [17]. More importantly, it has only been tested in the context of light-induced retinal injury [49]. Whether this approach is applicable to EAE remains to be shown. In previous attempts to improve the CD11b/CD45 strategy, CD39 [46] and CD44 [72] were suggested as additional microglia and monocyte/moMΦ markers, respectively. However, these markers were only validated using other marker-based approaches (Ly6C, CD45) during CNS inflammation [46,72]. Their specificity is therefore still insufficiently resolved.

New Microglial Markers
In-depth transcriptomic analysis of different macrophage populations recently identified several new markers that were termed microglia-specific, including Sall1, Tmem119, HexB, Tgfbr, Olfml3, P2ry12, P2ry13, Gpr34, Fcrls, and Siglech [14,30,31,45,73,74]. However, the significance of some of these studies with respect to EAE is limited by the choice of compared macrophage populations [73], the lack of an exhaustive comparison to other immune cells [73,74] or the missing evaluation under inflammatory conditions [30,73]. Moreover, all those studies used bulk RNA for transcriptome analysis and did not include non-parenchymal CNS macrophages. Other single-cell RNA sequencing studies have revealed highly cell type-specific markers for microglia (P2ry12, Gpr34, Cd83 and HexB) and pvMΦ (CD163, Hpgd, Mrc1, Slc40a1 and F13a1) in the steady state [17]. Among those markers stated above, Sall1 and Tmem119 have been extensively studied and will therefore be discussed here in more detail.
TMEM119 is a transmembrane protein of unknown function and has been characterized as a microglial marker in mice and humans [45,75]. It is developmentally regulated and seems to be expressed by microglia as they mature, starting at postnatal day 14 in mice. Tmem119 mRNA is expressed solely by parenchymal microglia in the CNS, but by no other neural or glial cell type. Importantly, also pvMΦ, mMΦ, and cpMΦ lack Tmem119 mRNA. Furthermore, it was absent at all peripheral sites tested, including the liver, BM, thymus, blood, spleen and peripheral nerve even in the context of inflammation and injury [45].
In contrast to Tmem119, expression of the zinc-finger transcription factor Sall1 was so far only demonstrated in murine microglia [15,31,76]. Sall1 is not expressed in CD45 + hematopoietic cells outside the CNS, but in CD45 − cells in liver, kidney and heart. Within the CNS, microglia are the only cell type that expresses Sall1, as this factor was not detected in neuronal progenitors, neurons, astrocytes and oligodendrocytes. As for Tmem119, cpMΦ lack Sall1 expression, while a small subpopulation (~5%) of pvMΦ and mMΦ was shown to be Sall1 positive [15].
Sall1 has been shown to control microglia identity in vivo by silencing an inflammatory program [15]. It is therefore important to note that intraperitoneal injection of LPS led to the downregulation of Sall1 in microglia [15]. As Sall1 seems to control the expression of other signature genes such as P2ry12, which were downregulated after conditional knock-out of Sall1 [15], this raises the question whether activated microglia lose the expression of most of the newly identified signature genes during CNS inflammation. Accordingly, P2ry12 has already been shown to become downregulated upon microglial activation [39]. Other investigators have also reported the downregulation of microglia signature transcripts following LPS injection [45] as well as in mouse models of amyotrophic lateral sclerosis (ALS) [74] and Alzheimer's disease (AD) [77]. However, Bennet and colleagues stated that activated microglia could still be specifically labeled with anti-TMEM119 antibody despite reduced Tmem119 mRNA levels [45]. It remains to be shown if this is also the case for other signature markers.
The potential of P2ry12, Fcrls, Tmem119, and Sall1 to discriminate microglia from monocyte-derived macrophages has already been tested in models of CNS injury as well as EAE [14,15,45], though these studies relied solely on CCR2 reporter mice as well as BM chimeras (which will be discussed later in more detail), to which certain limitations apply. As transfer of labeled BM for instance typically does not lead to complete chimerism, there is always a small population of non-labeled moMΦ present in the inflamed tissue. Sorting labeled cells from the CNS can still yield a pure moMΦ population, and it was unequivocally shown that these infiltrated cells indeed do not acquire the expression of microglia signature genes [14,15,45]. Conversely, in case a fraction of microglia at the lesion site loses expression of signature markers, the latter will always be misinterpreted as incomplete chimerism in immunohistochemical studies, especially as moMΦ outnumber microglia in lesions during EAE progression [24,72]. In flow cytometry or pooled RNA analysis, this population might even go unnoticed. So while it can be acknowledged that invading moMΦ do not express microglial signature genes and microglia do not lose signature gene expression on a global scale during CNS inflammation, it cannot be excluded that highly activated microglia in the lesion center might strongly shift their expressional profile towards an inflammatory phenotype making them indistinguishable from their monocyte-derived counterparts.
Even if these newly discovered signature genes lose some of their specificity under inflammatory conditions, they still allow an unprecedented definition of microglia in the healthy organism. In some cases, one marker alone seems sufficient for this definition, which opens up opportunities for conditional gene targeting. In the Sall1 CreER mouse line, tamoxifen pulsing leads to recombination of floxed target genes in 90% of microglia [15,78]. As microglia are long-lasting and potentially self-renew via division of differentiated cells, this recombination should be stably maintained within the microglia population [13,79]. Crossing these mice to e.g., a ROSA26 -stop-YFP reporter strain [80] would permit the specific labeling of microglia before the onset of CNS inflammation, while crossing them to e.g., conditional knock-out strains enables targeted manipulation of microglia [15].

Discrimination Based on Self-Renewal and Turnover
Whereas microglia as well as pvMΦ and mMΦ already reside and slowly self-renew inside the healthy CNS, monocytes are constantly produced from HSCs in the BM and first have to travel through the bloodstream to enter the CNS in the event of injury. These differences in the biology of CNS macrophages and monocytes can be exploited experimentally to discriminate between them and their progeny within CNS lesions, which will be discussed in detail in this section.

Bone Marrow Chimeras
Stable labeling of blood cells including monocytes can be achieved by exchanging their BM source with BM cells that have been collected e.g., from a ubiquitous GFP reporter mouse. Upon transfer of such cells into a myeloablated wildtype recipient, over 90% of circulating monocytes will usually be of donor origin, which can be easily distinguished from tissue macrophages by reporter expression [81,82]. It has been shown that the method for myeloablation has to be chosen carefully. Widely used whole body irradiation damages the BBB [83,84] and promotes artificial engraftment of transferred hematopoietic progenitors, which normally are not present in the bloodstream, into the CNS, where they can differentiate into macrophages [13,24,85]. This engraftment preferentially takes place around blood vessels and within the leptomeninges [86][87][88], where also EAE lesions develop [89,90]. Shielding the head during irradiation can avoid this problem in the brain [23], but not in the spinal cord, where much of EAE pathogenesis takes place [90,91]. It is controversial whether replacement of irradiation with the alkylating agent busulfan as myoablative agent prevents artificial CNS engraftment [81,82,85,92]. This might in fact depend on the exact treatment regimen. Alternatively, the related drug treosulfan has been suggested to prevent CNS engraftment after BM transplantation due to its lower CNS penetration [85].
BM transfer provides the advantage that modifications of blood cells can easily be introduced by using donor cells from knock-out or transgenic mice [23,42,93]. Where the desired models are not available, BM cells can be transduced with retro-or lentiviral vectors ex vivo, before transfer into the recipient [94]. As mentioned earlier, this is associated with the disadvantages of random integration and copy number effects. The development of protocols for the differentiation of murine HSCs from induced pluripotent stem cells in combination with Crispr/Cas9 technology might be able to circumvent this issue in the near future [95].

Parabiosis
Parabiosis enables partial labeling of blood cells, including monocytes. In this approach, the blood circulation of two syngeneic mice is connected surgically. Using one wildtype recipient and one transgenic donor animal, in which e.g., GFP is ubiquitously expressed as a reporter, 40-50% of circulating blood cells in the wildtype animal will be GFP + , while myeloid cells within the CNS remain GFP − . When inducing CNS injury in the recipient mice, infiltrating cells can be tracked via the GFP label [22,24].
Apart from the technical challenge, a major drawback of this model is the low level of chimerism that leads to an underestimation of infiltrating cells [81]. Ajami and colleagues reported that the rate of general chimerism can be increased to 80% by lethal irradiation of the recipient [22]. In this model, the ablated BM of the recipient is populated by physiologically circulating HSCs from the donor [96,97], which has been shielded from irradiation. Following this physiologic HSC transfer, joined mice can be separated again, yielding a BM transfer in the absence of artificial CNS engraftment [24]. However, the observed chimerism (>60% after separation) was still relatively low in comparison to BM transplantation [24]. More importantly, the exchange of blood-borne cells or proteins between parabiotic mice is directly related to their individual half-lives in the circulation [96][97][98]. This is why B and T cells reach equilibrium in the blood, in contrast to cells with high turnover such as polymorphonuclear leukocytes (~40% chimerism in the recipient) and Ly6C hi monocytes (~30% chimerism in the recipient) [99]. Finally, parabionts may experience a higher stress level, which is known to dampen EAE progression [100,101]. This was already confirmed in the parabiotic EAE studies of Ajami and colleagues [24].

Cx3cr1-Based Lineage Tracing
Currently, the most elegant way to discriminate CNS myeloid cells from moMΦ is based on the transgenic Cx3cr1 CreER mouse line, which was generated in parallel by two groups [52,102]. Here, tamoxifen-inducible Cre recombinase is knocked into the fractalkine receptor (CX 3 CR1) gene locus. Crossing these mice with e.g., a ROSA26 -stop-YFP reporter strain [80] enables the induction of stable yellow fluorescent protein (YFP) expression in CX 3 CR1 + cells via tamoxifen pulsing.
The Cx3cr1 promoter is active not only in CNS myeloid cells, but also in circulating monocytes and other peripheral mononuclear phagocytes, including macrophages and dendritic cells as well as myeloid progenitors in the BM [103,104]. Therefore, all of these cells are labeled after the initial tamoxifen pulse. Specificity is achieved based on differences in how these cell populations are maintained over time. While CNS myeloid cells are long-lasting and self-renew from a labeled CX 3 CR1 + population within the CNS [13,33], monocytes rapidly turn over in the circulation and have to be constantly replenished by HSCs. However, HSCs do not express CX 3 CR1 and are thus not affected by the tamoxifen-induced recombination. After a four-week wash-out period, no more YFP + cells can be found in the circulation, whereas resident macrophages inside the CNS retain the YFP label [105]. In health and disease, resident cells can now be distinguished from monocyte-derived macrophages via the expression of the induced reporter [13,33,49,105]. As long as recombination is induced at least four weeks in advance of CNS injury, infiltrating monocytes and their macrophage progeny will not acquire the YFP label, even when their marker signature including CX 3 CR1 becomes indistinguishable from resident CNS macrophages. This is why this strategy is preferable over direct fluorescent reporter models such as Cx3cr1 gfp/+ or Ccr2 RFP/+ . Further advantages of this approach are its non-invasive character, relatively low technical effort as well as its high specificity. High labeling efficiencies ranging from 80% to 99% of microglia can be achieved [13,33,102,105], which might probably depend on the exact protocol used for Cre induction. Still, this method cannot discriminate microglia from other CNS-resident myeloid cells. Perivascular macrophages are equally labeled in this approach, while Goldmann and colleagues reported a reduced labelling of mMΦ (40-50% YFP + ). Although cpMΦ were shown to be slowly replaced by circulating monocytes, about 40% of these cells retained the YFP label after more than 40 weeks post induction [33].
It has been shown that the Cx3cr1 CreER line represents an excellent system to decipher the role of certain factors specifically in CNS myeloid cells during inflammation, e.g., via conditional knock-out [93,105]. Nevertheless, while it achieves clear differentiation between resident and infiltrating macrophages in the CNS, one has to keep in mind that even after the wash-out some peripheral CX 3 CR1 + populations will also be targeted [52]. More importantly, gradual tamoxifen-independent recombination in microglia was observed in one of the generated Cx3cr1 CreER mouse lines [106]. While this did not impair the cellular specificity of this model [102], it obviously abolishes timed recombination. Hence, introduced manipulations can already affect CNS development, in which microglia are essentially involved [107].

Conclusions
In EAE, many attempts have been made to distinguish monocyte-derived from CNS-resident macrophages. These ambitious efforts have been rewarded with accumulating evidence that supports the specialized roles of these cells during disease progression and remission. However, the discrimination and selective targeting of different myeloid populations during EAE remains a significant task. While immunological markers and reporter lines can be used to obtain fast results, these are usually ambiguous and have to be confirmed by BM transplantation or genetic models. Newly identified microglial markers, Cre-based targeting systems as well as growing insight into the myeloid diversity within the CNS bear the potential to revolutionize macrophage research in EAE. Nonetheless, every experimental model holds its innate limitations. Scientists should properly consider these when interpreting previous data and designing future studies. Acknowledgments: We thank Süleyman Ergün for valuable discussions and Chee Keong Kwok for proofreading of the manuscript.

Conflicts of Interest:
The authors declare no conflict of interest.

AD
Alzheimer's disease ALS Amyotrophic lateral sclerosis Batf3 Basic leucine zipper transcription factor, ATF-like 3 BBB Blood-brain barrier BM Bone marrow CCR2 C-C chemokine receptor type 2 CNS