Introduction
Multiple sclerosis (MS) is a common neurological disease, characterized by the formation of inflammatory demyelinating lesions in the central nervous system (CNS) [
26]. Inflammation is driven by infiltrating lymphocytes and monocytes, in concert with resident activated microglia and astrocytes. Macrophages and reactive astrocytes are the most abundant cell types in acute lesions [
18,
30]. These cells are highly plastic and can adopt pro-inflammatory, anti-inflammatory, neurotoxic, neuroprotective, and tissue-regenerating functions [
4,
6,
7,
20,
23,
29,
30,
42]. Previous studies have identified macrophage phenotypes in MS lesions based on the expression of single classical (M1) and alternative (M2) activation markers; however, those studies have produced limited and sometimes conflicting results [
6,
42]. It is now increasingly clear that the M1/M2 polarization paradigm, which originated as an in vitro concept, is of limited value for distinguishing myeloid phenotypes in inflamed tissue [
35]. Recent studies, including one of our own, have used single-cell or single-nucleus RNA sequencing on CNS tissue to comprehensively assess the complex phenotypes of human glial cells in healthy and diseased brains [
14,
22,
27]. Similarly, in these studies, myeloid cell/microglial phenotypes did not separate into categories in which M1 and M2 markers are of organizational value.
Several novel histological techniques now make it possible to perform high parameter imaging of tissue sections and to evaluate complex cellular phenotypes in situ [
5,
8,
9,
11,
41]. We have used imaging mass cytometry (IMC), a technique that like mass cytometry (CyTOF) relies on metal isotope-labeled antibodies, and combines immunohistochemistry with high-resolution laser ablation followed by time-of-flight mass spectrometry [
9,
43]. This approach allows for simultaneous quantitative profiling with up to 37 antibodies on a single tissue section at subcellular resolution. Moreover, computational tools have become available to extract single-cell information from highly multiplexed histological data [
3,
24,
36]. In this proof-of-concept study, we applied IMC and single-cell analytics to two active MS lesions – one demyelinating and one post-demyelinating – to examine the cellular heterogeneity of myeloid cells and astrocytes based on thirteen markers known to be expressed by activated glial cells in MS lesions [
2,
6,
13,
15,
31,
42,
44,
45]. We demonstrate that multiplexed tissue imaging, in combination with the appropriate computational tools, can extract previously unattainable information from histological sections, including definition of cellular subpopulations, their distribution within the lesion environment, specific cell-cell interactions, phenotypic transitions and the impact of spatial sources on marker expression.
Discussion
Our study examines the landscape of myeloid and astrocyte phenotypes in early and late acute MS brain lesions using IMC. To our knowledge, this is the first application of highly multiplexed imaging to MS tissue. We applied thirteen markers that are known to be expressed by activated glial cells during MS lesion development. Clustering resulted in eleven myeloid cell and astrocyte phenotypes that localized to distinct lesion areas. Moreover, individual phenotypes interacted selectively with other cell types, and marker expression was driven by different factors in cells located at the lesion rim compared to the center. Thus, our approach provides a wealth of data on cellular spatial organization that is not accessible with standard histology.
The alignment of myeloid cell phenotypes with different lesional layers suggests functional specificity and validates our clustering approach. This spatial separation was most pronounced in the early lesion, and was reduced in the center of the late lesion where multiple phenotypes were intermixed. In addition, marker expression was the highest in myeloid phenotypes located at the lesion rim and diminished substantially towards the lesion center in both lesions. Consistent with the different stages of myelin phagocytosis and degradation, the myeloid phenotypes in the rim were larger than those in the lesion center. An additional feature of the late lesion was the presence of numerous highly activated macrophages in perivascular spaces throughout the lesion. Since these macrophages are believed to transition into the vasculature [
21], this may indicate that they exit the CNS in a highly activated state. In contrast to myeloid cells, marker expression in astrocyte phenotypes did not follow a rim-to-center gradient, but was consistent throughout the lesion.
Our findings argue that macrophages/microglia in MS lesions do not transition from a pro- to anti-inflammatory state, as previously suggested [
6], but convert from a highly activated to a less activated state as they move from the active edge to the lesion center. This is consistent with immunohistological results by Vogel and colleagues demonstrating that pro- and anti-inflammatory markers were simultaneously expressed by macrophages/microglia in MS lesions [
42], and with single nucleus/cell RNA sequencing data of microglial cells in MS and neurodegenerative diseases, which do not produce categories related to M1 or M2 marker expression [
22,
27]. Thus, our results add to the increasing evidence that activated macrophages and microglia in inflamed tissue do not follow a M1/M2 polarization dichotomy.
Using PHATE mapping, we found that myeloid cell but not astrocyte phenotypes followed a linear transition continuum from the G/WM outer rim to the WM outer rim and the lesion center (early lesion), and from lesion center phenotypes to the perivascular phenotype (late lesion). In contrast, phenotype trajectories on Monocle 2 Pseudotime showed no definite transition patterns. Although PHATE and Pseudotime provide biologically accurate transitions when applied to data sets with comparable parametric depth as ours, both methods have previously been shown to produce discrepant results, which may be attributed to their different computational approaches [
24]. Our results deviate from the predicted transition of myeloid phenotypes from the outer to the inner rim and lesion center. Based on the myeloid states defined by our marker panel, myeloid cells develop along several independent fates, rather than one linear phenotype trajectory. We can however, not exclude that inclusion of more or different activation markers may yield different results.
The neighborhood analysis demonstrated distinct cellular interaction signatures for both lesions, e.g. between phagocytic inner rim macrophages and center astrocytes in the early lesion, and between T cells and two myeloid phenotypes in the late lesion. This indicates that cellular interactions in this hypercellular lesion environment are not random, but occur between specific subpopulations and cell types such as lymphocytes. The low parametric depth of our study does not allow us to identify the functional implications of these interactions; however, they may represent nodal points of cellular communication critical for lesion formation and maintenance of low-grade inflammation.
Finally, spatial variance component analysis (SVCA) suggests that cell-extrinsic factors drive marker expression to a higher degree in the lesion rim than in the center. Conversely, cell-intrinsic factors have a more prominent influence on marker expression in the lesion center. This suggests that glia cells in the lesion rim respond to cues from the microenvironment, such as cytokines or receptor-ligand interactions, while glial activation in the lesion center is the result of cell-intrinsic programs set in motion e.g. by myelin phagocytosis.
Myeloid cell/microglial heterogeneity has recently been examined by us and others with single-cell RNA sequencing in the healthy CNS, MS lesions, and other neurological diseases such as Alzheimer’s disease, Parkinson’s disease and temporal lobe epilepsy [
22,
27]. These efforts have identified multiple myeloid cell/microglial phenotypes, comparable with our results. One of the microglia clusters, which was enriched for genes associated with MS susceptibility and characterized by high expression of CD74, was also enriched for genes that were highly expressed in our rim phenotypes (m1 and 5), suggesting that the MS-related CD74
+ phenotype corresponds to our rim myeloid phenotypes. We confirmed this congruency by staining our MS lesions with anti-CD74, which was expressed predominantly by myeloid cells occupying the lesion rim (Additional file
1: Figure S9). Other attempts to cluster myeloid cells in experimental autoimmune encephalomyelitis (EAE), a mouse model of MS, using single-cell cytometry [
25], and in MS lesions using single nuclear RNA sequencing [
14], have yielded substantially less myeloid cell heterogeneity.
Our study is limited by the small sample size and the low number of markers, which may result in inaccurate phenotype clustering. Moreover, we acknowledge that no definite conclusions can be drawn from a comparison of two lesions from different individuals. Nevertheless, as a proof-of-concept study it demonstrates the ability of multiplexed tissue imaging and appropriate single-cell analytics to reveal the heterogeneity and spatial properties of glial cell phenotypes in MS lesions. Future applications may combine cell clustering based on single-nucleus RNA sequencing data with highly multiplexed imaging to obtain maximal parametric depth and spatial resolution of phenotypes. This will help define the phenotypes and key interaction networks that drive acute demyelination and chronic low-grade inflammation in established lesions. This may ultimately provide novel targets for therapeutic intervention in relapsing-remitting and progressive MS.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.