Dominant-acting CSF1R variants cause microglial depletion and altered astrocytic phenotype in zebrafish and adult-onset leukodystrophy

Tissue-resident macrophages, including microglia, play a significant role in shaping brain development and connectivity as well as maintaining normal brain function [8, 24, 49, 67, 80, 86, 100, 111, 119]. Microglia in particular have been implicated in the pathogenesis of neurodegenerative diseases, including Alzheimer’s disease (AD) and multiple sclerosis (MS) [70, 89, 94, 120]. Strikingly, genetic variants in several microglia-specific genes cause ‘microgliopathies’ [94], a subgroup of leukodystrophies, which are defined as central nervous system (CNS) disorders primarily involving brain white matter. In both neurodegenerative diseases and microgliopathies, microglia increasingly gain interest as potential therapeutic targets, as they can be repopulated using hematopoietic stem cell transplantation (HSCT) or could be pharmacologically depleted by Colony-stimulating factor 1 receptor (CSF1R) inhibitors [17, 106, 107, 123]. It is, therefore, critical to establish a better understanding of microglial function and the consequences of microglial depletion for the human brain.

CSF1R is a protein primarily located on mononuclear phagocytes and considered a key regulator of macrophage and microglial biology [50, 110]. Bi-allelic variants in CSF1R can cause a severe, but variable, developmental disorder termed BANDDOS (brain abnormalities, neurodegeneration, and dysosteosclerosis) [40, 55, 74, 81, 114]. The few patients described presented with dysosteosclerosis, severe white matter lesions and other congenital brain abnormalities, leading to early death [40, 74, 81]. Analysis of post-mortem brain tissue of a patient with a homozygous variant causing disruption of a splice acceptor site showed brain calcifications, axonal spheroids and an almost complete absence of microglia [81]. Similarly, loss of CSF1R in rats, mice and zebrafish causes largely halted macrophage/microglia development, resulting in skeletal and CNS abnormalities [29, 34, 62, 91]. Heterozygous missense variants located in the tyrosine kinase domains (TKDs) of CSF1R have been found to cause the more prevalent adult-onset leukodystrophy, termed ALSP (adult-onset leukoencephalopathy with axonal spheroids and pigmented glia) [95]. Hallmarks of ALSP are motor and cognitive functional decline, often leading to death within several years after disease onset, and neuropathologically progressive white matter lesions, axonal spheroids, pigmented glia and cerebral calcifications [1, 11, 121]. Recent evidence showed that heterozygous CSF1R variants may also be associated with lower density of microglia in ALSP patients [83, 85, 113]. HSCT, which may act by supplying bone-marrow-derived cells that repopulate the microglial niche, was suggested for treatment of ALSP. In mice, HSCT is indeed capable of repopulating the microglial niche, but only after depletion of endogenous microglia [9]. More recently, it was shown that HSCT can halt disease progression in ALSP patients and even reduce pathology visible on MRI [39, 73]. It is possible that the lack of microglia in ALSP facilitates their re-introduction by HSCT. Nevertheless, it remains not fully established whether microglial depletion is the initial pathological event in ALSP, which is important in understanding the mechanisms underlying the highly beneficial effects of HSCT.

Intriguingly, parents of patients with bi-allelic variants, even those with complete loss of function, have not been noted to develop ALSP or neurological dysfunction [40, 55, 74, 81, 114]. While parents of patients with bi-allelic variants mostly have frameshift variants leading to nonsense-mediated mRNA decay (NMD), most ALSP-causing CSF1R variants are missense and located in the TKDs. This is consistent with a model where missense variants act dominantly and, at the biochemical level, cause dysfunctional heterodimers of wild-type and mutant CSF1R, which fail to show tyrosine kinase activity and lead to a ~ 75% suppression of CSF1R activity [50,92]. This is also consistent with findings in haploinsufficient Csf1r-mutant mice where microglia were not reduced [7, 19]. However, support for a dominant effect of ALSP-causing missense variants in vivo is lacking. Gaining insight into how these variants affect microglia, and their subsequent effect on brain health, is crucial to establish and refine treatment strategies for ALSP while also aiding in the development of microglia-focused therapies for other CNS diseases.

Here, we investigated the effects of ALSP-causing CSF1R variants on microglia and pathology correlating with microglial depletion, which could precede white matter degeneration. We integrated neuropathological and multi-omic analysis on post-mortem brain tissue of ALSP patients with in vivo applications in genetic zebrafish models. Zebrafish have been used extensively to study developmental genetics and function of microglia, as they show highly conserved basic properties including CSF1R function [45, 46, 62, 67, 81, 82, 88, 104, 105]. We provide evidence for microglial depletion both in ALSP post-mortem tissue and in vivo in zebrafish with pathogenic CSF1R variants. Furthermore, we identified an unexpected astrocytic phenotype correlating with microglial depletion both in zebrafish models and in ALSP patients. Altogether, our data show that microglial depletion caused by dominant CSF1R variants is a key hallmark and initial pathological event in ALSP. This further supports why HSCT in ALSP patients would be beneficial, possibly by repopulation of the microglial niche.

Here, we establish by independent approaches that microglial depletion due to dominant-acting CSF1R missense variants is an early hallmark of ALSP. Furthermore, by multi-omic analyses in post-mortem ALSP brain tissue and in vivo experimentation in zebrafish mutants, we explored putative consequences of microglial depletion, identifying an altered astrocytic phenotype characterized by increased endocytosis. We also demonstrate that in zebrafish heterozygous pathogenic CSF1R missense variants, but not a heterozygous null allele, result in microglial depletion already in embryonic brain development. This provides in vivo mechanistic evidence that such variants act dominantly, resulting in microglial depletion that may far precede onset of symptoms.

Previously, we found that there is a lower density of IBA1 + microglia in ALSP cortical tissue [83]. Our current analyses of brain tissue of late- and intermediate-stage ALSP extend these findings and show overall reduced homeostatic P2RY12 + and TMEM119 + microglia, particularly in white matter. Consistently, transcriptomic analyses of relatively spared occipital gyrus revealed downregulation of microglia-specific genes. Furthermore, our findings from two ALSP-causing CSF1R missense variants in genome-edited zebrafish models showed microglial depletion in early development. These findings provide in vivo evidence that microglial depletion may far precede onset of symptoms. Only in intermediate-stage patients, numbers of TMEM119 + microglia did not differ from controls in grey matter, in contrast to P2RY12 + microglia. A possible explanation for this is that TMEM119 is also expressed on amoeboid immunoreactive microglia, whereas P2RY12 is considered more exclusive to homeostatic ramified microglia [99, 127]. Therefore, subsets of microglia could be affected differently or possibly P2RY12 would be more sensitive to changes in microglial homeostasis. Previously, we and others observed a clustered distribution of microglia in multiple brain regions in ALSP, where some areas were completely devoid of IBA1 + microglia [83,113]. Kempthorne et al. (2020) also identified reduced expression of microglial genes in ALSP, but speculated that this was due to a loss of homeostatic gene expression rather than a loss of microglia [52]. However, this does not explain the local density differences we observe, including local clustering. A shortage of microglia in one area could perhaps lead to altered distribution by attracting microglia away from other areas, driven by an imbalance in the need and availability of microglia. Clustering of microglia, e.g., in response to cell death or plaques, is typical for various neuropathologies. However, in ALSP, these clusters of immunoreactive microglia, which could develop later in disease as a response to pathology, may further drive depletion in areas with an already low density of microglia by pulling away microglia towards areas of pathology. Microglial depletion caused by recruitment to areas in need could be further reinforced by the defective microglial proliferation. In csf1r-deficient zebrafish, we also observed this phenomenon, where microglial recruitment after neuronal ablation involved migration of resident microglia to the site of injury, in the absence of microglial proliferation [83]. Our data indicate that ALSP involves both a loss of homeostatic microglia as well as a general microglial depletion, predominantly in the white matter. Of note, we observed clusters of CD163 + macrophages/microglia, predominantly in white matter, and an upregulation of CD163 in ALSP tissue. CD163 is associated with an anti-inflammatory microglial signature driving remyelination and found in active MS lesions [72,125]. Since we also noticed brain areas with few CD163 + cells, the CD163 + clusters may again be a consequence of an overall depletion of microglia. Notably, we cannot exclude the possibility that a proportion of these CD163 + cells are infiltrating peripheral macrophages. Altogether, our findings further support the concept of HSCT as a treatment for ALSP, which may act by repopulation of a depleted microglial niche, comparable to the effect of HSCT in adrenoleukodystrophy where pre-lesion areas are also characterized by reduced numbers of microglia [12].

We demonstrate that 2 independent pathogenic CSF1R missense variants in the zebrafish csf1ra locus act dominantly in decreasing density of microglia, whereas a null allele, leading to protein production from only one allele, had no effect. This dominant effect in vivo is further supported by recent observations in mice carrying a heterozygous ALSP-causing CSF1R missense variant (Csf1r^E631K/+), showing reduced numbers of microglia [109]. Also, in vitro biochemical experiments show that pathogenic CSF1R missense variants act dominantly on CSF1R signaling, and parents of bi-allelic patients with heterozygous haploinsufficient CSF1R variants have not been noted to develop brain disease [40, 50, 55, 74, 81, 92, 114]. Furthermore, 86% of pathogenic variants in CSF1R are missense variants, predominantly located in the TKDs, whereas only 4% of pathogenic variants in CSF1R are frameshift variants (Suppl. Table 2, Online Resources). As most of these frameshift variants are located in the C-terminal end, they could in fact also lead to a dominantly acting truncating protein. It is not clear whether any ALSP case can be explained by haploinsufficiency. Indeed, haploinsufficient Csf1r+/- mice show no decrease in microglia, whereas complete knockout mice have no microglia [7, 19, 34]. Therefore, although such models can be very useful, they should be treated with caution when used to study ALSP. What is intriguing, however, is our observation of the differential effects on the number of microglia between CSF1R missense variants. As patients with the same CSF1R variant show major differences in age of onset and disease severity, it is likely that environmental factors and/or genetic factors play a disease-modifying role. This obscures possible genotype/phenotype relationships, and therefore, it has remained unclear whether pathogenic CSF1R variants correlate with ALSP age of onset and severity. Our results suggest that there likely are variant-specific effects contributing to the level of microglial depletion, and by extension possibly also disease severity. Together with possible environmental or genetic modifying factors, this could explain the variance in age of onset and disease course of ALSP patients.

Through unbiased analyses, we identified an altered astrocytic phenotype in ALSP brain tissue and in zebrafish already during embryonic development. In ALSP tissue of severely affected frontal gyrus, but also in relatively spared white matter of the occipital gyrus, astrocytes were more hypertrophic, which is associated with elevated reactivity [35]. Thus, astrocytic abnormalities may be an early consequence of microglial loss—although we cannot exclude that they originate from another cause—and possibly also play a role in early ALSP pathogenesis. Of note, since zebrafish astrocytes morphologically differ from those in mammals, it is not yet evident to what extent the phenotype we observed in radial astrocytes is analogous to that of astrocytes in ALSP, but it nevertheless gives an indication of early embryonic changes correlating with microglial depletion in the vertebrate brain. Notably, radial astrocytic lysosomal inclusions in heterozygous missense zebrafish mutants were transiently present, which could imply that the astrocytic compensatory response is needed temporarily when the phagocytic demand is high, since there is an increased number of apoptotic cells in the zebrafish brain around 3 dpf [45]. The astrocytic phenotype in our zebrafish mutants is reminiscent of the compensatory endocytic response by astrocytes in microglia-depleted mice [30, 57]. Indeed, csf1ra^V614M/+ larvae showed increased endocytosis of myriad substances, including NR dye, apoptotic particles and even myelin. We also observed phagocytized myelin debris inside astrocytes in ALSP brain tissue. This is reminiscent of observations in demyelinating lesions in MS and metachromatic leukodystrophy [90], although the amount of myelin debris within astrocytes appeared remarkably high in ALSP patient tissue. Astrocytes needing to process and degrade phagocytic waste could lead to neglected homeostatic functions, including energy supply to neurons and oligodendrocytes, blood–brain-barrier integrity, and uptake of neurotransmitters [16, 42, 54, 68, 84, 118]. Several leukodystrophies, e.g., Alexander’s disease, Megalencephalic leukoencephalopathy with subcortical cysts and Vanishing White Matter disease, are caused by genetic variants in astrocyte-associated genes or involve astrocytic dysfunction as a pivotal pathogenic mechanism [65, 121]. Therefore, abnormal astrocytic phenotypes appear sufficient to cause leukodystrophy. On the other hand, astrocytic endocytosis may partially rescue the shortage of microglial debris clearance, which could be important for prolonging the maintenance of a healthy CNS. Hence, compensatory endocytosis by astrocytes could have both beneficial and detrimental effects on brain health, but future research is needed to further establish this.

The importance of understanding the effects of microglial depletion in the human brain is increasingly relevant as multiple clinical trials are focusing on CSF1R inhibitors to deplete microglia in brain disease. We focused our omics analysis on occipital gyrus tissue, where white matter is relatively spared and pathology is relatively mild, allowing us to detect moderate changes that correlate well with an early stage of disease. Tissue analyzed included both grey and white matter, to allow detection of possible changes in both grey and/or in white matter. Of note, since ALSP is a rare disease and high-quality post-mortem brain tissue is scarce, our multi-omics analyses were performed on brain tissue of two ALSP patients who underwent autopsy with a very short post-mortem delay. It is unclear whether our analyses on two patients fully represent the general ALSP population. Nevertheless, we could replicate findings from these datasets by IHC in multiple patients and in zebrafish models. We found relatively few changes by transcriptome and proteome analysis in ALSP brain tissue, including downregulation of microglial genes, an elevated stress/heat shock response and increased abundance of astrocytic proteins. In particular, glutathione S-transferases (GSTs), including GSTM1, were upregulated at the transcript and protein level. In mice, GSTM1 is secreted by astrocytes to stimulate a pro-inflammatory microglial response [51, 103]. As we find abundant GSTM1 expression in astrocyte-like cells in AD and FTD post-mortem brain tissue, astrocytic GSTM1 upregulation may be present across neurodegenerative diseases. We also found differential expression of ECM proteins in ALSP brain. Recent studies reported microglial ECM remodeling, including pruning of aggrecan and brevican, of which the latter is slightly increased in ALSP brain tissue [7, 23, 80, 112]. Notably, we cannot exclude whether these changes might in part reflect region-dependent effects rather than disease stage differences, and in fact others have observed regional microglial changes in ALSP [52]. In our proteome and transcriptome analysis, we cannot distinguish cortical or white matter-specific changes. Nevertheless, based on immunohistochemical staining on the same, relatively spared tissue, we did not detect differences that were exclusive to either cortex or white matter, except for increased expression of LAMP1 + in S100β + astrocytes only in cortical tissue, a more hypertrophic morphology of S100β + astrocytes in spared white matter and astrocytic engulfment of myelin in demyelinating lesions. This could indicate that astrocytes in white versus grey matter respond differently to the lack of microglia. Microglial depletion could influence white matter health by multiple possible independent effects, including possibly (indirectly) affecting astrocytic functioning through their compensatory endocytic response to elevated phagocytic demand. Our results here provide initial steps to further understand consequences of microglial depletion in the human brain.

In addition to having fewer microglia and altered radial astrocytes, heterozygous csf1ra^V614M/+ zebrafish also had fewer sox10:RFP + oligodendrocytes in the hindbrain, but normal myelination and normal locomotor activity, whereas homozygous mutants showed a smaller myelinated area and reduced locomotor behavior. Csf1r-/- mice also present with reduced oligodendrocyte cell numbers, although the overall myelin patterns in both Csf1r^E631K/+ mice and Csf1r-deficient mice and rats appear relatively normal [34, 91, 109].  Cunha et al. (2020) reported normal myelin content in the zebrafish csf1r-mutant spinal cord, as we observed in larvae, but an impaired remyelination capacity [26]. Thus, as mentioned above, there could be regional- and density-dependent effects of microglial depletion on myelin health in zebrafish. Overall, it is currently unknown to what extent the clinical manifestation of ALSP is preceded by cellular phenotypes as we observe in zebrafish.

Zebrafish as a neurobiology model allows exploring cells in embryonic development, in their natural environment by non-invasive in vivo examination. This has led to new fundamental insight in glial biology [11, 71, 78]. In addition, options for genetic targeting have improved, including editing of specific genetic variants identified in human disease as we show here. Possible limitations, when used as a model for human brain disease, include dissimilarities at a more macro-scale such as the lack of a layered cortex and a low ratio of myelinated versus non-myelinated brain tissue, which may obscure pathology in adult animals. Furthermore, similar to rodent models, phenotypes in zebrafish models for leukodystrophies are relatively mild. For example, unlike in human disease, fully CSF1R-deficient mutant zebrafish and rats are viable, and have a mild phenotype [62, 91]. Therefore, comparisons between multiple models are needed to make conceivable conclusions on disease mechanisms. Nevertheless, zebrafish in particular are suitable to detect phenotypes that go unnoticed in other models, as we show here, to subsequently explore further in other model systems and post-mortem brain tissue.

Based on phenotypes described for complete loss of function of CSF1R in humans, rats, mice and zebrafish, most peripheral macrophages are also depleted, leading to various developmental abnormalities outside of the CNS [11, 29, 34, 50, 62, 81, 91]. As yet, it is still unclear what the effects of pathogenic CSF1R variants are on peripheral macrophages and CNS-myeloid cells other than microglia in ALSP patients, although two studies reported on effects on peripheral monocytes/macrophages, including impaired phagocytosis and reduced presence of SLAN + macrophages [41, 47]. Skeletal abnormalities are obvious in patients with bi-allelic CSF1R variants, but no skeletal abnormalities have yet been noted in ALSP patients [58]. Nevertheless, it is intriguing to investigate extra-cerebral tissues and macrophages in ALSP, including possible involvement of skull bone marrow-derived myeloid cells [25].

Concluding, we provide proof that depletion of (homeostatic) microglia in ALSP patients occurs as a result of pathogenic missense variants in CSF1R and additional evidence of the importance of microglia in the development and maintenance of a healthy brain. Furthermore, our findings indirectly support the beneficial effect of HSCT in ALSP—which may act by repopulation of an under-occupied microglial niche. It remains an open question how loss of microglia can result in severe white matter defects. We speculate that microglial depletion and/or dysfunction results in a cascade of effects on other brain cells, including astrocytes, due to lack of total phagocytic capacity and the disruption of microglial modulation of the CNS environment, that by an as yet unknown mechanism eventually leads to white matter abnormalities [11]. A better understanding of this cascade would improve our knowledge of ALSP and potentially other disorders where loss of microglial function is implicated. Taken together, further investigating ALSP and the effect of CSF1R missense variants in vivo may provide opportunities for improving microglia-focused treatment strategies and insight into the consequences of depletion of microglia, and possibly other macrophages, for the human brain.