Hematopoietic stem cell transplantation chemotherapy causes microglia senescence and peripheral macrophage engraftment in the brain

Hematopoietic stem cell transplantation (HSCT) is a therapy used for multiple malignant and nonmalignant diseases, with chemotherapy used for pretransplantation myeloablation. The post-HSCT brain contains peripheral engrafted parenchymal macrophages, despite their absence in the normal brain, with the engraftment mechanism still undefined. Here we show that HSCT chemotherapy broadly disrupts mouse brain regenerative populations, including a permanent loss of adult neurogenesis. Microglial density was halved, causing microglial process expansion, coinciding with indicators of broad senescence. Although microglia expressed cell proliferation markers, they underwent cell cycle arrest in S phase with a majority expressing the senescence and antiapoptotic marker p21. In vivo single-cell tracking of microglia after recovery from chemical depletion showed loss of their regenerative capacity, subsequently replaced with donor macrophages. We propose that HSCT chemotherapy causes microglial senescence with a gradual decrease to a critical microglial density, providing a permissive niche for peripheral macrophage engraftment of the brain. Hematopoietic stem cell transplantation chemotherapy with busulfan causes senescence of brain microglia, rapid loss of adult neurogenesis and engraftment of peripheral donor macrophages into the brain of mice post transplantation.

S ince 2000, over 400,000 HSCTs have been performed in the United States as a therapy for multiple malignant and nonmalignant diseases 1 , with chemotherapy agents commonly used for myeloablation 2 . Chemotherapy kills recipient bone marrow cells before transplantation to provide donor stem cells with a permissive niche to engraft the recipient's bone marrow. HSCT is also a therapy used for leukodystrophies [3][4][5] but, since peripheral cells do not breach the blood-brain barrier (BBB) in the normal brain, it is unclear how HSCT could be therapeutic for these white matter disorders. Recent studies in both mice and humans have demonstrated that donor cells gradually accumulate in the post-HSCT brain as parenchymal macrophages, suggesting they provide a supportive role in central nervous system therapies 1, 6 .
It has been proposed that brain conditioning with either irradiation or chemical myeloablative procedures is required for peripheral cell brain engraftment. HSCT, in combination with total body irradiation while shielding the head, or using the BBB-impermeable myeloablative drug treosulfan, both failed to engraft donor cells into the brain [7][8][9] . When irradiation was conversely focused on the head, without myeloablation, this was sufficient to allow donor cell brain engraftment 10 . These and other studies suggest inflammatory signaling, or BBB changes due to irradiation or chemotherapy-induced brain conditioning, as the mechanism engrafting peripheral cells into the brain.
Recent studies,using genetic methods to deplete microglia, were able to engraft peripheral cells without performing irradiation or chemical myeloablation. Genetic ablation of microglia under the control of the CX3C chemokine receptor 1 (CX3CR1) promoter, followed with either intravenous monocyte injections, HSCT with head-shielded total body irradiation or parabiosis, caused donor cells to engraft as brain parenchymal macrophages 11,12 . These studies support a brain microglia-depletion, permissive niche mechanism, yet it remains unclear whether this occurs with chemotherapy-based HSCT.
Here, we focused on a clinical dose of busulfan chemotherapy myeloablation to determine how this HSCT regimen affects microglia and other brain regenerative populations, to determine the mechanism of peripheral cell brain engraftment.

HSCT donor cells become brain-resident macrophages.
To first determine the fate of donor cells in the post-HSCT brain, recipient wild-type C56BL/6j mice, or mice with expression of green fluorescent protein (GFP) in monocytes/macrophages/microglia (CX3CR1-GFP), were myeloablated with busulfan for 4 days (25 mg kg -1 day -1 intraperitoneally (i.p.; 100 mg kg -1 total dose)), a dose replicating clinical levels 13 , and transplanted with tdTomato (tdTom)-expressing, lineage-negative donor cells 1 day after the final busulfan dose (Fig. 1a). For the in vivo imaging cohort, a chronic cranial window was created over the somatosensory/ motor cortex (dorsal cortex) of CX3CR1-GFP mice 4 weeks before transplantation, and host microglia/macrophages (GFP + ) and donor-derived cells (tdTom + ) were imaged over multiple time points (Fig. 1a,b). Immunohistochemistry analysis was performed on the dorsal cortex for Iba-1, a microglia/monocyte/macrophage marker, and TMEM119, a marker specific to brain microglia, 24 weeks post HSCT (24 w ; Fig. 1c). TdTom + donor-derived cells first appeared in the brain parenchyma at 6 weeks post HSCT and increased significantly over the subsequent 18 weeks (Fig. 1d). All tdTom + donor cells were TMEM119 negative and most (~90%) were Iba-1 positive, as observed previously 6,11 , with the remaining tdTom + cells most probably myeloid-derived cells trapped in the vasculature (Extended Data Fig. 1a) 6,11 . There were regional differences in donor engraftment efficiency, with the olfactory bulb having the highest and the striatum the lowest donor cell density (Extended Data Fig. 1b).
In the first week post HSCT, in vivo two-photon imaging detected tdTom + dynamic meningeal phagocytes in the arachnoid space (Supplementary Video 1). The first tdTom + parenchyma donor cells were found in vivo at 8 weeks post HSCT (Supplementary Video 2). The size of donor macrophages did not significantly change over the observation period, and they maintained their radial resting morphology (Extended Data Fig. 1c,d). Single-cell three-dimensional (3D) tracking of 43 donor cells in the same brain volume showed their accumulation (Fig. 1e,f), with 40 cells persisting ( Fig. 1f lower plot, black circles) and three cells either dying or migrating out of the imaging volume ( Fig. 1f lower plot, red circles). Thus, with their continued presence, we conclude that most donor cells became resident.
Imaging every 10 min showed the macrophage protrusions to be highly dynamic ( Fig. 1g and Supplementary Video 3), persistently surveilling their environment with ~30% change (Fig. 1h), higher than control host microglia (~18% change). These in vivo results show macrophages becoming resident, with a resting morphology and surveilling their environment, possibly fulfilling a similar role to microglia despite having a significantly different transcriptome profile 6,11,14,15 .
Post-HSCT microglia partially deplete and enlarge. We next tracked host microglial changes in response to busulfan chemotherapy. Using the same experimental outline (Fig. 1a) host microglia were characterized, adding time points at 1 day and 1 week post HSCT for immunohistochemistry (tdTom − /Iba-1 + /TMEM119 + ; Fig. 1c). Host microglial density insignificantly declined after the final busulfan administration at 1 day to 1 week ( Fig. 2a). By 2 weeks post HSCT, a significant (~50%) loss of host microglial density was sustained up to 24 weeks. These results were confirmed in vivo with two-photon imaging of CX3CR1 − GFP HSCT mice (Extended Data Fig. 2a,b). Host-derived macrophages were also found in the brain parenchyma after HSCT due to incomplete donor chimerism, but this population was >100 times rarer than donor macrophages (Extended Data Fig. 2c,d), demonstrating that peripheral engraftment is not exclusive to donor cells.
Convex hull tracing of individual host microglial processes was performed (Fig. 2b, dashed lines) and, from 4 to 24 weeks post HSCT, their individual coverage area doubled (Fig. 2c). Complementary tiling between host microglia and donor macrophages was observed in vivo, with borders clearly delineated (Fig. 2d). There was an inverse correlation between host microglial density and cell area (Fig. 2e), indicating that microglia enlarge in response to a decrease in density. At 24 weeks post HSCT, higher donor macrophage density also correlated with decreased host microglial density, as expected with complementary tiling (Extended Data Fig. 2e). The higher presence of donor macrophages in the brain also correlated with increased host microglial area at 24 weeks (Fig. 2f).
Donor macrophage area remained the same between 12 and 24 weeks (Extended Data Fig. 2f), comparable to that of host-derived macrophages (Extended Data Fig. 2g). Since all individual donor cells were of the same area, while never having been exposed to busulfan, they probably proliferated to maintain brain tiling. In contrast, host microglial density decline, coupled with increased process coverage, could have been a result of microglial proliferation deficits due to busulfan 16,17 .
HSCT chemotherapy disrupts cell proliferation in the brain. Since busulfan chemotherapy causes crosslinking of DNA and has been shown to induce cell cycle arrest 18 , we first looked at the effect of HSCT on adult neurogenesis, a persistently mitotic cell population. In the hippocampal dentate gyrus, mitotic stem/progenitor cells-as demonstrated by the proliferation marker Ki67 in the subgranular zone-were observed in control mice (Fig. 3a), but at 1 day post busulfan there was a ~80% decrease. This declined to near-complete loss of proliferation, sustained to 24 weeks post HSCT (Fig. 3a,b), diminishing more rapidly than expected with normal age-related decline 19 (Fig. 3b, orange circles). The same loss occurred in subventricular zone (SVZ) Ki67 + stem/progenitor cells (Extended Data Fig. 3a,b). Labeling for doublecortin, an immature neuron/progenitor marker, showed ablation in regions of adult-born neuronal integration: the hippocampus and olfactory bulb (Extended Data Fig. 3c-g). Thus we demonstrate a chemotherapy agent having a permanent effect on extinguishing all adult neurogenesis. Interestingly, oligodendrocyte progenitor cells (NG2), another regenerating population, suffered near-complete loss of cell density at 2 and 4 weeks post HSCT but had begun to recover by 6 weeks post HSCT (Extended Data Fig. 4), demonstrating diverse effects of busulfan chemotherapy on different regenerating cell populations of the brain.
We next investigated the proliferation of donor macrophages and host microglia post HSCT. Donor macrophages showed an insignificant increase in Ki67 expression at 2-24 weeks post HSCT (Extended Data Fig. 5a). Host microglia showed a transient, Fig. 1 | HSCT causes the accumulation of brain parenchymal donor macrophages that become resident and surveillant. a, Experimental timeline for myeloablation with busulfan, HSCT of tdTom + /Lincells with histological and in vivo analysis at indicated time points. b, Coronal brain section illustration of approximate area for histological analysis (orange) and in vivo two-photon imaging through a chronic cover glass cranial window (green dashed line). c, Confocal images of immunohistochemistry at 24 weeks for Iba-1 and TMEM119 showing donor macrophages (tdTom + /Iba-1 + /TMEM119 -, closed arrowheads) and host microglia (tdTom − /Iba-1 + /TMEM119 + , open arrowheads). Scale bar, 20 μm. d, Plot of donor macrophage density in the dorsal cortex at 2, 4, 6, 12 and 24 weeks post HSCT (mean ± s.e.m.; one-way two-sided ANOVA with Tukey's multiple comparisons test; **P = 0.0078, 0.0078 and 0.0029 for 2, 4 and 6 weeks, respectively versus 24 weeks; n = 4 mice for 2 and 4 weeks, n = 6 mice for 6, 12 and 24 weeks). e, In vivo two-photon images of the same donor macrophages at 12, 16, 19 and 23 weeks post HSCT. Scale bar, 20 μm. f, Plot of accumulation of donor cells (top) and individual donor cell tracking over 25 weeks (bottom) in one mouse post HSCT with cells that persisted to the final imaging session (black circles/lines) and those that died or migrated out of imaging volume (red circles/lines). g, Three in vivo donor resident macrophages in brain parenchyma imaged every 10 min for 60 min, showing the dynamics of their processes. Last panel is an overlay of the first (0', red) and last timepoints (50', green) showing net dynamic changes. Scale bar, 10 μm. h, Average percentage change in process area during 10 min imaging interval between host microglia and donor macrophages, with significantly greater area change in donor macrophages (mean ± s.d.; unpaired two-sided t-test, ****P < 0.0001; n = 40 and 20 cells for host microglia and donor macrophages, respectively). NS, not significant. significant increase in Ki67 expression at 1 week post HSCT and a significant, gradual increase at 4 and 6 weeks post HSCT (Fig. 3c,d). Ki67 expression had dropped to control levels by 12 and 24 weeks, coinciding with the period of significant donor cell engraftment (Fig. 1d). These results were perplexing since no associated increase in host microglial density (Fig. 2a) concurred with the increased Ki67 expression period.
Using a rate model based on host microglial density (Extended Data Fig. 5b), the initial daily loss rate of microglia was ~5% during the first 2 weeks. Since density was maintained after 2 weeks, the host daily microglial death rate would have needed to increase to ~10% to compensate for increased Ki67 expression, assumed to indicate cell proliferation, thereby forming an almost perfect equilibrium between loss and proliferation to maintain microglial density (Extended Data Fig. 5b). Moreover, the modeled 10% death rate at 6 weeks would infer almost complete depletion of host microglia by 12 weeks with loss of Ki67 expression (Extended Data Fig. 5b, inferred density), which was not experimentally observed. We therefore investigated whether Ki67 expression in host microglia post HSCT truly indicated cell proliferation, since DNA crosslinking renders cells incapable of DNA synthesis and thus causing cell cycle arrest while expressing Ki67 (ref. 20 ). We explored the cell cycle with multiple markers (Fig. 3e) to determine at which stage microglia undergo arrest, using Ki67 and MCM-2 for G1 through G2, 5-ethynyl-2'-deoxyuridne (EdU) for S phase, proliferating cell nuclear antigen (PCNA) for S phase through G2, phospho-gamma-H2A.X (pγH2A.X) for DNA damage in S phase and phospho-histone H3 (pHH3) as a late G2 marker before mitosis 21 . At 6 weeks post HSCT, the peak of Ki67 expression, mice were injected with the DNA analog EdU 2 days before sacrifice and the ratio of EdU + /Ki67 + cells was measured to determine whether Ki67 expression indicated DNA synthesis 22 . In control mice, the ratio was measured in the SVZ (Fig. 3f) since the cortex was almost completely devoid of Ki67 + cells (Fig. 3d, ctrl) and an approximately fivefold greater number of EdU + cells was detected compared to Ki67 + cells (Fig. 3g). In mouse dorsal cortex 6 weeks post HSCT this ratio was the opposite, with around sixfold more Ki67 + cells than EdU + cells (Fig. 3f,g), demonstrating near-complete absence of EdU incorporation into host microglial DNA. Immunostaining for the DNA damage marker pγH2A.X showed the near absence of Ki67 colabeling in control, but it was significantly coexpressed at 6 weeks post HSCT in the dorsal cortex of host microglia (Fig. 3h,i). Both results strongly suggested cell cycle arrest 20 .
Ki67 + host microglia at 6 weeks post HSCT were almost completely coexpressed with MCM-2 ( Fig. 3j), which was insignificant in control SVZ Ki67 + cells (Fig. 3k), supporting the finding that cells had entered G1. Less than half of 6-week post-HSCT microglia coexpressed Ki67 with PCNA ( Fig. 3l), as opposed to near-complete coexpression in control SVZ Ki67 + cells (Fig. 3m), suggesting induction of S phase and possible arrest. Finally, to determine whether there was a transition to G2, Ki67 + and pHH3 coexpression was evaluated. Ki67 + control cells showed ~12% coexpression with pHH3 ( Fig. 3n,o) whereas no Ki67 + host microglia at 6 weeks post HSCT expressed pHH3 (Fig. 3o). Taken together, these results show microglial cell cycle arrest at S phase, probably due to busulfan-mediated DNA crosslinking 18 .
HSCT chemotherapy causes senescence in the brain. We next stained for senescence-associated beta-galactosidase (SA-β-gal), whose expression was prevalent at 2 and 4 weeks post HSCT across the dorsal cortex as compared to control, but decreased to control levels between 6 and 24 weeks (Fig. 4a,b). Since SA-β-gal is not an absolute marker of senescence 23 , we also stained for the nuclear membrane protein lamin B1, which is lost when cells undergo senescence 24 . We saw a significant loss at 2 weeks post HSCT, sustained up to 12 weeks across the dorsal cortex but insignificant at 24 weeks (Fig. 4c,d).
We next stained for the cyclin-dependent kinase inhibitor p21 (Fig. 4e), an indicator of cell cycle arrest and senescence that can result from DNA damage 25 . Microglial nuclear staining for p21 was observed at 6 weeks post HSCT (Fig. 4e) yet was absent in control microglia ( Fig. 4f). At 1, 4 and 6 weeks post HSCT, Iba-1 + host microglia showed significantly increased expression of p21 ( Fig. 4f). At these time points, the percentage of Iba-1 + combined expression with p21 and/or Ki67 was not significantly different ( Fig. 4g), with ~65% of host microglia being p21 + /Ki67 + , ~27% p21 − /Ki67 + and ~7% p21 + /Ki67 − . Taken together, these results suggests that host microglia commence cell division but undergo cell cycle arrest at S phase, activate p21 and become senescent. We therefore decided to test in vivo the extent of post-HSCT microglial senescence.
Post-HSCT microglia lose all regenerative capacity. To test the regenerative capacity of microglia post HSCT, the CSF1R antagonist PLX3397 was employed to deplete microglia and observe their ability to recover. PLX3397 was previously shown to cause near-complete loss of microglia within 4 weeks of administration and, after drug withdrawal, microglia recovered within 1 week, being exclusively repopulated with microglia with no peripheral cell brain engraftment 26 . CX3CR1-GFP mice with a cranial window were utilized for in vivo two-photon imaging with a control group receiving neither busulfan nor donor cells, and a HSCT group receiving busulfan and transplanted with tdTom + lineagedonor cells (Fig. 5a). Both groups were given a PLX3397-containing diet for 4 weeks to deplete microglia, followed by drug withdrawal (normal diet) for recovery. Throughout the experiment the same brain volume for each mouse was two-photon imaged over multiple time points (Fig. 5a,b).
Mice were imaged 4 weeks after cranial window surgery (-7 w) for baseline microglial density and again 3 weeks later (-4 w), immediately before PLX3397 administration. Control microglial density was sustained, but HSCT mice showed a significant decrease due to busulfan treatment (-4 w; Fig. 5c), as in our previous results (Extended Data Fig. 2b). After 4 weeks of PLX3397 treatment, both groups showed significant microglial density loss (0 days; Fig. 5c). The control group recovered within 4 days after PLX3397 withdrawal due to proliferation of host microglia (Fig. 5c), as shown previously 27 . With HSCT there was no recovery of host microglia, even up to 9 days after PLX3397 withdrawal. Moreover, 3D single-cell tracking of all microglia within individual brain volumes demonstrated no cell divisions out of 481 tracked cells in six mice (Extended Data Fig. 6).
At 5 days after PLX3397 recovery in the HSCT group, donor macrophages were first observed in the brain with varying rates of cell engraftment between individual mice (Fig. 5d). Of interest, mouse 2 had the lowest microglial depletion after PLX3397 treatment (microglial density of ~200 cells mm -2 ; Fig. 5d) and no donor cells were observed engrafting within the imaging window, even at 19 days post PLX where, in contrast, mouse 1 (microglial density of ~50 cells mm -2 ; Fig. 5d) showed almost complete donor engraftment (Extended Data Fig. 7a). This suggests that a minimal host microglial density threshold is required for successful donor engraftment. Despite this greater host microglial density in mouse 2, none of  Time-lapse 10-min-interval in vivo imaging showed a period of massive proliferation of donor macrophages within the parenchyma at 6 days after PLX3397 withdrawal in mice with donor engrafted cells ( Fig. 5e and Supplementary Video 4). Cells migrated, condensed to a mitotic sphere, developed a cleavage plane, extended their processes and separated a daughter cell (Fig. 5e). There were ~33 proliferating cells mm -3 per 10 min (Fig. 5f), peaking to ~115 cells mm -3 per 10 min depending on the time period imaged for each mouse, with 7 days post PLX showing the highest level of proliferation (Extended Data Fig. 7b). Of importance, no host Utilizing the tdTom cell signal in the bloodstream, blood vessels were reconstructed (Supplementary Video 4) and we searched in 3D for transmigration events of tdTom donor cells into the brain parenchyma, but were unable to find any crossings despite tracking in six mice consisting of 27 imaging sessions (10-min intervals) for a total of 28 h in a combined volume of ~0.5 mm 3 . The migration front of donor cells was also detected in one mouse at 6 days post PLX, moving at ~40 µm h -1 (Extended Data Fig. 7c and Supplementary Video 5) and, by the next day, donor cells filled the imaging field (Extended Data Fig. 7d). At 21 weeks post HSCT, with PLX depletion and recovery, ~91% of the brain was replaced with donor cells, even within white matter, showing the accelerated depletion of host microglia with PLX to be effective in enhancing donor macrophage engraftment 15 (Extended Data Fig. 8a). Interestingly, the striatum was consistently less engrafted, similar to normal post-HSCT engraftment density (Extended Data Fig. 1b), suggesting that the natural microglia turnover rate, which is variable among brain regions 28 , probably correlates with engraftment efficiency.
The surveillant process dynamics of donor macrophages in post-PLX HSCT brains were not significantly different from the dynamics of control microglia (Extended Data Fig. 8b), yet post-PLX, HSCT-treated sparse microglia showed decreased dynamics (Extended Data Fig. 8b). There was also no significant difference in neuron density in layers 2/3 between HSCT brains 15 weeks post PLX3397 recovery and control (Extended Data Fig. 8c). Combined, these results indicate that dense donor macrophages may fulfill the role of host microglia and do not have a major detrimental effect, supporting a previous study that saw no effect on behavior with macrophage engraftment 11 . Summarizing these results, busulfan chemotherapy HSCT causes complete, irreversible loss of adult neurogenesis, loss of lamin B1 and depletion of half of the host microglial population within 2 weeks post HSCT, which corresponds to increased individual host microglial area, a transient period of Ki67 ramped expression that arrests in parallel with the gradual engraftment of donor macrophages (Fig. 6a). We propose a depletion model that allows the permissive engraftment of peripheral macrophages in response to the partial loss and complete senescence of host microglia (Fig. 6b). Under normal conditions, microglia have a low turnover rate 29,30 with exclusive self-renewal (Fig. 6b, before HSCT). HSCT busulfan chemotherapy depleted the microglial population by 50%, and busulfan-induced DNA damage caused cells to become senescent (Fig. 6b, 2 w post HSCT). Six weeks post HSCT, microglial processes expanded to maintain brain tiling or, due to senescence-induced cell enlargement 31 , undergoing cell cycle arrest (Ki67 + /p21 + ; Fig. 6b, 6 w post HSCT). A minimal microglial density threshold was reached and the permissive transmigration of peripheral cells ensued, causing their division to re-establish brain tiling (Fig. 6b, 12 w post HSCT), and potentially trophic (interleukin 34) 32 or contact feedback mechanisms prevent host microglia from further initiating cell division.

discussion
We show that busulfan chemotherapy causes pervasive host microglial senescence and cell cycle arrest, exhausting their regenerative capacity and resulting in brain engraftment of peripheral macrophages, which become resident. Post-HSCT microglia also expand their processes to potentially maintain brain tiling, which is not in accordance with chronic brain inflammation as the engraftment mechanism where an ameboid/condensed microglial morphology would be expected 33 . A previous study demonstrated that parabiosis of a GFP mouse to an inducible CX3CR1-CSF1R-KO mouse, followed by depletion of microglia specifically in the knockout (KO)   mouse, was sufficient to cause GFP macrophages to engraft the KO brain without the induction of global inflammation 11 . Instead, the brain engraftment mechanism appears to have the similar goal of HSCT myeloablation, where suppression of host hematopoietic stem cells provides a niche for donor cells to engraft the bone marrow. Here we propose that busulfan chemotherapy stops microglial regeneration, thereby causing cell loss that then reaches a critical density, providing a permissive niche for the engraftment of donor macrophages into the brain. Brain conditioning by either irradiation or chemotherapy was proposed to cause inflammation and BBB disruption and thereby being essential for donor cell engraftment, which was based on head protection during irradiation and a BBB-impermeable chemotherapy agent treosulfan showing no engrafted donor-derived macrophages in the central nervous system after HSCT [7][8][9] . We instead propose that both these manipulations merely protected the host microglia from irradiation or chemotherapy damage, and thus the population was not depleted/senescent, which consequently deprived donor cells of a permissive niche to engraft the brain. Additionally, busulfan treatment (80 mg kg -1 ) had no effect on BBB integrity 34 , and genetic microglial depletion with transplantation resulted in no change in BBB integrity during macrophage brain engraftment 12 . Taken together, our findings suggest a non-inflammation-or non-BBB-disruption-induced permissive engraftment process, but further direct investigation would be needed.
Our observation that host microglia become senescent after HSCT is confounding, since many cells expressed the commonly used proliferation marker Ki67. Ki67 protein is expressed at the start of G1, being maintained through all subsequent phases of the cell cycle 35 . Nevertheless, Ki67 expression was observed after DNA damage in G 1 /S-phase arrested cells 20 . We showed near-complete loss of DNA incorporation in post HSCT microglia, expression of the DNA damage marker pγH2A.X, coexpression of Ki67 with MCM-2 and partially with PCNA, expression of the senescence marker p21 and a complete lack of coexpression with pHH3, a G2 marker. Taken together, we conclude that microglia initiate cell division but that busulfan induces DNA crosslinking 18,36 , causing them to become incapable of DNA synthesis and thereby arresting the cell cycle at S phase and becoming senescent via the p21 pathway. Busulfan also causes hematopoetic stem cell senescence without undergoing apoptosis 16,36 , and a similar mechanism may occur with microglia 37 . Interestingly, we and others observed a 50% loss of microglia within the first 2 weeks 38 , and it would be intriguing to compare the lost and surviving populations. 27 A similar microglial senescence mechanism may occur with full-body-irradiation myeloablation, since this induces hematopoietic stem cell senescence 36 . Further study on the effect of irradiation on microglial senescence would be important to pursue.
We also discovered busulfan as a chemotherapy agent that rapidly and irreversibly arrests all adult neurogenesis. One day after the 4-day busulfan regimen, almost all cell proliferation was absent in the SVZ and dentate gyrus, accompanied by loss of doublecortin progenitor/immature neurons in the olfactory bulb and hippocampus, persisting up to 24 weeks post HSCT. Previous studies ablated adult neurogenesis to unmask its function in utilizing irradiation 39 and chemotherapy agents, including b-arabinofuranoside 40 and temozolomide 41 , yet residual neurogenesis still occurred or recovered with these methods. Clinically, it would be intriguing to determine whether patients who received busulfan chemotherapy lack adult neurogenesis and, moreover, whether this could explain the cognitive deficits associated with HSCT 42 .
We also observed a transient loss of NG2 oligodendrocyte precursor cells that, unlike adult neurogenesis and microglia, gradually recovered. It would be interesting to pursue what mechanism protects the NG2 population and whether busulfan treatment causes white matter disruption. This would be particularly important for HSCT treatments in the case of leukodystrophies, since donor cells have been detected in clinical post-mortem brain and spinal cord tissue, and a similar mechanism probably exists in humans 6,43 .
We demonstrated in vivo the mechanism of peripheral cell brain engraftment undergoing presumably rare transmigration events followed by massive proliferation and parenchymal migration, albeit after a microglial chemical depletion and recovery protocol. Our inability to observe peripheral transmigration events after recovery from microglial depletion suggests that BBB-crossing events are rare. Furthermore, visualization of migration fronts supported this finding, and future studies with clonal analysis could potentially determine transmigration frequency.
Numerous studies have compared the transcription profile between brain parenchymal macrophages and microglia, with about 90% of transcripts having similar expression 6 , yet significant differences were found in the remaining transcripts 6,11,14,15 . When microglia isolated from a donor brain were injected into the hippocampus of post PLX brains, their transcriptome was highly similar to control microglia, suggesting that PLX depletion and recovery had not drastically changed the niche 15 . In contrast, post-HSCT macrophages maintained a more macrophage-like, rather than microglia-like, transcriptome, showing that even near-complete macrophage replacement in the brain environment is not sufficient to change their transcriptome 15 . Although we observed no obvious detrimental effect on cortical neurons with macrophage replacement of microglia, it would be important to determine whether parenchymal macrophages can fulfill the role of microglia, with considerable relevance for future cell therapy strategies.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41591-022-01691-9.
In vivo two-photon Imaging. For in vivo two-photon imaging studies, cranial window surgery was performed no later than 4 weeks before transplantation, as previously described 46 . The animals were anesthetized (100 mg kg -1 ketamine/10 mg kg -1 xylazine) and a 5-mm cranial window created over the right dorsal cortex between bregma and lambda sutures. Dental cement was used to seal both the coverglass and a stainless steel head bar to the skull 47 . For each imaging session, mice were lightly anesthetized with isoflurane (0.8%) and the metal head bar was attached to a custom stage. Anesthesia continued throughout imaging, and body temperature was maintained with a rectal thermometer feedback heating pad. A fiducial point of reference on the stage was used to zero x-y-z positions to find the same brain volume over multiple sessions. Imaging was performed on an Investigator microscope (Prairie View/Bruker) with a ×16/0.8 numerical aperture objective (Olympus) and a Ti:Sapphire femtosecond pulsed two-photon laser (DeepSee, Spectra Physics) tuned to 1,000 nm. Multiple Z-plane stack images were acquired at a 2-3-μm z-step at a resolution of either 0.82 × 0.82 2 or 0.41 × 0.41 μm 2 per pixel X-Y.

Immunohistochemistry.
A cohort of animals at 6 weeks post HSCT was injected with EdU (Sigma-Aldrich, no. 900584; 50 mg kg -1 i.p.) 2 days before sacrifice. Animals were anesthetized with a 2× dose of ketamine/xylazine and transcardially perfused with heparinized saline followed by 4% paraformaldehyde. Tissues were dissected and post-fixed in paraformaldehyde for 2-4 h, followed by 24 h in 30% sucrose in PBS. Tissues were microtome sectioned at 40 μm and stored at -20 °C in antifreeze solution. For immunolabeling, brain sections were washed in PBS, underwent antigen retrieval (30 min at 80 °C; citrate buffer pH 6.0), washed in PBS and incubated in various combinations of the following primary antibodies: Iba-1 (1:500, guinea pig; Synaptic Systems, no. 234004), TMEM119 (1:250, rabbit; Abcam, no. ab209064), Ki67 (1:250, rat; Invitrogen, no. 14-5698-80), doublecortin (1:500, guinea pig; Millipore, no. AB2253), NG2 (1:400, rabbit; Chemicon, no. AB5320), phospho-gamma-H2A.X (1:1,000, rabbit,; Abcam, no. ab2893), MCM-2 (1:250, mouse; BD Biosciences, no. 610701), PCNA (1:250, rabbit; Abcam, no. EPR3821), pHH3 (1:500, rabbit; Invitrogen, no. PA5-104936), lamin B1 (1:500, rabbit; Abcam, no. AB16048), NeuN (1:500, mouse; Millipore, no. MAB377) and p21 (1:250, rabbit; Abcam, no. AB188224). Primary antibody was diluted in PBS ++ (PBS, 5% normal donkey serum, 0.25% triton) and incubated overnight at 4 °C. Sections were then washed in PBS and incubated for 2 h at room temperature with combinations of the following secondary antibodies in PBS ++ : donkey anti-guinea pig Alexa Fluor 648 (1:500; Jackson Immno Research, no. 706-605-148), goat anti-rabbit Alexa Fluor 568 (1:500; Invitrogen, no. A11036), goat anti-rabbit Alexa Fluor 488 (1:500; Invitrogen, no. A11034), goat anti-rabbit Alexa Fluor 647 (1:500; Invitrogen, no. A21244), goat anti-rat Alexa Fluor 568 (1:500; Invitrogen, no. A11077) and goat anti-mouse Alexa Fluor 586 (1:500; Invitrogen, no. A11031). Sections were then washed in PBS. EdU was labeled with an Alexa Fluor 488 Click-iT cell proliferation kit following the manufacturer's protocol (ThermoFisher, no. C10337). Fluorescent sections were mounted, coverslipped and confocal imaged (Zeiss LSM 700, Zen Black software) at a resolution of 0.63 × 0.63 μm 2 per pixel for image analysis. Brain sections were incubated at 37 °C overnight for SA-β-gal staining (Cell biolabs, no. CBA-230) using the manufacturer's protocol. Sections were mounted, dehydrated and coverslipped, with brightfield images acquired using a Zeiss Observer Z1 microscope with a Hamamatsu ORCA-ER camera at a resolution of 0.65 × 0.65 μm 2 per pixel. Image analysis. For in vivo two-photon data, 3D registration was performed with 3D slicer software v.4.10.2 (www.slicer.org) and/or Imaris software (Bitplane, v.9.1.2). Images were normalized across time points and z-planes using Matlab 2018a (Mathworks) code 48 . Donor macrophage 3D surface renderings were manually formed using the magic wand tool in the surface toolkit. For the PLX3397 depletion and recovery time series, supervised spot analysis in Imaris was used for tracking of individual cells across days. Cells that were within 40 μm of the x, y or z volume edges were excluded from analysis, to eliminate migration-or registration cropping-associated loss/gain artifacts. Blood vessel reconstruction took advantage of the tdTom signal in circulating blood cells. A simple horizontal spatial filter was applied to images due to the microscope x-scan speed being greater than on the y axis, and therefore cells in the bloodstream were x-distorted from their rapid trajectory. Multiple timepoints of extracted cells from time-lapse imaging were registered and summed to reconstruct the intraluminal vasculature structure. To search for peripheral cell transmigration events, in Imaris software using 3D glasses (Nvidia 3dVision) cells were followed within the bloodstream in a 10-min-interval image series, searching for their transmigration into the parenchyma (each imaging session was between four and ten volumes). Quantification of host microglia and donor macrophage confocal images was performed with Fiji ImageJ (imagej.net/Fiji, NIH, v.1.53) software. Confocal images were indexed with a number upon acquisition, with the observer performing subsequent quantification blind to the image reference. Images were cropped to exclude large blood vessels and cortical edges. The image was split to show overlays of all fluorescent channels separately, and the 'synchronize windows' tool was used to show the pointer location in all images. Using the 'multi-point' tool, cell types were manually counted based on the combination of antibody markers. For each brain section, convex hull circumference was manually traced and measured for five random cells using the 'polygon selection' tool. Donor macrophage and host microglial process motility was recorded in vivo by acquisition of image stacks every 10 min. In Imaris, a 70-μm projection was made for each field of view (FOV) at each time point, the series was normalized and registered ('image stabilizer';https://www.cs.cmu.edu/~kangli/code/Image_ Stabilizer.html) and five cells from each FOV with complete structure underwent convex hull tracing of their process area. In Matlab, a contiguous spatial filter of ten pixels was applied and iterative subtraction of images within the series ((n + 1) -n) was performed for detection of motility. The subtraction product images were binarized, summed and divided by the area (pixels) of the complete cell projection to calculate percentage change per 10 min.
Statistical analysis. Plots and statistics were performed with Prism software v.9 (www.graphpad.com). A Shapiro-Wilk normality test was first performed and, for normal data, one-way two-sided analysis of variance (ANOVA) with Tukey's multiple comparisons test was performed; otherwise, for non-normal data, a two-sided Kruskal-Wallis or Kolmogorov-Smirnov test was performed. For microglial process area x-y plots, a simple linear regression was performed with the extra sum-of-squared F-test to determine whether the fit was significant from the null hypothesis of slope = 0.
Microglial Ki67 density age-decline neurogenesis modeling. Statistical analysis and modeling were performed in Python, and goodness of fit was assessed with R 2 and root mean square error (RMSE). Since Ki67 is expressed for ~32 h in microglia 30 , microglial proliferation rate was fit over time based on the mean of measured data for each time point. Spline interpolation of order 3 was used, meaning that the first-order derivative is null on the first and sixth points (weeks 0 and 12), to ensure that the proliferation rate remained positive (R² = 0.74 and RMSE = 2.64). Microglial density (cells mm -2 ) was fit over the first five timepoints, until week 6. It is assumed that microglia divide in two separate populations: N 1 , the resident microglia and N 2 , newly generated microglia. A distinct rate model was used to fit each population evolution, the rate of death being λ 1 and λ 2 for N 1 and N 2 , respectively; µ(t), the proliferation rate for N 2 , was interpolated. N 1 : dN1 dt (t) = −λ1N1(t) leading to N1(t) = N 0 1 × e −λ1t . N 2 : dN2 dt (t) = μ(t) − λ2N2(t), which was solved iteratively using Euler's method, with a time step (dt) of 1 day: N2(t + dt) = N2(t) × (1 − λ2)dt + μ(t) × dt. To fit total microglial density N(t) = N 1 (t) + N 2 (t), we optimized successively λ 1 and λ 2 using the least-squares method, obtaining λ 1 = 5.1%/day, λ 2 = 10.8%/day and N 0 1 = 281.8 (R 2 = 0.84 and RMSE = 27.46). To fit our control neurogenesis Ki67 + (Fig. 3b) and DCX + (Extended Data Fig. 3e) cell quantification to previously published, age-related decline results 19 , data were fitted to a distinct rate model with N being the population and λ the corresponding monthly death rate: dN dt (t) = −λN(t) leading to N(t) = N 0 × e −λt . For Ki67 + cells, λ = 39.7%/day and N 0 = 8,250 (R 2 = 0.91 and RMSE = 642); and for DCX + cells, λ = 33.7%/day and N 0 = 28,566 (R 2 = 0.89 and RMSE = 2,435).
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.