Leukocyte dynamics after intracerebral hemorrhage in a living patient reveal rapid adaptations to tissue milieu

Intracerebral hemorrhage (ICH) is a devastating form of stroke with a high mortality rate and few treatment options. Discovery of therapeutic interventions has been slow given the challenges associated with studying acute injury in the human brain. Inflammation induced by exposure of brain tissue to blood appears to be a major part of brain tissue injury. Here, we longitudinally profiled blood and cerebral hematoma effluent from a patient enrolled in the Minimally Invasive Surgery with Thrombolysis in Intracerebral Hemorrhage Evacuation trial, offering a rare window into the local and systemic immune responses to acute brain injury. Using single-cell RNA-Seq (scRNA-Seq), this is the first report to our knowledge that characterized the local cellular response during ICH in the brain of a living patient at single-cell resolution. Our analysis revealed shifts in the activation states of myeloid and T cells in the brain over time, suggesting that leukocyte responses are dynamically reshaped by the hematoma microenvironment. Interestingly, the patient had an asymptomatic rebleed that our transcriptional data indicated occurred prior to detection by CT scan. This case highlights the rapid immune dynamics in the brain after ICH and suggests that sensitive methods such as scRNA-Seq would enable greater understanding of complex intracerebral events.


INTRODUCTION
Intracerebral hemorrhage (ICH), brain bleeding often caused by hypertension, has a worldwide incidence of over 3 million cases per year, limited treatment options, and high mortality rate (40-60% worldwide). 1,2 To date, surgical approaches to remove the hemorrhage have not improved functional outcomes. 3,4 The Phase III clinical trial of Minimally Invasive Surgery with Thrombolysis in Intracerebral Hemorrhage Evacuation (MISTIE III) was designed to test whether minimally invasive catheter evacuation followed by thrombolysis would improve functional outcome in patients with ICH. The trial's approach used image-guided minimally invasive surgery to aspirate the liquid component of the hemorrhage followed by placement of a sutured indwelling catheter to liquify and drain remaining hemorrhage over several days with the aid of recombinant tissue plasminogen activator (tPA). The trial found that surgery was safe and reduced mortality, but did not improve overall functional outcomes 365 days after ICH. 5 As part of a trial sub-study (ICHseq), daily samples of both hematoma effluent from the catheter and peripheral blood were collected. These samples provided a unique opportunity to study the immune response to ICH over time in living patients.
Here, we describe longitudinal cellular characterization of hematoma effluent and blood from a patient who was enrolled in MISTIE III using single-cell RNA-sequencing (scRNA-seq). 6 We define distinct leukocyte phenotypes that may dynamically reflect local inflammatory responses in the brain as well as the emergence of cells resembling peripheral blood leukocytes that we propose identify the onset of repeat tissue exposure to blood (in this case, an asymptomatic re-bleeding event) later detected by CT scan. Additionally, we show that leukocyte transcriptional programs in the patient's blood return to a baseline comparable to age-matched control blood 2.5 years after ICH. This case report is the first longitudinal single-cell genomic characterization of cellular infiltrate in acute brain injury in a living patient and, importantly, underscores the utility of single-cell analyses for understanding dynamic immune responses in the brain and other tissues.

RESULTS AND DISCUSSION
A 74-year old female with hypertension presented to the emergency room with difficulty speaking and right-sided weakness (Supplemental Table 1). Her initial National Institutes of Health Stroke Scale Score (NIHSS) was a 19, and symptoms included aphasia and right-sided hemiplegia, sensory loss, and visual field deficit. A CT scan of the brain showed a large left ICH. Two days after ICH onset, she was enrolled into the MISTIE III trial and randomized to the surgical ! arm. At 51 hours post-hemorrhage, 45 mL of liquid hematoma were aspirated and the drainage catheter was placed. The following day, instillation of tPA (1 mg every 8 hours) was initiated per trial protocol with subsequent effective drainage of the remaining hematoma ( Figure 1A). Beginning on day three posthemorrhage, an increasing volume of drainage was noted from the catheter and tPA was discontinued. The patient consistently improved neurologically. The catheter continued to drain the hematoma for two more days before being removed. The patient was subsequently discharged to an acute rehabilitation facility. Thirty days after ICH onset, she had improved to mild aphasia (difficulty speaking), right visual field deficit, and mild right hemiparesis (weakness) and continued to recover functionally. At 2.5 years after onset, she was active with minimal aphasia and right-sided vision loss (NIHSS 3).
Overall, during the patient's hospital stay, her hematoma was effectively drained ( Figure 1A) and she improved clinically. Midway through her treatment increasing hematoma volume by CT scan and an increase in hematoma drainage suggested that she experienced an asymptomatic re-bleeding event and treatment with tPA was halted ( Figure 1A, Supplemental Table 1).
Volumetric quantification by serial CT scans suggests that this event occurred between 93 and 105 hours after the onset of the hemorrhage. Increases to the volume and cellularity of hematoma drainage, however, were observed prior to hematoma expansion as measured by imaging, beginning to rise by 73 hours after onset (Figure 1B, Supplemental Table 1). These data suggest that the patient's re-bleeding event occurred prior to the increased volume seen on CT scan.
We hypothesized that detailed analysis of the cellular composition of the hematoma effluent in this patient would provide insight into the molecular features of the hematoma microenvironment, as well as her re-bleeding event, and determine whether immune responses are brain tissue-specific. Using Seq-Well, we performed scRNA-seq on individual hematoma cells and peripheral blood collected longitudinally during her hospitalization as well as peripheral blood collected after 2.5 years, and control blood from age-matched healthy donors ( Figure 1C). 6 We generated single-cell transcriptional profiles for 31,722 cells spanning seven time points after onset, and these are displayed along with control samples using a t-distributed stochastic neighbor embedding (tSNE) ( Figure 1D, Supplemental Figure 1 and 2) along the first two components.
Replicates (performed for most time points) clustered together across each time point and have comparable quality control metrics (Supplemental Figure 1).
Overall, we found that blood and hematoma-derived cells were transcriptionally distinct at almost all time points, and separated in tSNE space by compartment, suggesting that the local milieu within the hematoma is a major driver of leukocyte transcriptional state and function after ICH.   Table 2). Our mass cytometry results further corroborate our scRNA-seq frequencies, showing that T cells and monocytes were the predominant cell types profiled in both blood and hematoma. Finally, in our scRNA-seq data, we found a large cluster of neurons that appeared exclusively in hematoma effluent at high frequency prior to increased hematoma drainage at 66 hours ( Figure 1D and E, Supplemental Table 2). These neurons were the sole cluster positive for KIF5A, a neuron-specific kinesin subunit (Supplemental Figure 3C). 7,8 To assess whether the leukocyte populations identified in the peripheral blood after ICH differed from baseline, we also generated single-cell transcriptional profiles from the patient's blood collected 2.5 years post-ICH and from the blood of four age-matched healthy donors (Figure 1D and E). The ! 6 follow-up blood and healthy control blood generally overlap, and may suggest an overall return to baseline in circulating leukocytes in the blood of this patient (Supplemental Figure 2). Bulk RNA-seq analysis on sorted populations of immune cells, including CD4 + T cells and monocytes (Supplemental Figure 5 A and B), from this patient's blood and hematoma at several time points (Supplemental Table 1), combined with hallmark gene set enrichment analysis (Supplemental Figure 5C), corroborates the substantial transcriptomic variation over time despite little variability in broad immune cell frequencies assessed by flow cytometry (data not shown).
In the MISTIE III trial, 2% of patients in the surgical arm had a symptomatic re-bleeding event, however 32% had asymptomatic re-bleeding. 5,9 Hematoma effluent volume drainage is an imperfect measurement of rebleeding, as some hematoma cavities have variable connections with the ventricular or subarachnoid space and, at times, volume may increase due to CSF drainage. This event is difficult to identify or predict at the bedside. The implications of rebleeding on the local immune response and long-term outcomes remain unknown but have an important impact of treatment (i.e., cessation of tPA). We hypothesized that a deeper analysis of the predominant cell types in our scRNAseq data, including T cells and myeloid cells, could better delineate the compartment-specific response to an asymptomatic re-bleed. To accomplish this, we performed sub-clustering analyses on each population separately (Supplemental Methods).
Three of these (myeloid sub-clusters 2, 4, and 5) were almost exclusively found in hematoma effluent ( Figure 2B, Supplemental Figure 6B). We found that the frequency of myeloid sub-clusters shifted over time, with several appearing in either blood or hematoma around the time of re-bleeding ( Figure 2B).
Functional gene enrichment analyses on myeloid sub-cluster defininggenes revealed distinct pathways that were active in each ( Figure 2C, remaining clusters in Supplemental Figure 7 and Supplemental Table 6). Blood prior to re-bleeding was predominantly comprised of myeloid sub-cluster 3 at 51 hours (enriched for granzyme signaling and glucocorticoid signaling pathways) and myeloid sub-cluster 12 at 66 hours (enriched for iron homeostasis and several T cell interaction pathways). Blood during the re-bleed as detected by catheter drainage (73-89 hours) was predominantly comprised of myeloid sub-cluster 1, enriched for LXR/RXR activation and IL-10 signaling, and then by myeloid subcluster 10, enriched for interferon signaling, at 112 and 137 hours. Sub-clusters unique to control and the patient's long-term follow-up blood (myeloid subclusters 0 and 13, Supplemental Figure 6B) were enriched for autophagy.
Overall, sub-clusters predominantly in blood were enriched for a spectrum of pathways involved in innate immune cell functions that changed over time, including TLR signaling, glucocorticoid signaling, EIF2 signaling, and interferon signaling.
Hematoma prior to re-bleeding was predominantly comprised of myeloid sub-cluster 4, which was enriched for glycolysis and HIF-1α signaling. We only recovered 12 myeloid cells at 66 hours in hematoma, limiting our resolution at this time point. There was then a time-dependent shift from dominance by myeloid sub-cluster 2 to myeloid sub-cluster 5. Myeloid sub-cluster 2 was enriched for pathways related to myeloid cell extravasation and tissue infiltration, possibly indicating new infiltration after rebleeding, ( Figure 2C) and myeloid subcluster 5 was enriched for pathways related to tissue repair and senescence including phagosome maturation, FGF signaling, autophagy, and senescence. This shift is consistent with a transition between dominance of pro-inflammatory activation and recruitment to phagocytosis and perhaps anti-inflammatory functions. Taken together, we found alterations in the overall transcriptional states of myeloid cells in both blood and hematoma, and gene set enrichments suggest these cells, particularly in hematoma, may have differing functions altered by their unique physiological niche that shift over time.
To better contextualize the phenotypes of myeloid sub-clusters, we performed module scoring against a list of curated human monocyte/macrophage gene sets (Supplemental Table 6, Supplemental Figure 8A). [11][12][13][14] We performed clustering and principal component analysis (PCA) on these scores to identify modules that explain differences across myeloid clusters. We found that most of these known signatures did not explain differences between myeloid subclusters (Supplemental Figure 8A). Nonetheless, we identified four gene signatures that predominantly explain differences between myeloid sub-clusters that were either hematoma or blood in origin, well-describing blood monocytes/macrophages but not those from hematoma (Supplemental Figure   8B). Specifically, we found that the predominantly blood sub-clusters (myeloid sub-clusters 0, 1, 3, 6, 7-13) and the sub-cluster that emerged in the hematoma at 73 hours (myeloid sub-cluster 2) scored similarly for a CD14 + monocyte gene signature identified recently in blood 12 (Figure 2D, Supplemental Figure 9A).
Taken together, our data suggest that myeloid sub-cluster 2 may represent monocytes newly entered into the hematoma due to a recent re-bleed, while other hematoma clusters, like myeloid clusters 4 and 5, are defined by activation states induced by local signals in the hematoma and do not resemble clusters previously defined in human blood or by in vitro stimulations. [12][13][14] When considered in the context of the timing of the increased draining volume from the catheter and the emergence of a neuron-like cluster at 66hrs, our data may suggest a re-bleeding event prior to the clinically identified changes in hematoma volume detected by a daily CT scan.
Our data also revealed substantial involvement of T cells in the hematoma at all time points. Sub-clustering over all hematoma and peripheral blood T cells from the patient and control blood resulted in 12 T cell sub-clusters that also separated overall by compartment ( Figure 3A, Supplemental Figure 9). As with monocytes, we found that T cell frequencies were dynamic over the course of ICH in hematoma (Supplemental Table 7). In the blood, there was an initial shift in the dominant cluster from T cell sub-cluster 3 to T cell sub-cluster 2 prior to the re-bleed, which then remained highly frequent for the remaining time points ( Figure 3B). Several sub-clusters were also predominantly blood in origin (subclusters 0, 2, 4, 10, and 11). T cell sub-clusters 2 and 3, which were enriched for IL-17A and GADD45 signaling, respectively, also emerged in the hematoma at 73 hours. The emergence of these clusters in hematoma at this time paralleled the emergence of a more blood-like monocyte sub-cluster at the same time point, further supporting evidence this was the beginning of the rebleeding.
Several sub-clusters of T cells were predominantly found in hematoma (Supplemental Figure 9B) (e.g., T cell sub-clusters 1, 3, 5, 6, and 8). T cell subcluster 5, prevalent in early hematoma samples, was enriched for heat shock proteins, protein ubiquitination, and eNOS signaling ( Figure 3D, Supplemental Table 8). T cell sub-cluster 3, which emerged at 73 hours, was enriched in GADD45 signaling and several immune activation pathways ( Figure 3C, additional clusters in Supplemental Figure 10). T cell sub-cluster 1, emerging at later time points in hematoma, shows expression of genes involved in transcriptional regulation, anti-proliferation, and iron homeostasis. Interestingly, we did not find clear patterns of chemokine and receptor expression 15 in specific clusters in either myeloid or T cell sub-clusters; we did find, however, elevated expression of CXCR4 consistently across myeloid and T cell sub-clusters (Supplemental Figure 11). Overall, we found that T cells that originated in blood are enriched for innate immune interactions, IL-17 signaling, chemokine signaling and sirtuin signaling, whereas those predominantly from hematoma are enriched for stress response pathways, diapedesis and neuroinflammation. Both the patient's long-term follow up blood sample and the healthy control blood samples were predominated by T cell sub-cluster 0, which was defined by two marker genes (TXNIP and LTB).
To better contextualize our T cell sub-clustering results, we again performed module-scoring analysis with known T cell gene signatures previously derived from human transcriptional data (Supplemental Table 10).
Unsupervised clustering followed by PCA on module scores revealed several modules that describe variation across T cell sub-clusters (Supplemental Figure   12A). We found that three gene modules that describe effector functions in T cells drove major axes of variation in our dataset ( Figure 3D). One signature is associated predominantly with T cell sub-cluster 4 and is defined by T cell effector genes like GNLY, CCL4, and GZMK (Effector Module Score 1, Supplemental Figure 12B). The other is associated with T cell sub-cluster 6, which is a hematoma cluster emerging during rebleeding, and is defined by T cell effector genes like CXCR4, TXN, and CSTB (Effector Module Score 2, Figure 3d, Supplemental Figure 12B). Generally, cells in clusters on the right in the tSNE plot shown in Figure 3D, including T cell sub-clusters 0, 2, and 7 (found predominantly in blood), score higher for a naïve T cell module. Overall, these data -combined with our pathway enrichment results -suggest that T cells in the predominantly hematoma-derived clusters may be enriched for unique T cell functions, including several related to effector functions and stress responses.
Based on the emergence of several blood-like monocyte and T cell subclusters in the hematoma at 73 hours, as well as increased drainage volume and cellularity, we estimate that the time of re-bleeding was between 66 and 73 hours after ICH onset, a full day prior to detection by CT scan. It is also intriguing that neurons emerged in the hematoma effluent immediately prior to this event, but the consequences of this is unknown as the patient continued to improve.
Monitoring of specific peripheral blood biomarkers found in effluent could thus potentially enable earlier detection of a re-bleeding event to guide a more effective treatment paradigm. We caution, however, that our study presents only one patient's ICH trajectory; nevertheless, it suggests that larger scale scRNAseq studies could provide more insight into specific blood cell states that may better inform patient treatment and outcome and highlight pathways for future research.
Our transcriptional analyses of leukocytes over the course of ICH in this patient suggest coordinated, dynamic stages of monocyte and T cell responses within the hematoma that were altered after asymptomatic re-bleed, indicating rapid adaptation and response to local changes in the hematoma tissue milieu.
Additionally, we studied the cellular composition of the blood of this patient more than two years after ICH, and our data suggest that the peripheral blood of this patient, who had an excellent outcome, is comparable to that of age-matched controls.
The dynamics of leukocyte infiltration and activation after ICH have been difficult to study in patients, especially at the site of injury. 16 At present, our understanding is derived primarily from rodent models and peripheral measures; how the kinetics in these models compares to those in humans is unknown. 17,18 Using the infrastructure of the MISTIE III trial, we characterized the local cellular response to ICH and an associated re-bleeding event in the brain of a living patient at single-cell resolution for the first time. We characterized thousands of single cells from both blood and hematoma effluent at several points after ICH onset, allowing us to identify cellular phenotypes that were dynamic in both the T cell and myeloid lineages. Most notably, the emergence of myeloid sub-cluster 2 and T cell sub-cluster 2 in the hematoma after re-bleeding, followed by the sequential emergence of new cell states, suggests that newly infiltrating immune cells adapt transcriptionally to dynamic injury events in the brain.
Crucially, our work demonstrates how rapidly immune cell activation states change after brain injury in humans and sets the stage for framing critical time periods for future studies. Our data, while we caution is a small case study, provide proof-of-principle that, in the near future as personalized medicine evolves, this and other approaches could be used clinically to interrogate cellular responses and uncover meaningful clinical events in patients to help guide treatment decisions.

METHODS
The methods are described in the Supplemental Materials.

Patient enrollment
The patient was enrolled in the MISTIEIII trial (NCT01827046) under an IRB approved protocol including sample collection. Control donors (n=4, 2 males and 2 females, ages 63-93) were recruited and samples were also collected under a IRB approved protocol.

Patient hematoma effluent and blood collection and processing
Each sample was centrifuged at 500g for 10 minutes and the supernatant was removed.
Then each sample was resuspended in HBSS (Gibco) and treated with 2. 50 bases) and custom sequencing primers.

scRNA-seq computational analyses
Raw sequencing data was demultiplexed and aligned to the Hg19 genome using publicly available scripts on Terra (scCloud/dropseq_workflow, version 11, https://app.terra.bio/). The resulting UMI-collapsed digital gene expression matrices were used as input to Seurat (v3) for further analyses in R (v3.6.2). Initial clustering was performed, and on the basis of marker genes identified using Seurat's Wilcoxon ranksum test, red blood cells were identified by expression of hemoglobin genes and excluded from further analysis, resulting in a total of 31,722 cells across all conditions.
All data were normalized, the top 2,000 variable genes identified, and the data was clustered at a resolution of 2. Marker genes for each cluster were identified using Seurat's Wilcoxon rank-sum test, and broad immune cell types were labeled by comparing to known marker genes that have been previously published, and clusters were collapsed into labels as shown in Figure 1 Table 6 and 10) and scores were generated in R using Seurat.

Mass cytometry sample preparation
Cell suspensions from each time point was stained for mass cytometry and fixed immediately upon isolation. 4 Antibody mixes (Supplemental Table 2 Intercalator-IR (Fluidigm #201192A) diluted in Fix/Perm buffer, and incubated overnight at 4°C. Immediately before running each sample on the mass cytometer, that sample was centrifuged, supernatant was removed, and the pellet was resuspended in deionized water.

Data analysis for mass cytometry
Initial gating was performed to identify CD45 hi cisplatin lo viable leukocytes, after first using Iridium intercalator to exclude doublets and non-cellular debris. Cells meeting these criteria were then normalized using bead standards to account for variations in cytometer sensitivity as previously described. 4 Cell populations were identified using

Flow cytometry analysis of blood and hematoma leukocyte populations
Cell suspensions were generated as described above, except granulocytes were not depleted prior to red blood cell lysis.

Data availability
Raw sequencing data will be made available through Gene Expression Omnibus (Study ID: #####) concurrent with publication, as well as through The Alexandria Project (https://singlecell.broadinstitute.org/single_cell?scpbr=the-alexandria-project).  from each myeloid sub-cluster is visualized separately (sub-clusters 0-11), or as a function of time in either blood or hematoma, and as patient derived or control blood derived. Plots are grouped by metadata characteristics, predominantly blood, predominantly hematoma, or both. Gene signatures used for module scoring are described in Table S10.

Gene Set Name Description Reference
Szabo gene sets Generated from resting and stimulated human T cells isolated from blood and tissues (lungs, lymph nodes, bone marrow).
Szabo et al. 6 Miao gene sets Generated from summary of marker genes across the literature (Table S1).