A hyperacute immune map of ischaemic stroke patients reveals alterations to circulating innate and adaptive cells

Stroke affects millions of people across the globe and infections following a stroke often lead to death or disability in patients. Looking rapidly after a stroke in patients, we identify early alterations to the innate and adaptive cells in circulation. Our findings highlight novel changes to dendritic cells, monocytes, haematopoietic stem and progenitor cells, B and T cells to be further investigated as they could have implications the development of post‐stroke infection and poorer patient outcomes.


Introduction
Stroke (cerebral ischaemia) is the leading cause of adult disability and a major cause of death worldwide [1,2]. In addition to initial neurological impairments, such as dysphagia, stroke can rapidly exacerbate the risk of bacterial pneumonia which is independently associated with increased mortality and worse functional outcome in survivors [3][4][5][6]. Bacterial pneumonia could, in part, result from a pneumonitis with an inflammatory component [7,8].
Together with reduced motor control (e.g. contralateral respiratory muscle/diaphragmatic weakness contributing to hypoventilation) [9] and increased vagal drive [10], treatment in a high infection risk environment could also exacerbate an environment for the development of bacterial pneumonia. The ineffectiveness of candidate primary interventions such as prophylactic β-blockers and antibiotic therapy in preventing stroke-associated pneumonia (SAP) has necessitated a clearer understanding of the systemic immune landscape in stroke, to identify mechanisms underlying early complications and new therapeutic approaches to improve patient outcome [11][12][13]. As most infections, particularly SAP, manifest within the first 72 h of stroke onset [14], novel strategies to prevent SAP need to be applied as soon as possible after stroke occurrence. Defining changes in the circulating immune cell compartment very early (within hours) after stroke onset, prior to manifestation of infections, is therefore critical.
Cerebral ischaemia drives local microglial activation in tandem with peripheral immune cell (e.g. monocytes, neutrophils, B cells and T cells) recruitment into the brain, as well as the production and release of cytokines and chemokines [15]. The ensuing inflammation is adversely involved in the evolution of stroke pathology and through its impact on systemic immunity leads to an increased risk of infection by impairing anti-microbial function. The mechanisms underlying systemic immune changes after stroke are not fully understood, but increased catecholamine release by the sympathetic nervous system, glucocorticoids and changes in cholinergic output are proposed [4,[16][17][18].
To this end, we performed a small, prospective, observational immunophenotyping study where we explored the profile of forty immune parameters across three panels on whole blood cells from ischaemic stroke patients at a median time of 168 min following stroke symptom onset and in controls of similar age and sex distribution. Our data identify for the first-time decreased frequencies of a specific subset of dendritic cells (DC) (type I conventional dendritic cells, cDC1), altered expression of HLA-DR, CD64 and CD14 in distinct myeloid populations and alterations to haematopoietic stem and progenitor cells (HSPC). We also show a previously undescribed modulation of memory B cells as well as terminally differentiated effector memory T cells re-expressing CD45RA (TEMRA). Based on CD69 + expression, we also observed a rapid activation of naive and central memory T (TCM) CD4 + T cells, the latter of which are capable of migration to lymphoid organs and orchestrate antigen recall responses [45]. The frequency of CD69 + CD4 + T cells inversely correlated with stroke severity. Our findings highlight unappreciated early changes in both the innate and adaptive immune compartments to be validated in larger cohorts of patients, and provide valuable hypothesis generation regarding how diverse immune cell populations could independently and differentially change following stroke.

Participants, study design and sample acquisition
Samples in this study were part of the flow cytometry substudy and obtained from participants recruited as part of the Subcutaneous Interleukin-1 Receptor Antagonist (IL-1Ra) in Stroke Study [46] Hyper Acute Stroke Unit (HASU) were recruited as previously described [46]. Approval was obtained to draw a blood sample immediately on admission (median time to sample was 168 min), prior to thrombolysis (when eligible), randomization to treatment and receipt of consent for participation in the Phase II trial. The National Institutes of Health Stroke Scale (NIHSS) was used to assess stroke severity at admission; investigational treatment was administered < 6 h within symptom onset. Community-dwelling control participants with similar age and sex distribution to the patients and with no prior history of transient ischaemic attack or stroke, or infection treated with antibiotics within the preceding 6 weeks and capable of informed consent were also recruited. The demographics and baseline characteristics of the cohort are summarized in Table 1.

Cell isolation from blood
Briefly, 3 ml of venous blood collected in ethylenediamine tetraacetic acid (EDTA) tubes (Sarstedt, Leicester, UK) was washed in phosphate-buffered saline (PBS) and resuspended in 9 ml of sterile water (Hyclone, Cramlington, UK) for 10 sec at room temperature twice. Cell suspensions were washed, resuspended in PBS and counted before staining.

Flow cytometry
Single-cell suspensions of blood (~3 × 10 6 -5 × 10 6 total cells) were incubated in PBS for 15 min at 4°C in the dark with the Zombie UV™ or Zombie Aqua™ Fixable Viability Kit (BioLegend, London, UK) as appropriate and immunoglobulin (Ig)G from human serum (Sigma Aldrich, Gillingham, UK). Cells were washed in PBS and stained for a further 15 min in three cocktails of fluorochrome-conjugated antibodies to identify myeloid, B or T cell subsets. Our immunophenotyping panels were adapted from Haniffa et al. [47], Tsang et al. [48], Thome et al. [49] and Thome et al. [50], which are summarized in Table 2. Cells were fixed in 2% paraformaldehyde (Sigma Aldrich) at room temperature for 10 min, washed and resuspended in PBS prior to acquisition. Samples were acquired on an LSR Fortessa using facsdiva version 8 software (BD Biosciences) and, typically, all cells in the sample were collected. Data were analysed using FlowJo software (Treestar, Inc., Ashland, OR, USA). We identified all immune cells based on their expression of CD45 and the lineages were determined based on the indicated markers: Data are median (interquartile range) m where m is the number of missing data points.

Dimensional reduction of flow cytometry data.
Flow cytometry data were gated to exclude debris, dead cells and doublets leaving behind live cells for Boolean gating and high-dimensional analyses. We employed a combination of expert guided manual gating and dimensional reduction using the uniform manifold approximation and projection (UMAP) algorithm, which we implemented in FlowJo. We adopted UMAP over t-distributed stochastic neighbor embedding (t-SNE), as it faithfully visualizes cell clusters in the high-dimensional space following dimensional reduction with a shorter run time [51,52]. UMAP plots were generated separately for individual samples, such that either the same number of cells or all available cells were sampled to avoid artefacts due to insufficient representation of cells in the sampled gate. Immune cells manually identified based on previously described characteristics were projected on the UMAP to generate a global immune map.

Statistical analyses
Statistical analyses were performed using Prism version 7/8 software (GraphPad, San Diego, CA, USA) and data are presented as median with individual data points. Statistical comparisons were performed using a Mann-Whitney U-test and correlations using Spearman's ranked coefficient correlation test. Following sample processing, surface staining and data acquisition, if insufficient cells were present in individual samples for specific panels they were excluded from analyses and appropriate samples numbers are indicated in the Figure legends. One control and stroke patient were excluded from all analyses as indicated in the results and a further three controls and two stroke patients lacked data for up to two panels.

Study cohort
We established a research protocol to rapidly sample peripheral blood from ischaemic stroke patients upon admission to the HASU at the MCCN for deep immune profiling as well as assaying acute-phase proteins. In total, we recruited 13 patients who were stratified for stroke severity according to the NIHSS, of which we analysed 12 samples. We also recruited 16 control participants of similar age and sex distribution in our study (Table 1),  of which we analysed 15 samples. One control was excluded from all analyses due to a previous diagnosis of chronic lymphoid leukaemia, while one stroke patient was excluded, as sample processing yielded no viable cells for analysis. Our cohort of stroke patients was 62% male, with a median age of 75 [interquartile range (IQR) = 65-81] years whose characteristics are summarized in Table 1. The median time to blood sampling following stroke symptom onset was 168 min, at which point 85% of the patients were also thrombolysed after blood draw was obtained.

Stroke alters the composition and phenotype of circulating myeloid cells in the hyperacute phase
To characterize the heterogeneity of myeloid cells, we gated them to identify granulocyte, monocyte and dendritic cell subsets using well-established markers [53][54][55] (Supporting information, Fig. S1a). We identified all DCs based on their expression of CCR2 (Supporting information, Fig. S2a) and divided them into functionally distinct populations based on surface marker expression, notably types 1 and 2 conventional DC (cDC1 and cDC2) and plasmacytoid DC (pDC) [47,56]. The quantification of manually gated populations revealed a modest decrease in the frequency of cDC1s (Fig. 1a,b), while the frequencies of other myeloid populations remained unaltered (Supporting information, Fig. S2b-d). The frequencies of cDC1s poststroke, however, did not correlate with the severity of stroke (Supporting information, Fig. S2e). Enumerating the numbers of immune populations revealed a modest increase in monocytes, particularly classical monocytes (Fig. 1c). We also identified HSPCs based on CD34 expression and the lack of canonical lineage markers and observed a modest decrease in the frequencies of CD34 + HSPCs ( Fig. 1D and Supporting information, Fig. S1b) that could be suggestive of alterations in haematopoietic output or potential.
Alterations to the surface phenotype myeloid populations have been linked to impairments in the ability to generate appropriate immune responses following infection. For example, the expression of HLA-DR has been shown to be critical for presenting processed antigens across a variety of cell lineages [57][58][59], while the FcγR has been shown to be indispensable for antigen uptake by cDC2s [60,61]. Therefore, we next investigated the surface phenotype of myeloid cells which highlighted HLA-DR, CD64 and CD14 as markers that were altered hyperacutely ( Fig.  1e-g). Indeed, we identified a modest down-regulation of HLA-DR on intermediate monocytes, cDC2s and pDCs post-stroke, correlating with age but not stroke severity for cDC2s ( Fig. 1e and Supporting information, S2f). We also observed a significant down-regulation of CD64 (FcγRI; high-affinity Fc receptor for monomeric IgG) on classical monocytes and cDC2s (Fig. 1f). Furthermore, we identified increased CD14 (Toll-like receptor 4 signalling co-receptor) expression across all monocyte subsets (Fig.  1g). CD14 up-regulation has been linked to the acquisition of a tolerance to endotoxin challenge [62]. Taken together, our data could imply rapid alterations in antigen presentation, phagocytic capacity and the ability to secrete cytokines in response to infections following ischaemic stroke.

Stroke rapidly decreases the frequency of unswitched memory B cells in circulation
Stroke has been shown to drive the loss of B cells in experimental stroke [26], and they have been linked to cognitive decline post-stroke due to their recruitment to the ischaemic brain in patients [27]. To understand the early impact of cerebral ischaemia on B cells, we identified B cells in whole blood of stroke patients and observed that at a hyperacute time-point, frequencies of B cells were largely unaltered (Fig. 2a). Emerging research has highlighted impairments in antibody-mediated immunity as a key driver of post-stroke infections and the critical role that B cell subsets play [26,40,63]. We next identified memory B cell subsets based on their expression of IgD and CD27 [48], which revealed a decrease in the frequency of unswitched memory B cells (Fig. 2b), but this did not correlate with the stroke severity (Fig. 2c). Although plasmablasts and other memory B cell populations remained unaffected (Fig. 2b), a larger cohort might be required to identify changes to the pool of memory B cells. Given that unswitched B cells can rapidly mount IgM-driven anti-microbial responses and enter the germinal centre reaction [64][65][66][67], our observations suggest that stroke rapidly drives deficiencies in humoral immunity that could modulate infection susceptibility in patients.

Stroke rapidly alters the phenotype of the memory T cell compartment in circulation
Clinical studies have identified the loss of T cells in circulation [25,68,69] as well as long-term functional alterations [20] as a characteristic of stroke-induced immunosuppression. We first identified CD3ε + T cells in the peripheral blood of controls and stroke patients. In accordance with previous studies, we identified a decrease in the frequency of T cells, but this did not correlate with stroke severity (Fig. 3a). Due to the segregation of functions between TCR-αβ + and TCR-γδ + T cells and their subsets [70][71][72], as well as the role of IL-17 + γδ T cells in exacerbating ischaemic injury [22,73], we characterized T cell subsets based on TCR as well as CD4 and CD8 expression. However, stroke did not impact the frequencies of CD4, CD8 or TCR-γδ + T cell subsets at a hyperacute time-point (Supporting information, Fig. S3a).
To investigate the activation and exhaustion characteristics acquired by T cell subsets following stroke, we implemented the UMAP algorithm [51,52] on CD4 + and CD8 + T cells and projected memory T cell subsets identified by their expression of CCR7 and CD45RA [49] on the dimensional reduction, and observed a decrease in TEMRA cells among CD4 + but not CD8 + cells ( Fig.  3b and Supporting information, S3b). However, we observed a high degree of variability in the frequencies of memory CD8 + T cell subsets amongst patients, potentially ascribed to co-morbidities that were not matched with controls (Supporting information, Fig. S3b). Both the CD4 + and CD8 + compartments were rapidly activated by stroke, evidenced by the increased proportion of cells expressing CD69 (Fig. 3c-f), a marker of early T cell activation [74]. Increased frequencies of CD69 + T cells inversely correlated with stroke severity only in the CD4 + compartment (Fig. 3d). The increase in CD69 + CD4 + cells was restricted to naive and central memory T cells (TCM), which also inversely correlated with stroke severity ( Fig. 3e and Supporting information, S3c), while all memory subsets up-regulated CD69 in the CD8 + compartment (Fig. 3g). Finally, examining programmed cell death (PD)-1 + T cells within the CD4 + and CD8 + compartments as a measure of T cell exhaustion [75,76]  were performed using a Mann Whitney U test, ****P < 0·0001, ***P < 0·001, **P < 0·01, *P < 0·05.
revealed no alterations in PD-1 + cells within either compartment at a hyperacute time point (Fig. 3h). Combined, our data outline how stroke swiftly alters the composition and phenotype of myeloid and lymphoid cells in circulation and could have implications for antigen presentation and circuits of humoral immunity as well as memory T cell responses.

Discussion
Infectious complications following ischaemic stroke present a significant barrier to recovery and are thought to be driven by alterations to the systemic immune landscape. Here, we present a hyperacute map of circulating immune cells in ischaemic stroke patients where we show decreased frequencies of cDC1s, HSPCs, unswitched memory B cells and TEMRA cells. We also identify concomitant alterations in the expression of HLA-DR, CD64 and CD14 in distinct myeloid subsets and a rapid activation of CD4 + T cells based on CD69 expression. Interestingly, this CD69 + CD4 + T cell phenotype inversely correlated with stroke severity and was associated with naive and TCM cells.
The vast majority of CD4 + or CD8 + T cells have been shown to be CD69 − in circulation [77] and the upregulation of CD69 is thought to be a marker of early T cell activation, probably in response to the cytokine milieu and TCR engagement, which regulates cytokine production [74]. Although we observe an activation of T cell subsets post-stroke, the pathways driving their activation remain to be determined. Similarly, while we also implicate CD4 + TEMRA cells in the hyperacute immune response following stroke, further studies in larger patient cohorts are essential to determine their functional consequence and temporality, particularly in the context of long-term cognitive decline, as increased CD8 + TEMRA cells in circulation have been recently identified as an immunophenotype in patients with mild cognitive impairment or Alzheimer's disease [78]. Given that TEMRA cells possess the capacity to migrate to peripheral tissues and take up residence [79], mechanistic studies are critical to determine whether the decreased frequencies observed could be attributed to their migration or loss via cell death. Studies have documented increased CD69 + T cells in the palatine tonsils and cervical lymph nodes of stroke patients 76 h following (a) stroke onset [80], and that activated T cells reactive to myelin oligodendrocyte glycoprotein accumulate in the brain following experimental stroke, implicating activated T cells in driving autoimmunity post-stroke [81,82]. Conversely, an exhausted T cell phenotype characterized by an increased frequency of PD-1 + CD4 + T cells has also been observed in patients 48 h post-stroke [83].
Given the diverse T cell phenotypes and their pleiotropic roles post-stroke, it becomes crucial to dissect mechanisms that control immunoregulatory programmes in T cells balancing the requirement for a tightly regulated anti-microbial immune response versus initiating an autoimmune reaction. Our observations concur with previously reported features of the disease and, by sampling patients soon after stroke onset, illustrate the rapid effects of stroke on systemic immunity. As such, alterations to the abundance of monocytes as well as their HLA-DR expression have been widely reported [31,69,84,85]; however, we demonstrate that it occurs within 3 h of symptom onset. Our study also highlights that the down-regulation of HLA-DR is a wider phenomenon, affecting cDC2s and pDCs that play key roles in priming T cell responses and type I interferon production. While myeloid DC precursors have previously been shown to be decreased in circulation following ischaemic stroke [33], in recent years our understanding of DC biology and their associated subsets has grown [86][87][88]. Consequently, specific subsets that have been recently identified have not been investigated. In this context, our data outline phenotypical alterations to cDC1s, cDC2s and pDCs in the hyperacute phase post-stroke. Our data add to mounting evidence that implicate pDCs in the ensuing immune response post-stroke and illustrate that pDCs could acquire a functionally altered state within hours of stroke, in addition to their role in the sustained immune response post-stroke [13]. Further, the modest decrease in cDC1 frequencies we observe soon after stroke could result from their recruitment to the ischaemic brain, where they could prime detrimental T cell responses. However, only cDC2s (human: CD1c + , mouse: CD172a + ) have been detected in the brain using post-mortem brain tissue and following experimental stroke [33,73]. Regardless, DCs have been shown to be reduced in circulation and exhibit an impaired capacity to secrete cytokines in patients following subarachnoid haemorrhage, a condition that also drives systemic immunosuppression [89]. Emerging evidence has also highlighted the existence of a novel inflammatory subset of cDC2s which are CD5 − CD163 + CD14 + and has implicated their expansion in systemic lupus erythematosus [90]. Given that we implicate stroke in modulating HLA-DR expression on cDC2s, it remains to be determined whether this phenotype could be ascribed to the newly described inflammatory cDC2 subset.
Altered monocyte phenotypes have previously been reported in stroke patients [30][31][32]84,91] and, congruent with previous data, we also identify a decreased expression of HLA-DR on intermediate monocytes. Although we report no alterations to monocyte proportions in the hyperacute phase of stroke, unlike previous data [31], our sample size and the observed variability could limit our ability to detect these phenotypes. Nevertheless, we identify novel alterations to the expression of CD64 on classical monocytes and CD14 across all monocyte subsets. Enhanced CD14 expression on monocytes has been linked to the acquisition of a tolerance to endotoxin exposure, characterized by an inability to secrete proinflammatory cytokines [62], a phenotype observed in stroke [92]. However, it is unclear if the increased levels of CD14 on monocytes are driven by up-regulation or the generation and release of CD14 high monocytes from the bone marrow, analogous to the concept of innate immune training [15,93,94]. Taken together with alterations to CD64 expression, these alterations could probably compromise the ability of patients to response to infectious challenges.
Examining the heterogeneity in immune cells also demonstrated a modest reduction the frequency of HSPCs as well as unswitched B cells in circulation. Studies have showing an overlay of CD69 + cells on memory CD8 + T cell subsets, quantification of the proportion of CD69 + CD8 + T cells and their correlation with stroke severity (NIHSS). (g) Quantification of the proportion of CD69 + cells among memory CD8 + T cell subsets in controls and stroke patients. (h) Frequency of programmed cell death (PD)-1 + CD4 + and CD8 + T cells in controls and stroke patients. Data are presented as bars showing median values and dots represent individual data points for control (n = 12) and stroke patients (n = 10). In correlations, the regression line and standard error are shown. Statistical comparisons were performed using a Mann-Whitney U-test and correlations using Spearman's ranked coefficient correlation test; ***P < 0·001, **P < 0·01, *P < 0·05.
shown that in experimental stroke there are alterations in bone marrow haematopoiesis that confers a myeloid bias to HSPCs at the expense of lymphopoiesis, affecting B cell development [17,95]. Equally, it is plausible that HSPCs could be retained in inflamed tissues such as the ischaemic brain as they circulate and undergo local haematopoiesis [96,97]. Thus, while we identify a decrease in the frequency of IgD + CD27 + unswitched memory B cells, it is unclear if alterations in HSPC frequencies could modulate the balance of memory B cell subsets as migration and entry into the germinal centre reaction could occur more rapidly. What is clear, however, is that stroke swiftly alters the dynamics of the circulating memory B cells and potentially humoral responses. Further studies are required to causally implicate the decreased frequencies of unswitched memory B cells in driving the observed hypogammaglobulinaemia post-stroke [40,63]. Moreover, it also remains to be determined if stroke directly modulates the frequency and developmental potential of circulating HSPCs.
Our study has several limitations, including a small sample size recruited from a single hospital which might not be representative of the wider population of stroke patients. Serial blood samples in the patients were not obtained for immunophenotyping, meaning that insights into the changes in immune phenotype beyond the hyperacute phase were not possible. Our control and stroke groups are also not precisely matched for co-morbidities and our data could over-estimate the effect of stroke on immune function. This is evidenced by recent work that highlighted how risk factors for cerebrovascular disease and their genetic susceptibility loci, e.g. hypertension, obesity, atherosclerosis and hyperlipidaemia, can modulate haematopoiesis and innate immunity [98]. As a result, larger appropriately powered, co-morbidity-matched cohorts assessing multiple immune parameters are critical to validate our findings and to determine the relationships with post-stroke infection and clinical outcomes. Our small sample size and low numbers of patients with infection limit speculation on immunophenotypes that could be predictive of infection, further reinforcing the need for larger studies. Although we profile immune cell subsets, we were unable to concurrently map the systemic cytokine, complement, autonomical and hypothalamic-pituitary adrenal axis response with a similar level of depth due to insufficient blood volumes. Thus, mechanistic studies are essential to determine the differential roles of the adrenergic, cholinergic and glucocorticoid pathways driving early and delayed alterations to immune function. Although we present a snapshot of hyperacute immune alterations, it is equally important to analyse the temporality of immune changes over the acute and post-acute phases and how they shape outcome following the ischaemic insult; for example, in the context of cognitive decline [13]. Regardless, our study is the first to place an equal emphasis on the innate and adaptive immune compartments and prospectively identifies novel immunophenotypes that track with disease severity during the hyperacute phase of stroke, and that warrant more detailed follow-up studies.

Supporting Information
Additional supporting information may be found in the online version of this article at the publisher's web site: Fig. S1. Immune cell types identified in stroke patients. (A,B) Representative FACS plots showing flow cytometric gating strategy employed to identify myeloid cell subsets (A) and HSPCs (B) in control and stroke patients. Lineage contains CD3, CD19, CD20, CD15, CD56, CD66b.