Characterization of Monocyte Subsets in a Prospective Cohort of Patients with Acute Stroke Suspicion: Results of BOOST Study

Abstract

Introduction

Rapidly sorting patients with large vessel occlusion (LVO) ischemic stroke is crucial to ensure efficient transfers to stroke units. Peripheral monocyte subsets (classical Mon1, intermediate Mon2, non-classical Mon3) could be interesting candidate biomarkers in this setting: their profiles in the first hours after stroke symptom onset are unknown.

Aim

To characterize monocyte subsets in patients admitted to emergency units for acute stroke suspicion.

Methods

BOOST (“Biomarkers-algOrithm-for-strOke-diagnoSis-and-Treatment-resistance-prediction,” NCT04726839) is a prospective multicenter cohort. Adult patients with symptoms suggesting acute stroke within the last 24 hours were included. Blood was collected upon admission before brain imaging. Flow cytometry (FCM) was performed on fresh blood with gating based on CD45/CD14/CD16/CD91 as well as on activation markers (CD62L/CD11b/CD86/HLA-DR/CCR2/ICAM-1/CX3CR1/TF).

Results

Of the 298 consecutive patients tested, mean age 64.0 ± 18.7 years, 64 (21.5%) had LVO stroke versus 234 (78.5%) other diagnosis (non-LVO ischemic stroke, cerebral venous thrombosis, intracranial hemorrhage, transient ischemic attack, and stroke mimics). The median time from symptom onset to sampling was 2.3 hours. We found a significantly lower proportion of Mon3 (geometric mean: −47%, p = 0.0093) and a higher proportion of Mon1 (+1.6%, p = 0.0296), suggesting earlier Mon1 mobilization and patrolling Mon3 consumption in LVO patients versus those without. Using linear-mixed-effect model, significant differences in ICAM-1 and HLA-DR expression on monocyte subsets were evidenced between LVO and other patients.

Conclusion

This is the first study to evidence monocyte subset differences in LVO versus non-LVO patients at the time of admission, indicating an acute systemic response in LVO. Whether Mon assessment would add value for LVO diagnosis remains to be determined.

Keywords

monocytes - stroke/prevention - clinical studies

Introduction

Stroke is a major public health issue and one of the “incoming epidemic of the century.” By 2035, with population aging, stroke's incidence is expected to increase by 30% in Europe, that is, 40,000 new cases, reaching more than 2.58 million cases per year.[1] Stroke is the leading cause of disability and the second leading cause of death. Approximately 85% of acute strokes are ischemic. Acute ischemic strokes (AIS) due to large vessel occlusion (LVO) are at higher risk of disability and mortality. Time is running out in this acute clinical setting because a 30-minute delay in the recanalization of the ischemic area increases mortality by 20%.[1] [2] In addition to medical imaging, there is currently a major interest to search for blood biomarkers differences in patients with LVO stroke among those with symptoms consistent with acute stroke.

Several clinical studies suggest that peripheral monocyte subsets could be interesting cell biomarker candidates of severity and prognosis of cerebral and cardiovascular disease.[3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] According to the nomenclature described in 2010, circulating monocytes, which are cells of innate immunity, are subdivided into three major phenotypic subsets, based on the expression intensity of the surface molecules CD14 (lipopolysaccharide receptor) and CD16 (FcγRIII immunoglobulin receptor): “classical” (Mon1, CD14⁺⁺CD16⁻), “intermediate” (Mon2, CD14⁺⁺CD16⁺), and “non-classical” monocytes (Mon3, CD14⁺CD16⁺⁺), accounting for 85 ± 6%, 5 ± 2%, and 10 ± 4% of total monocytes, respectively.[14] [15] The European Society of Cardiology (ESC) working groups on “Atherosclerosis and vascular biology” and “Thrombosis” have recommended using additional cell surface markers expression to improve the monocyte subset gating by flow cytometry (FCM).[16] Indeed, the difficulty lies in the phenotypic heterogeneity of surface CD14 and CD16 expression during monocyte differentiation due to the continuous nature of the process; particularly, the discrimination between Mon2 and Mon3 subsets have been significantly improved.[16] [17] [18] [19] [20] [21] [22] Furthermore, the inter-operator variability in monocyte subset gating strategy should be thoroughly evaluated, and the operator effect taken into account in the final analysis of FCM data. Using an original monocyte subset gating strategy we recently developed, we have demonstrated that the impact of the operator effect on monocyte subtype quantification was weak.[22]

Clinical studies regarding the characterization of monocyte subsets in stroke patients differed according to study design, stroke etiology, and clinical settings.[2] [3] [4] [5] [8] [12] Importantly, the first ones have been conducted before the recommendations on monocyte subset-gating strategy in cardiovascular disease, possibly leading to inaccurate classification.[16] Moreover, most studies have been performed on patient samples collected beyond 24 hours following stroke symptom onset.

However, in the first few hours after the onset of stroke symptoms, monocyte subset profiles (including number and activation markers) are not known. We conducted the BOOST (“Biomarkers algorithm for strOke diagnOSis and Treatment resistance prediction”) study, including patients with suspected acute stroke during the early phase (i.e., within 24 hours following symptom onset). In patients included in BOOST study, the main objective was to investigate a potential association between peripheral monocyte subsets and the final diagnosis, especially in patients with LVO stroke versus all other patients with suspected acute stroke. We characterized the peripheral monocyte subsets including their activation state using FCM on fresh blood sample drawn upon admission to hospital.

Methods

Patients

BOOST (NCT04726839) is a non-interventional multicenter cohort, funded by the BOOSTER project (ANR-18-RHUS-0001). It was approved by the committee for the protection of individuals on September 11, 2020, and by the French national commission for information technology and liberties (CNIL) on November 3, 2020. BOOST patients included in the present study were recruited in stroke units of three hospitals in Paris: Lariboisière University hospital (AP-HP.Nord), Adolphe de Rothschild Foundation, and Foch hospital. Inclusion criteria were: (i) patients aged 18 years or older with symptoms suggesting acute stroke within the last 24 hours; (ii) consent obtained (emergency inclusion procedure; all patients provided non-opposition to the use of their clinical data for research purposes); (iii) affiliation to the French social security system. The non-inclusion criteria were patient under guardianship or curatorship. For the present study, an additional inclusion criterion was a maximal time of 24 hours between blood collection and sample processing (see below). According to standards of care, patients had magnetic resonance imaging (MRI) or a brain computed tomography (CT) scan on admission if needed. In the three emergency stroke units, magnetic resonance angiography (MRA) (with TOF) was the first-line imaging test (except for patients with pacemakers). LVO diagnosis was determined on MRI, particularly on the TOF sequence. In case of doubt due to the lower quality of MRA, additional examination by computed tomography angiography (CTA) was performed, more often in patients with suspected LVO. Images were systematically reviewed by both a senior neuroradiologist and a vascular neurologist. LVO included intracranial internal carotid artery, M1, M2, M3 segments of the middle cerebral artery, basilar artery, P1, P2 of posterior cerebral arteries, and intracranial vertebral arteries. The final diagnosis was recorded by the neurologist responsible for the patient, after collecting data from the clinical examination and imaging, laboratory test results, and additional tests if necessary. In case of uncertain diagnosis, a second neurologist adjudicated the final diagnosis. Patients included in BOOST cohort were finally classified as follows: LVO patients, non-LVO AIS, cerebral vein thrombosis (CVT), intracranial hemorrhagic stroke (ICH), transient ischemic attack (TIA) or “stroke mimics” (epilepsy, migraine, metabolic disorders…).

Data Collection

Physicians prospectively collected demographic data (age, gender, weight), clinical data (time of first acute stroke symptom onset, National Institutes of Health stroke scale [NIHSS] upon admission, type of brain imaging, final diagnosis), therapeutic data (anticoagulant or antiplatelet drug upon admission), and laboratory data (complete blood count [CBC], C reactive protein [CRP], fibrinogen, D-dimer, creatininemia upon admission). Physicians were blinded to the results of FCM analysis, and pathologists were blinded to the results of the final diagnosis.

Blood Collection and Transport

On patient admission, 4 mL of blood was collected by venipuncture into a single potassium EDTA extra-tube during the initial sampling before brain imaging, in addition to those required for standard care. Whole blood was rapidly transported to the accredited Hematology laboratory of Lariboisière hospital at controlled temperature (15–25 °C) for centralized FCM analysis. Times of blood collection, FCM staining, and acquisition were systematically recorded.

Complete Blood Count

CBC, including leukocyte differential, was performed on XN-3000 analyzer (Sysmex, Kobe, Japan) before FCM analysis.

FCM Analysis of Peripheral Monocyte Subsets

FCM analysis of monocyte subsets was performed on a dual-laser six-color FACS-Canto-II cytometer (Becton-Dickinson, Franklin Lakes, NJ, USA) as previously described in detail.[22] The choice and titration of the monoclonal antibodies panel, the lysis conditions, and the monocyte subset gating FCM strategy were optimized in collaboration with BioCytex (Marseille, France). In a preliminary study performed on six healthy subjects with normal CBC, we checked the stability of backbone and activation markers over 24 hours. A compensation matrix was calculated and applied to the data before analysis to account for spectral spillover.

The panel included 13 antibodies split into four tubes (numbered 1 to 4) for compatibility with our 6-color cytometer: master mix for monocyte subset identification, namely, CD45-PC7 (pan-leukocyte), CD91-FITC (pan-monocyte), CD14-PerCP-Cy5.5, and CD16-APC-H7; HLA-DR-FITC in all tubes except for tube 2; and eight activation markers (CD62L-APC and CD11b-PE in tube 1; HLA-DR-APC and CD86-PE in tube 2; CCR2-APC and ICAM1-PE in tube 3; CX3CR1-APC and Tissue-Factor-PE in tube 4) ([Supplementry Table S1], available in the online version only).

Into each tube, 100 µL of whole blood was added and samples were incubated (15 min), lysed (15 min), and washed in a BD-FACS Lyse Wash Assistant (Duo-Lyse protocol, Becton-Dickinson). Color-labeled leukocytes were acquired using FACS DIVA software coupled with the FACS-Canto-II and analyzed with FlowJo software version 10.8.1 (Becton-Dickinson, Franklin Lakes, NJ, USA). The ideal condition was to acquire at least 10,000 monocyte-like cells selected on morphological parameters[23] ([Supplementry Fig. S1], available in the online version only). Initial gating strategy was common to the four tubes ([Supplementry Fig. S1], available in the online version only). First, we separate leukocytes from small cell debris; then, singlets were gated by plotting the height against the area for forward scatter (FSC-H vs. FSC-A). Based on SSC +/CD45+ bright properties, a wide gate was created to select monocytes; CD91 marker then allowed separation of monocyte from remaining selected leukocytes. In tubes 1, 3, and 4, HLA-DR-FITC also contributed to monocyte gating in combination with CD91 ([Supplementry Table S1] and [Supplementry Fig. S1], available in the online version only). The expression of some activation markers (CD62L, CD11b, HLA-DR, ICAM1, CCR2, and CX3CR1) was also used for monocyte subset discrimination ([Supplementry Fig. S1], available in the online version only). Two experimented operators were trained to gate monocyte subsets on FlowJo using the same template.[23] A preliminary study with gating performed by three operators on the first 20 patients had shown an inter-operator gating agreement regarding Mon1,2,3 counts within the tolerance limits of the CLSI[22] [23]: in summary, Mon1 count variability was below the CLSI-recommended 10% threshold, whatever the FCM tube, resulting in an excellent inter-operator agreement; although we evidenced slight inter-operator variabilities, CVs of Mon2 and Mon3 counts were below the CLSI-recommended 20% threshold in most patients.[22] The inter-operator variability was lowest in tube 4, encouraging us to choose the antibody combination of tube 4 for Mon1, Mon2, Mon3 subset fraction results.[22] Nevertheless, the slight operator effect[22] was accounted for in the statistical analysis.

Subsequently, two among the three trained operators blindly analyzed BOOST patient FCM raw data files; each of them analyzed half of the patient cohort, and visually checked on FlowJo the gating results of the other operator for the other half of the patient cohort. In case of disagreement, gating was reviewed by the two operators together. Mon fractions were finally converted into G/L (10⁹/L) based on the CBC results.

Median fluorescence intensities (MFI) results were expressed as raw values and relatively to negative cells (lymphocytes for tubes 1, 3, and 4; neutrophils for tube 2).

Statistical Analyses

All analyses were performed using the R software, version 4.0.2,[24] and appropriate packages. Cytometry data after gating were summarized as count of events for each cell type and as MFI of each marker. Analysis was done separately for each tube. Patients were classified, based on their diagnosis, first as “LVO” (namely, LVO-AIS) versus “non-LVO” patients (i.e., with all other diagnosis), and comparisons were done between these two groups. Non-LVO patients were then further classified into four groups as pre-specified: AIS/CVT, ICH, TIA, and stroke mimics. Count data for Mon1, Mon2, and Mon3 were analyzed after log-transformation for the four FCM tubes. Analysis was done separately for the three monocyte subsets, using a linear model including the operator effect. Model assumptions, especially the Gaussian assumption, were checked graphically. Results are given as raw p-values, effect-sizes, and 95% confidence intervals (CI) in the original scale. A difference was considered significant if p < 0.05 (type I error: 5%). As a secondary analysis, a joint analysis of the proportions of the different populations was done, using the disjoint-graph approach.[25] Each node is a population; two nodes are connected by an edge if the ratio of the two population amounts does not change significantly between LVO and non-LVO patients, using p < cutoff as rejection rule; all links between nodes are tested using a T-test on the log scale, with a cutoff p < p _min. Populations are assumed to change if the final graph is disjoint (i.e., there is at least two subsets of nodes with no path connecting the two subsets). The cutoff was defined by simulations to control the type I error rate (wrongly obtaining a disjoint graph) at 5%, leading to p _min = 0.07 for three populations (Mon1, Mon2, Mon3) and p _min = 0.12 for five populations (Mon1, Mon2, Mon3, lymphocytes, neutrophils). Ability of Mon subsets to discriminate between LVO and non-LVO patients was explored with ROC curves (package pROC for R); best threshold was defined as giving the highest Youden index. For this threshold, sensitivity, specificity, and positive and negative predictive values are given, with their exact 95% confidence interval. Using boostrap, 95% confidence intervals for the area under the ROC curves (AUC) were obtained.

MFI data for the eight activation markers were also analyzed after log-transformation. Each marker was analyzed separately. Analysis was done jointly for the three monocyte populations, using a mixed-effect model with the patient as a random effect:

where MFI _i, j is the MFI observed on monocytes of type i [Moni] in patient j, µ₁ the mean log-MFI for Mon1 in non-LVO patients, δ _k the change in mean log-MFI for Monk in non-LVO patients, δ_LVO the change in mean log-MFI for Mon1 in LVO patients, δ _k,LVO the additional change in mean log-MFI for Monk in LVO patients (interaction term), δ_Op the operator effect, Z _j the patient effect (random variable assumed to follow a normal distribution of mean 0 and of standard deviation σ_Z), and ε the residual error (random variable assumed to follow a normal distribution of mean 0 and standard deviation σ); 𝕀_X is the indicator of X (equals to 1 if X is true, 0 otherwise). Z _j and ε were assumed to be independent. This model was fitted using the lmer function of the lme4 package[26]; assumptions were checked graphically. Through profiling 95% confidence intervals were obtained; p-values were obtained using likelihood ratio tests (LRT) for nested models, using the asymptotic chi-square distribution.

For all three analyses (linear model for Moni subpopulation count; disjoint graphs for Moni proportions; mixed-effect linear model for activation markers), the analysis was made first with no additional cofactor (except the operator effect), then by adding adjustment for patient's age and time between sampling and technical processing as additional variables in the model.

Results

Patient Characteristics

Among patients admitted to stroke units for stroke suspicion and included in BOOST study from April 2021 to June 2022, 330 had an FCM analysis for monocyte subset characterization: 196 from Lariboisière hospital, 102 from Adolphe de Rothschild Foundation, and 32 from Foch hospital. A total of 5 patients withdrew their consent, 2 did not sign the non-opposition form, and 25 were excluded for FCM technical failure or for being out of time (>24 hours between sampling and FCM acquisition). Among the 298 patients included in the final analysis, 64 (21.5%) had an LVO AIS and 234 (78.5%) had other diagnosis, i.e., 59 patients (19.8%) non-LVO AIS, 2 (0.7%) CVT, 18 (6.0%) ICH, 19 (6.4%) TIA, and 136 (45.6%) stroke mimics (migraine, hypoglycemia, vasovagal syncope, epileptic seizure…) ([Fig. 1]).

Fig. 1 BOOST flowchart. AIS, acute ischemic stroke; CVT, cerebral venous thrombosis; FCM, flow cytometry; ICH, intracranial hemorrhage; LVO, large vessel occlusion; TIA, transient ischemic attack.

Patient demographic, and clinical and laboratory characteristics on admission are summarized in [Table 1]. LVO patients were older (mean age 68.4 years, vs. 62.7 years for non-LVO patients, p = 0.0528), with 48.5% of female LVO patients versus 41.5% for non-LVO (p = 0.3172). The mean NIHSS was significantly higher in LVO patients (12.3 ± 7.5) compared to non-LVO patients (5.2 ± 6.0; p < 0.0001). The mean whole blood neutrophils and monocyte counts were slightly higher in LVO patients versus non-LVO patients: 7.3 versus 5.9 G/L (i.e., 10⁹/L) (p = 0.0090), and 0.73 versus 0.63 G/L (p = 0.0109), respectively. The mean levels of inflammation parameters were also higher in LVO patient versus non-LVO patients: fibrinogen 4.0 g/L versus 3.6 g/L (p = 0.0207), CRP 19 mg/L versus 10 mg/L (p = 0.0124), and D-dimer (1791 vs. 1031, p = 7.10⁻⁴). Upon admission, 32.8% of LVO patients were taking antithrombotic drugs versus 39.1 of non-LVO patients, mainly antiplatelet agents ([Table 1]). The median time between the onset of symptoms and sampling was 2.30 (IQR 1.33–3.02) hours for LVO patients and 2.33 (IQR 1.58–4.22) hours for non-LVO patients (p = 0.410).

Table 1 BOOST patient characteristics (n = 298)

	LVO patients (n = 64)	Non-LVO patients (n = 234)	P value
Demographic and clinical data
Age (m ± SD) (years)	68.4 ± 17.4	62.7 ± 18.8	0.052
Females, n (%)	31 (48.5)	97 (41.5)	0.323
Smoking  Smokes daily/occasionally (%)  Ex-smoker (%)  Never smoked (%)	17 (26.3) 9 (14.0) 38 (59.7)	45 (19.2) 37 (15.8) 152 (65.0)	0.504
NIHSS (mean ± SD) at admission	12.3 ± 7.5	5.2 ± 6.0	<10−4
Imaging  MRA (%)  CTA (%)	60 (93.8) 11 (17.2)	196 (83.8) 41 (17.5)	0.043 1.0
History of stroke or TIA, n (%)	5 (7.9)	65 (27.8)	0.0007
Arterial hypertension, n (%)	36 (56.3)	126 (53.7)	0.777
Dyslipidemia, n (%)	15 (23.8)	63 (26.8)	0.746
Diabetes mellitus type 2, n (%)	8 (12.5)	55 (23.1)	0.081
Coronary artery disease, n (%)	14 (21.9)	23 (9.8)	0.018
Heart valvular disease, n (%)	3 (4.8)	11 (4.8)	1.000
Cancer, n (%)	1 (1.5)	11 (4.8)	0.474
Laboratory data
Hemoglobin (g/dL)	13.6 ± 1.9	13.5 ± 1.9	0.8816
Leukocytes (G/L)	10.1 ± 4.1	8.6 ± 3.1	0.0110
Neutrophils (G/L)	7.3 ± 3.8	5.9 ± 2.8	0.0090
Lymphocytes (G/L)	1.83 ± 0.9	1.81 ± 0.9	0.9569
Monocytes (G/L)	0.73 ± 0.32	0.63 ± 0.27	0.0109
Platelets (G/L)	259 ± 77	264 ± 88	0.8091
D-Dimer (ng/mL)	1,791 ± 4,532	1,031 ± 1,960	0.0007
Fibrinogen (g/L)	4.0 ± 1.2	3.6 ± 0.9	0.021
CRP (mg/L)	19 ± 33	10 ± 22	0.012
Renal function  creatininemia (μmol/L)  GFR (CKD-EPI) (mL/min/1,73 m2)	89 ± 56 75.6 ± 23.8	84 ± 29 81.9 ± 25.2	0.916 0.101
Time between symptom onset and sampling (hours)—median (IQR)	2.30 (1.33–3.02)	2.33 (1.58–4.22)	0.410
Time between sampling and FCM processing (hours)—median (IQR)	4.95 (1.17–18.8)	1.43 (1.17–12.0)	0.1495
Therapeutic data
Antithrombotic drug upon admission,  n (%) namely  antiplatelet agent, n (%)  DOAC, n (%)  VKA, n (%)  prophylactic LMWH, n (%)	21 (32.8)   15 (23.4) 5 (7.8) 0 0	91 (39.1)   63 (27.0) 18 (7.9) 6 (2.6) 2 (0.9)	0.384   0.251 1.000

Abbreviations: CKD-EPI, Chronic Kidney Disease Epidemiology Collaboration; CRP, C-reactive protein; CTA, computed tomography angiography; DOAC, direct oral anticoagulant; FCM, flow cytometry; GFR, glomerular filtration rate; G/L, Giga/L (10⁹/L); IQR, interquartile range; LMWH, low-molecular-weight heparin; LVO, large vessel occlusion; MRA, magnetic resonance angiography; NIHSS, National Institutes of Health Stroke Scale; TIA, transient ischemic attack; VKA, vitamin K antagonist.

Monocyte Subsets

The median time between sampling and FCM analysis was 4.95 (IQR 1.17–18.8) hours for LVO patients and 1.43 (IQR 1.17–12.0) hours for non-LVO patients (p = 0.1495). Since FCM tube 4 had the best inter-operator gating agreement regarding Mon1, Mon2, and Mon3 fractions,[22] monocyte subset results obtained with this tube are presented here ([Table 1]). Noteworthy, results obtained with other tubes were qualitatively similar when comparing LVO and non-LVO patients.

Mean Mon1, Mon2, and Mon3 proportions were 89.0, 5.7, and 5.3% in LVO patients versus 87.5, 6.1, and 6.4% in non-LVO patients. Overall, we found a significantly lower proportion of Mon3 (geometric mean −47%, p = 0.0093) and a higher proportion of Mon1 (+1.6%, p = 0.0296) in patients with LVO stroke versus those without, but no significant difference for Mon2 (−13%, p = 0.2448). The disjoint graph approach confirmed that these changes between LVO and non-LVO patients were not a result of the compositional nature of the data ([Fig. 2A, B]), with a separation far from the cutting threshold. Considering these three subsets together with lymphocytes and neutrophils, only the Mon3 node remained isolated, with a separation far from the cutting threshold; the Mon2 was connected only to lymphocytes, with a separation very close to the cutting threshold, whereas lymphocytes, neutrophils, and Mon1 formed a fully connected graph ([Fig. 2C, D]).

Fig. 2 Disjoint graph approach showing for Mon3 a different pattern than other populations between large vessel occlusion (LVO) and non-LVO patients. (A, B) Analysis with the three monocyte subsets; (C, D) analysis with the three monocyte subsets, neutrophils (Neu), and lymphocytes (Lym). (A, C) Graph obtained based on the selected cutoff. Each node is a population; two nodes are connected by an edge if the ratio of the two population amounts does not change significantly between LVO and non-LVO patients, using p < cutoff as rejection rule. The cutoff was obtained by simulation, to control the type I error (α < 5%) of falsely observing a disjoint graph if no population change between LVO and non-LVO acute ischemic stroke (AIS) patients. (B, D) Dendrogram of populations, displaying the proximity in population changes between LVO and non-LVO patients based on the cutoff needed to separate the populations in the graph; the horizontal bar shows the selected cutoff. (A) The three nodes of the graph are isolated (no edge in the graph), meaning that the three monocyte subsets evolve differently between LVO and non-LVO patients (p < 0.05). (B) The separation of the three nodes occurs well below the selected cutoff, meaning that the result is robust. (C) The Mon3 node is isolated, the Mon2 node has only one edge (with Lym), whereas the Lym, Mon1, and Neu nodes form a clique: this shows that Mon3 amount changes differently than other populations between LVO and non-LVO patients (p < 0.05), and suggests that Mon1, Neu, and Lym have similar changes. (D) The separation of Mon3 occurs far below the selected cutoff, whereas the separation of Mon2 from other populations is very close to the cutoff, and further separations is far above the cutoff, suggesting that the Mon3 separation is a robust result and that different behavior of Mon2 may be missed due to a lack of power. All these results are consistent with a similar amount of Mon2 in LVO and non-LVO patients, a lower level of Mon3, and a higher level of Mon1, Lym, and Neu.

After conversion to counts using CBC results, the median count of Mon1 was significantly higher in LVO than non-LVO patients, 0.59 versus 0.49 G/L (+19%, p = 0.0087), respectively, with a wide dispersion of results ([Fig. 3]). The median count of Mon3 was slightly lower in LVO than non-LVO patients: 0.022 versus 0.031 G/L, respectively (−28%, p = 0.0075). No significant difference was observed between LVO and non-LVO patients for the median count of Mon2 (+5.0%, 0.034 vs. 0.032 G/L, respectively; p = 0.94).

Fig. 3 Monocyte subset counts in large vessel occlusion (LVO) versus non-LVO patients (after conversion of monocyte fractions with complete blood count).

Both Mon1 and Mon3 were able to discriminate LVO patients better than random, whenever considered as proportions of total monocytes, or as counts in G/L ([Fig. 4] and [Tables 1] and [2]; AUC significantly higher than 0.5: resp. 0.589 [0.506; 0.671] and 0.606 [0.523; 0.690] for proportions, and 0.607 [0.523; 0.686] and 0.609 [0.526; 0.692] for counts in G/L). Conversely, Mon2 fraction or count could not discriminate LVO patients (AUC: 0.548 [0.463; 0.632] and 0.497 [0.412; 0.582]). Finally, using a multivariate logistic regression to predict LVO versus non-LVO, we found that CRP measurement was not significantly worse than using CRP and Mon1 or Mon3.

Fig. 4 Receiver operating characteristic (ROC) curves for prediction of large vessel occlusion (LVO) or non-LVO using monocyte subsets (in %). The best threshold (according to Youden's definition) is given as a red dot. (A) Using Mon1 fraction. (B) Using Mon2 fraction. (C) Using Mon3 fraction. (D) Using the Mon3/Mon1 ratio.

Table 2 Operating-point metrics for monocyte subsets (in %) (sensitivity, specificity, positive predictive value, negative predictive value) for the prediction of LVO or non-LVO

	Threshold	Sensitivity 95% CI	Specificity 95% CI	Positive predictive value 95% CI	Negative predictive value 95% CI
Mon1 (%)	0.9524	0.2812 [0.176; 0.4076]	0.9103 [0.8661; 0.9436]	0.4615 [0.3009; 0.6282]	0.8224 [0.7703; 0.8669]
Mon2 (%)	0.04073	0.3906 [0.271; 0.5207]	0.7265 [0.6646; 0.7825]	0.2809 [0.1907; 0.3862]	0.8134 [0.7539; 0.8638]
Mon3 (%)	0.01134	0.2344 [0.1375; 0.3569]	0.953 [0.9174; 0.9763]	0.5769 [0.3692; 0.7665]	0.8199 [0.7689; 0.8637]
Mon1/Mon3	0.01201	0.2344 [0.1375; 0.3569]	0.953 [0.9174; 0.9763]	0.5769 [0.3692; 0.7665]	0.8199 [0.7689; 0.8637]

Abbreviation: LVO, large vessel occlusion.

Results were not qualitatively different when taking into account time between sampling and CMF analysis, sex or patient age.

Relative distribution of monocyte subset counts according to the different diagnosis, i.e., LVO stroke, non-LVO AIS/CVT, ICH, TIA, and stroke mimics ([Supplementry Fig. S2], available in the online version only), was similar in patients with ICH and LVO stroke, both subgroups showing the highest median values of Mon1, and the lowest median values of Mon3.

Monocyte Activation Marker Expression

HLA-DR and ICAM-1 expression levels were higher on Mon3 and much higher on Mon2 than on Mon1, as expected. We found a significant interaction term between monocyte subsets and diagnosis (p = 0.0016, [Fig. 5]), corresponding to a lower expression of ICAM- 1 on Mon1, a similar expression on Mon2, and a higher expression of ICAM-1 on Mon3 in LVO patients. Similar results were obtained when MFI results were expressed relatively to negative cells. There was also a significant interaction term between monocyte subsets and diagnosis (p = 0.0191, [Fig. 5]) for expression of HLA-DR, corresponding to a lower expression on Mon1, and a similar expression on Mon2 and Mon3, in LVO patients.

Fig. 5 ICAM-1 and HLA-DR expression on monocyte subsets in large vessel occlusion (LVO) versus non-LVO patients. Each dot represents the monocyte activation marker median fluorescence intensity (MFI), expressed in a logarithm scale, of each patient (raw on the left, and normalized on negative cells on the right). For the expression of both markers, the arrows show a significant interaction between monocyte subsets in LVO versus non-LVO patients.

A slightly lower expression of CD86, around −12%, was observed in the three monocyte subsets in LVO patients compared to non-LVO patients (p = 0.001, similar for the three populations: no interaction [p = 0.161]). Similar results were obtained for CX3CR1 and CD11b: −11.6% (p = 0.029; no interaction [p = 0.509]) and −9.6% (p = 0.0281; no interaction [p = 0.285]), respectively. When considering MFI relative to negative cells, only the decrease in CD86 and CX3CR1 remained significant (CD86: −10%, p = 0.030; CX3CR1: −14%, p = 0.034; CD11b, −0.05%, p = 0.994; [Fig. 6]).

Fig. 6 CD86, CX3CR1, and CD11b expression on monocyte subsets in large vessel occlusion (LVO) versus non-LVO patients. Each dot represents the monocyte activation marker median fluorescence intensity (MFI), expressed in a logarithm scale, of each patient (raw on the left: A, C, E, and normalized on negative cells on the right: B, D, F).

No other activation marker (CCR2, CD62L, TF) showed significantly different pattern, either for absolute MFI or normalized MFI ([Supplementry Fig. S3], available in the online version only).

Discussion

BOOST is the first study evidencing early profiles of monocyte subset fractions and counts (in G/L) analyzed from fresh blood in a substantial number of LVO versus non-LVO patients within the 24 hours following the onset of stroke symptoms.

The first strength and novelty of our study was the focus on the quantification of monocyte subsets in the first hours following the onset of stroke symptoms (median 2.3 hours) at patient admission to stroke units in real-life conditions. Indeed, studies assessing monocyte subsets focused on later phases (i.e., days/weeks) following stroke symptoms, hence analyzing subsequent monocyte subset evolution and being complementary to our study.[3] [4] [5] [6] [8] [12] [27] [28]

The second strength of our study lies in the quality of FCM gating strategy for monocyte subset analysis from fresh blood: first, we used the CD91 panmonocytic marker in addition to size, granularity, CD14, and CD16[18] [22]; second, we used different activation markers, hence refining the gating of the three monocyte subsets, especially to allow better separation of the Mon2 and Mon3 subsets, based for instance on their respective expression of HLA-DR or the chemokine receptor CCR2, as recommended.[16] The three subsets were distinct from one another, in their expression of CD11b, with higher expression on the classical and intermediate subsets, and in their expression of CX3CR1, with higher expression on the intermediate and non-classical subset, as expected.[29] Finally, we took into account the slight inter-operator gating variability that we previously evidenced.[22] [23] Remarkably, we observed consistent results in the four FCM analysis tubes regarding the monocyte subset results, thus strengthening our findings. Noteworthy, results obtained using “old” FCM gating approaches used to identify monocyte subsets should be interpreted with caution given the difficulties in accurately separating the different subsets.[2] [3] Inconsistencies between studies limit comparisons between them: optimized and standardized gating strategies should facilitate comparisons between studies in the future.

We evidenced both lower proportion and count of non-classical monocytes (Mon3) in LVO versus non-LVO patients. These findings are consistent with an early recruitment of these cells in case of severe endothelial injury, hypoxemia, and vascular inflammation, which occur after a cerebral LVO.[6] [27] Indeed, the early and rapid recruitment of Mon3 on the site of vascular injury helps in maintaining vascular integrity and clearing debris, especially in large arteries as shown in previous studies in different territories.[12] The early extravasation of Mon3 could lead to rapidly lowering in the count of circulating Mon3 in LVO stroke patients versus non-LVO patients since vascular damage is more severe in LVO patients. Moreover, their patrolling behavior depends on adhesion molecules such as ICAM-1, more expressed in Mon3 of LVO versus non-LVO patients as found in our study. Finally, Mon2, and to a lesser extent Mon3, are involved in antigen processing and presentation functions.

Besides, a higher proportion and a higher count of Mon1 were observed in LVO versus non-LVO patients along with higher inflammation parameter levels (leukocytes, neutrophils, CRP, fibrinogen, and consequently D-dimers). Our study supports the hypothesis of a hyper acute and early response orchestrated by the peripheral immune system following a LVO-AIS stroke, in line with findings of previous studies.[6] [28] [29] Mon1 originate in bone marrow and are released into the peripheral circulation depending on proinflammatory cytokines (e.g., interleukin 6). Their circulating lifespan is physiologically known to be very short, around 24 hours.[3] In addition, Mon1 are stored as substantial marginal reservoir in the spleen and lung that can be mobilized upon need. The higher count of Mon1 in LVO AIS patients versus non-LVO patients could result from a rapid release of classical monocytes into the circulation in LVO patients secondary to early systemic response. Indeed, Mon1 are characterized by their ability to adhere to the endothelium and to migrate,[30] explaining their early mobilization in case of severe vessel injury. Interestingly, we did observe the highest levels of Mon1 in LVO and ICH patients, compared to non-LVO AIS stroke patients, TIA and stroke mimics, highlighting the immune response intensity in the two first conditions.

It is admitted that a small fraction of Mon1 (1% in physiological conditions) turns into Mon2 with the intermediate subset proposed as monocytes in transition from a classical to a non-classical monocyte. Physiologically, after a circulating lifespan of around 4 days, Mon2 turn into Mon3, which have a lifespan of 7 days before disappearing through apoptosis or tissue migration.[30] [31] The kinetics of circulating monocytes is tightly regulated in steady state situations, but stress can lead to rapid release of classical monocytes into the circulation and thus alter the distribution of monocyte subsets.[28] Our findings support the latter scenario, with early Mon1 mobilization and Mon3 consumption in LVO patients compared to patients with other diagnosis, but no significant change in Mon2 subset. Remarkably, these results are consistent with those of previous studies with blood collected beyond the first 24 hours following stroke symptoms (from one to several days), and showing an elevation of circulating pro-inflammatory Mon2.[3] [4] [9]

Studying chemokine and cytokine profiles in peripheral blood in our patient cohort might be of interest, in addition to Mon subpopulation assessment whose performances to predict LVO versus non-LVO were rather poor as shown with ROC curves. Indeed, the combination with other biomarkers might be much more time efficient than FCM. This remains to be addressed in specific future studies.

Our study has several limitations. First, differences of some of the markers (HLA-DR, CX3CR1) were measured on each monocyte subset, but these were (in part) also used to define each subset, thus potentially introducing biases in the analysis. Nevertheless, regarding monocyte subsets, results obtained with the four analysis tubes were qualitatively similar, thus strengthening our results. Second, we did not analyze the potential association of monocyte subset variations with LVO stroke severity and prognosis. Third, patients were sampled once, upon admission: thus, we could not investigate monocyte subset profiles at several times post-stroke to further explore the dynamics and release of monocytes after ischemic injury. Finally, our gating strategy and patient results have to be validated in external labs and cohorts.

In conclusion, this is the first study evidencing modifications in monocyte subsets in LVO versus non-LVO patients within 24 hours following the onset of symptoms, with a significant higher count of Mon1 and a significantly lower count of the patrolling Mon3. Whether Mon assessment would add value for LVO diagnosis in the first hours after the onset of stroke symptoms remains to be determined.

Contributors' Statement

G.C., V.S., and M.M. designed the research; P.R., M.M., B.L., and J.P.D. were clinical investigators; E.H., M.N., F.M., M.D., T.B., M.S., and C.B. performed research; E.H. and F.M. performed FCM gating; E.H., F.M., V.S., F.M., T.B., G.J., M.M., and E.C. analyzed the data; E.H. and V.S. wrote the manuscript; all authors reviewed the manuscript.

Conflict of Interest

The authors declare that they have no conflict of interest.

• References

• 1 Wafa HA, Wolfe CDA, Emmett E, Roth GA, Johnson CO, Wang Y. Burden of stroke in Europe: thirty-year projections of incidence, prevalence, deaths, and disability-adjusted life years. Stroke 2020; 51 (08) 2418-2427
• 2 Mazighi M, Chaudhry SA, Ribo M. et al. Impact of onset-to-reperfusion time on stroke mortality: a collaborative pooled analysis. Circulation 2013; 127 (19) 1980-1985
• 3 Urra X, Villamor N, Amaro S. et al. Monocyte subtypes predict clinical course and prognosis in human stroke. J Cereb Blood Flow Metab 2009; 29 (05) 994-1002
• 4 Kaito M, Araya S, Gondo Y. et al. Relevance of distinct monocyte subsets to clinical course of ischemic stroke patients. PLoS One 2013; 8 (08) e69409
• 5 ElAli A, Jean LeBlanc N. The role of monocytes in ischemic stroke pathobiology: new avenues to explore. Front Aging Neurosci 2016; 8: 29
• 6 Tahir S, Steffens S. Nonclassical monocytes in cardiovascular physiology and disease. Am J Physiol Cell Physiol 2021; 320 (05) C761-C770
• 7 Ptaszyńska-Kopczyńska K, Eljaszewicz A, Marcinkiewicz-Siemion M. et al. Monocyte subsets in patients with chronic heart failure treated with cardiac resynchronization therapy. Cells 2021; 10 (12) 3482
• 8 Krishnan S, O'Boyle C, Smith CJ. et al. A hyperacute immune map of ischaemic stroke patients reveals alterations to circulating innate and adaptive cells. Clin Exp Immunol 2021; 203 (03) 458-471
• 9 Williams H, Mack CD, Li SCH, Fletcher JP, Medbury HJ. Nature versus number: monocytes in cardiovascular disease. Int J Mol Sci 2021; 22 (17) 9119
• 10 Noz MP, Ter Telgte A, Wiegertjes K. et al. Pro-inflammatory monocyte phenotype during acute progression of cerebral small vessel disease. Front Cardiovasc Med 2021; 8: 639361
• 11 Boidin M, Lip GYH, Shantsila A, Thijssen D, Shantsila E. Dynamic changes of monocytes subsets predict major adverse cardiovascular events and left ventricular function after STEMI. Sci Rep 2023; 13 (01) 48
• 12 Hansen RB, Laursen CCH, Nawaz N. et al. Leukocyte TNFR1 and TNFR2 expression contributes to the peripheral immune response in cases with ischemic stroke. Cells 2021; 10 (04) 861
• 13 Hristov M, Weber C. Monocyte subsets in cardiovascular disease: a biomarker perspective. Thromb Haemost 2025; 125 (02) 93-96
• 14 Ziegler-Heitbrock L, Ancuta P, Crowe S. et al. Nomenclature of monocytes and dendritic cells in blood. Blood 2010; 116 (16) e74-e80
• 15 Wong KL, Tai JJ, Wong WC. et al. Gene expression profiling reveals the defining features of the classical, intermediate, and nonclassical human monocyte subsets. Blood 2011; 118 (05) e16-e31
• 16 Weber C, Shantsila E, Hristov M. et al. Role and analysis of monocyte subsets in cardiovascular disease. Joint consensus document of the European Society of Cardiology (ESC) Working Groups “Atherosclerosis & Vascular Biology” and “Thrombosis.”. Thromb Haemost 2016; 116 (04) 626-637
• 17 Hudig D, Hunter KW, Diamond WJ, Redelman D. Properties of human blood monocytes. II. Monocytes from healthy adults are highly heterogeneous within and among individuals. Cytometry B Clin Cytom 2014; 86 (02) 121-134
• 18 Hudig D, Hunter KW, Diamond WJ, Redelman D. Properties of human blood monocytes. I. CD91 expression and log orthogonal light scatter provide a robust method to identify monocytes that is more accurate than CD14 expression. Cytometry B Clin Cytom 2014; 86 (02) 111-120
• 19 Zawada AM, Fell LH, Untersteller K. et al. Comparison of two different strategies for human monocyte subsets gating within the large-scale prospective CARE FOR HOMe Study. Cytometry A 2015; 87 (08) 750-758
• 20 Thomas GD, Hamers AAJ, Nakao C. et al. Human blood monocyte subsets: a new gating strategy defined using cell surface markers identified by mass cytometry. Arterioscler Thromb Vasc Biol 2017; 37 (08) 1548-1558
• 21 Williams H, Mack C, Baraz R. et al. Monocyte differentiation and heterogeneity: inter-subset and interindividual differences. Int J Mol Sci 2023; 24 (10) 8757
• 22 Heng E, Neuwirth M, Mas F. et al. Assessment of inter-operator variability in peripheral monocyte subset gating strategy using flow cytometry in patients with suspected acute stroke. Cytometry A 2024; 105 (03) 171-180
• 23 Clinical and laboratory standards institute (CLSI). Validation of Assays Performed by Flow Cytometry. 1st ed.. CLSI guideline H62. CLSI; 2021
• 24 Core Team. (2020). R: A language and environment for statistical computing.R Foundation for Statistical Computing, Vienna, Austria. Accessed at: https://www.R-project.org/
• 25 Curis E, Courtin C, Geoffroy PA, Laplanche JL, Saubaméa B, Marie-Claire C. Determination of sets of covariating gene expression using graph analysis on pairwise expression ratios. Bioinformatics 2019; 35 (02) 258-265
• 26 Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Softw 2015; 67: 1-48
• 27 Farooqui M, Ortega-Gutierrez S, Hernandez K. et al. Hyperacute immune responses associate with immediate neuropathology and motor dysfunction in large vessel occlusions. Ann Clin Transl Neurol 2023; 10 (02) 276-291
• 28 Blank-Stein N, Mass E. Macrophage and monocyte subsets in response to ischemic stroke. Eur J Immunol 2023; 53 (10) e2250233
• 29 Boyette LB, Macedo C, Hadi K. et al. Phenotype, function, and differentiation potential of human monocyte subsets. PLoS One 2017; 12 (04) e0176460
• 30 Patel AA, Zhang Y, Fullerton JN. et al. The fate and lifespan of human monocyte subsets in steady state and systemic inflammation. J Exp Med 2017; 214 (07) 1913-1923
• 31 Tak T, Drylewicz J, Conemans L. et al. Circulatory and maturation kinetics of human monocyte subsets in vivo. Blood 2017; 130 (12) 1474-1477

Publication History

Received: 22 July 2025

Accepted after revision: 04 February 2026

Accepted Manuscript online:
09 February 2026

Article published online:
17 February 2026