This exploratory analysis is part of an observational biomarker sub-study nested within the multicenter, randomized controlled, endpoint blinded PHYS-STROKE trial. The details of the PHYS-STROKE study design and the main results have been published previously (21, 22). In total, 200 patients in the subacute phase of ischemic or hemorrhagic stroke (5–45 days after the index event) were recruited at 7 rehabilitation centers in and around Berlin, Germany, between 2013 and 2017. Patients were randomized 1:1 to receive either 4 weeks of treadmill-based, aerobic physical fitness training or a 4-week relaxation program in addition to standard rehabilitation therapy. For inclusion into the trial, patients had to be at least 18 years of age, present with a baseline Barthel Index (BI) of 65 or less and be capable of participating in an aerobic fitness program. In the clinical setting, the BI is a widely used tool to measure impairments in activities of daily living (23). Patients were excluded if they needed assistance in walking before stroke, were unable to sit unsupported for at least 30 s or had severe psychiatric or cardiac comorbidities. A full list of inclusion and exclusion criteria to the PHYS-STROKE study can be found in Supplementary Table 1. If patients met the eligibility criteria, written informed consent was obtained. During the intervention, all additional therapy sessions were recorded and therapists and patients were instructed to note the time of additional therapy they received outside of their participation in the PHYS-STROKE study (22). No standard treatment policy was set for the post-intervention period (22). Aerobic fitness training was performed using treadmill-based training to achieve a target heart rate using the formula “180 min years of age”. A period of 4 weeks was chosen to ensure that intervention session took place during the in-patient stay at the rehabilitation centers (21). A subsample of patients were additionally enrolled in the accompanying observational biomarker study BAPTISe (“Biomarkers And Perfusion – Training-Induced Changes After Stroke”) and received additional blood biomarker measurements and MRI scans before and after the intervention (24). PHYS-STROKE participants with at least baseline inflammatory marker measurements were included in this analysis. Ethical approval was obtained from the institutional review board of Charité–Universitätsmedizin Berlin (PHYS-STROKE: EA1/138/13; BAPTISe: NCT01954797).

Clinical follow-up visits took place at study enrollment (baseline), after the 4-week trial intervention period (v1), as well as 3- (v2) and 6-months (v3) after stroke. Screening of patients was performed by the trial physician of the respective site. Study outcomes were assessed and documented by trained study assessors at each visit. Study assessors and the trial statistician were blinded to the intervention. Detailed descriptions of outcome parameters were reported previously (21, 22). For this project, functional outcome was analyzed using the modified Rankin Scale (mRS), BI and maximal walking speed (MWS) at 3 and 6 months after stroke. The mRS assesses the degree of handicap in the clinical setting (25).

Fasting blood samples were retrieved at the recruiting rehabilitation center at each follow-up visit. Laboratory analyses of serum inflammatory markers were performed within 6 h at the “Labor Berlin” in Berlin, Germany. Blood levels of hs-CRP, IL-6 and TNF-alpha were determined using solid-phase, chemiluminescent immunometric assays (IMMULITE® 1,000, Siemens Healthcare Diagnostics). Measurement of fibrinogen levels was performed based on the Clauss method in citrated blood plasma by using the HemosIL® Q.F.A. Thrombin (Bovine) kit, Instrumentation Laboratories. Standard laboratory reference values were used to define elevated serum levels for each inflammatory marker: <3.0 mg/l for hs-CRP, <3.8 ng/l for IL-6, <8.1 pg/ml TNF-alpha, and 2–4 g/l for fibrinogen.

Primary research questions of this project were the dynamics of inflammatory markers over a time course of 6 months and differences in inflammatory biomarkers between the two treatment groups. Secondary aims were the associations of inflammatory markers at baseline with functional outcome parameters at 3- and 6-months after stroke.

Using Generalized estimating equation (GEE)-models, we investigated the dynamics of inflammatory markers, as well as associations of levels of inflammatory markers over time with respective patient characteristics and cerebrovascular risk factors. GEE-models allow estimating associations for repeated measurements (26). We used an exchangeable correlation matrix and calculated models for each biomarker individually. For GEE-models, we used levels of inflammatory markers over time (from baseline to v3) as dependent variables. As independent variables we chose a list of clinically relevant baseline parameters and cerebrovascular risk factors: center, visits/time, age, sex, intervention group, TOAST-classification, baseline NIHSS, smoking, atrial fibrillation, diabetes mellitus, arterial hypertension, cerebro- and cardiovascular disease. Cerebrovascular disease was defined as previous stroke and/or transient ischemic attack (TIA). Cardiovascular disease was defined as history of chronic heart disease (CHD), peripheral artery disease (PAD) and/or myocardial infarction (MI).

We used multivariate logistic regression models to analyze associations between levels of inflammatory biomarkers at baseline and 3 months after stroke and functional outcome parameters up to 6 months. Functional outcome parameters (mRS, BS and MWS) were dichotomized by the median and used as dependent variables in the multivariate models. As independent variables, we used absolute levels of single inflammatory markers (as continuous variables). Furthermore, we added sex, National Institutes of Health Stroke Scale (NIHSS) scores at baseline and age, as well as arterial hypertension (p ≤ 0.1 in univariate analyses) to our models. Multivariate models were calculated using backward selection.

For all statistical analyses we used SPSS Version 25 and 27. Non-normally distributed data were log-transformed. P-values ≤ 0.05 are considered significant. Due to the exploratory character of the study, no Bonferroni correction has been performed.

Between September 2013 and April 2017, 200 participants were enrolled in the PHYS-STROKE study of which 105 (53%) were randomized in the fitness and 95 (48%) in the relaxation group. The mean age of the study cohort was 69 years (SD ± 12) and 119 (60%) of our participants were male. One hundred eighty-one (90.5%) participants suffered from ischemic stroke, 19 (9.5%) from hemorrhagic stroke. At study enrollment, our cohort presented with a median NIHSS of 8 (IQR: 5–12), a median mRS of 4 (IQR: 4.00–4.00), a median BI of 50 (IQR: 35–60) and a median MWS of 0.30 m/s (IQR: 0.13–0.66 m/s). No significant differences between the two groups regarding the baseline characteristics were determined. Table 1 shows a detailed list of baseline characteristics of the study cohort.

The median time from stroke onset to the start of intervention was 28 days (IQR: 17–40). At baseline, the mean levels of all investigated inflammatory markers were above the clinical cut-offs. The mean serum concentrations of inflammatory markers showed a falling tendency over the 6-month observation period. The mean values of hs-CRP, IL-6, TNF-alpha, and fibrinogen over a time course of 6 months are depicted in Figure 1. In GEE-models, hs-CRP, IL-6, and fibrinogen were significantly lower at all follow-up visits when compared to baseline visits (p < 0.01 for all three inflammatory markers and all time points from v1 to v3 as compared to baseline). TNF-alpha was elevated post-stroke and declined over the study period; however, no statistically significant decline was observed (v1: Coef. 0.01; v2: Coef. −0.001; v3: Coef. −0.02; p-values between 0.09 and 0.9).

GEE models demonstrated that elevating hs-CRP (Coef. 0.01, p = 0.03), IL-6 (Coef. 0.01, p < 0.01) and TNF-alpha (Coef. 0.004, p < 0.01) levels over time correlated with increasing age. Additionally, levels of inflammatory markers over time were associated with prevalent cardiovascular risk factors at baseline: participants with pre-existing arterial hypertension had significantly increased levels of hs-CRP (Coef. 0.23, p = 0.04) and smokers had significantly elevated hs-CRP (Coef. 0.18, p = 0.04) and IL-6 (Coef. 0.09, p = 0.04) levels over the course of 6 months after stroke. Participants presenting with a history of cardiovascular, but not cerebrovascular, disease had significantly increased TNF-alpha levels over time (Coef. 0.07, p = 0.02), compared to participants without cardiovascular disease. Fibrinogen levels over time did not show any associations with investigated risk factors in GEE-models. Table 2 shows an overview of the results of the GEE-models.

The dynamics of inflammatory biomarkers over the 6-month observation period separated by treatment groups are presented in Figure 2. In GEE-models, absolute levels of investigated inflammatory markers over a time course of 6 months did not significantly differ between the two treatment groups of PHYS-STROKE (hs-CRP: Coef. −0.04; IL-6: Coef. 0.004; TNF-alpha: Coef. −0.02; fibrinogen: Coef. −0.03; p-values between 0.29 and 0.91).

In univariate models, increased baseline levels of all investigated markers were associated with worse functional outcome measurements at 3 or 6 months after stroke. Median splits of functional outcome parameters and the results of univariate analyses can be found in the Supplementary Material (Supplementary Tables 3, 4). Table 3 shows an overview of all results from multivariate logistic regression models. In multivariate logistic regression analyses, increased levels of IL-6 and fibrinogen at baseline were associated with a lower MWS (MWS ≤ 0.57 m/s) 6-months after stroke, independently of age, sex, baseline NIHSS and arterial hypertension [OR 0.34, 95% CI (0.13, 0.88); OR 0.70, 95% CI (0.51, 0.96), respectively]. Elevated baseline IL-6 and fibrinogen were also independently associated with a worse mRS (mRS ≥ 3) at 6 months after stroke [OR 3.02, 95% CI (1.01, 9.08); OR 1.59, 95% CI (1.08, 2.35), respectively]. In addition, elevated baseline levels of fibrinogen were associated with a lower BI (BI ≤ 90) 6 months post-stroke [OR 0.66, 95% CI (0.47, 0.94)]. Elevated fibrinogen at 3 months was independently associated with a worse outcome on all clinical scales at 3 and 6 months [mRS: OR 1.63, 95% CI (1.12, 2.38) and OR 1.75, 95% CI (1.14, 2.70); BI: OR 0.62, 95% CI (0.42, 0.92) and OR 0.59, 95% CI (0.34, 0.88); MWS: OR 0.65, 95% CI (0.45, 0.93) and OR 0.61, 95% CI (0.40, 0.91)]. Elevated hs-CRP at 3 months was associated with a higher mRS (mRS ≥ 3) at 6 months [OR 1.89, 95% CI (1.02, 3.51)] but no independent associations of baseline hs-CRP with outcome were identified. TNF-alpha did not show any independent associations with investigated outcome parameters at follow-up.

Hs-CRP is the marker that has received the most attention over the past decades when it comes to inflammatory biomarkers in stroke research. Many studies have underlined an elevation of hs-CRP after stroke and an association of higher levels with an unfavorable outcome (12). We found elevated hs-CRP after stroke followed by a significant decrease up to 6-months after the event. These findings are in line with previous studies describing an elevation of a panel of inflammatory markers in stroke patients (6, 7). However, in contrast to the literature, we could not show independent associations of hs-CRP levels with functional outcome parameters. One possible explanation is that the time point of blood draws differs from previous studies. Most studies focused on CRP blood levels in the acute phase after stroke or on measuring peak levels of CRP (12, 27, 28). Moreover, post-stroke infections are associated with higher levels of CRP, especially in the early phase after stroke (29). In our study, baseline visits varied from five to 45 (median 28 days) days post-stroke and acute phase reactions (including levels of CRP) most likely have already decreased.

There is growing evidence that elevated blood levels of IL-6 in acute stroke are associated with stroke lesion volume, stroke severity, post-stroke infection as well as worse short- and long-term outcome and death (1, 13–15, 30–32). Whiteley et al. underlined that adding blood values of IL-6 and NTproBNP to a validated score including age and NIHSS could improve outcome prediction (1, 14). Nevertheless, the impact was not substantial enough for this model to be useful in clinical practice. In contrast, the Linz Stroke Study claimed that the combination of several inflammatory markers can be a useful approach to predict post-stroke outcome in a clinical setting (15). Mouse models, however, showed a substantial role of IL-6 for angiogenesis after middle cerebral artery occlusion (17). No favorable effects of increasing IL-6 levels were stated in this analysis. Instead, we provide further evidence that upregulated IL-6 after stroke is correlated with worse functional outcome. We can specify the potential advantages of measuring IL-6 for prediction of impaired outcome, yet its clinical significance needs to be further validated.

There is good evidence that fibrinogen is involved in cardiovascular disease and negatively associated with clinical outcome (33–36). Evidence about the role of fibrinogen is still inconsistent and various studies were not able to show an independent impact on outcome (37). Moreover, adding fibrinogen to a model of age and NIHSS did not improve outcome prediction in stroke patients (38). Del Zoppo et al. showed that stroke patients with initial hyperfibrinogenemia had a worse outcome up to 90 days (39). Our study contributes further evidence of understanding the dynamics of fibrinogen after stroke and the advantages of measuring fibrinogen levels in clinical stroke management. In our cohort, fibrinogen levels remained elevated in the subacute phase of stroke and associations between higher inflammatory marker levels and impaired outcome were most profound for fibrinogen (Table 3). In addition, in contrast to other inflammatory markers, levels of fibrinogen over time were not associated with (common) pre-existing comorbidities, which makes it a more reliable marker to measure in the clinical setting.

Studies on TNF-alpha and stroke have yielded inconsistent results. Zaremba et al. indicated that elevated levels of TNF-alpha within 24 h of stroke-onset (in cerebrospinal fluid and serum) are correlated with lower BI scores up to 2 weeks (40). In contrast to these findings, Vila et al. were unable to show a significant correlation between raised serum values of TNF-alpha on admission and early neurological worsening up until 48 h after stroke (41). Our study expanded the knowledge to the subacute phase after stroke and could demonstrate that levels of TNF-alpha showed no dynamic change over the course of 6 months after stroke. We were not able to detect a specific expression pattern of TNF-alpha in the blood of patients after stroke. Moreover, we could not find any independent associations of serum TNF-alpha levels and functional outcome.

Exercise potentially downregulates inflammatory activity in the long-term but only few interventional studies focus on its chronic effects on inflammatory markers. So far, study results either remain inconsistent or cannot show any favorable effects (42–44). In our cohort, fitness training did not have a substantial effect neither on levels of inflammatory biomarkers nor on clinical outcome. Furthermore, no significant effects of fitness training on primary outcome (MWS or BI) were observed in the main PHYS-STROKE study analyses (21). A possible reason for our lack of evidence might be that our aerobic fitness training sessions were not long and/or intense enough to initiate anti-inflammatory processes. Training-induced anti-inflammatory effects might be present only after longer, more intense training sessions. Given our study population with functional impairment post-stroke, longer training interventions could not have been performed.

We investigated levels of inflammatory biomarkers within the PHYS-STROKE study, representing a multicenter, high quality trial. To our knowledge, most studies investigating post-stroke inflammatory markers focused on blood levels in the acute phase of stroke and only a few provide serial long-term measurements. In this study, however, we collected blood samples at four defined time-points until 6 months after stroke. We were able to depict the progression of inflammatory markers more precisely, even beyond the acute phase, covering the early and late subacute as well as early chronic phase. Additionally, we were able to examine a cohort with relatively severe impairments after stroke.

Certainly, this study has some limitations. Firstly, inclusion and baseline visit vary from a period of five to 45 days after the index stroke. This results in variances in the timing of the first intervention, as well as the post-intervention visit (v1). Blood levels of inflammatory markers might therefor vary between participants, with participants included at a later time-point post-stroke showing decreased inflammatory activity. Secondly, the effect that aerobic fitness training has on inflammatory markers might differ with different time points of training-initiation, resulting in us finding no significant effects in our cohort. Thirdly, we did not adapt for possible differences in activities outside the study intervention, as well as activities post-intervention. That could have influenced our study results. Moreover, we did not exclude patients with a diagnosis of infection or other conditions that lead to an upregulation of blood inflammatory markers.

Our study contributes to the literature on understanding the dynamics of inflammatory processes in stroke. In association with clinical outcome, the observed fluctuations can help to comprehend the role of inflammatory processes post-stroke. Including the measurement of inflammatory markers in routine stroke management can help identify patients that are at risk for long-term functional impairment. Models of outcome prediction including blood levels of inflammatory markers (single markers as well as a panel of markers) need to be validated in large, independent cohort studies. To understand the impact of fitness training on inflammation in stroke patients, researchers must first identify the ideal training program for counteracting inflammation. This research will support the important goal of identifying treatment options to blunt the neurotoxic effects of post-stroke inflammation in clinical practice. It remains unclear whether targeted anti-inflammatory treatment for stroke patients could prevent long-term stroke deterioration and this work is certainly deserving further research.