This study comprises a cohort of 62 post-stroke patients recruited in two periods within the prospective, observational, longitudinal, and multicenter SMARRTS study (Studying Markers of Angiogenesis and Repair during Rehabilitation Therapy after Stroke): ischemic stroke patients under IRT (between February 2014 and May 2015) (15) and both ischemic and hemorrhagic stroke patients under IRT or conventional rehabilitation therapy (between February 2017 and June 2018) from seven Spanish hospitals. Inclusion criteria were: first-ever ischemic or hemorrhagic stroke, age ≤ 75 years, modified Rankin scale (mRS) ≤ 2 before stroke and mRS post-stroke from three to five, stable medical condition, and the signature of informed consent. The exclusion criteria were: previous stroke or transient ischemic attacks, malignant infarct, global aphasia, hemorrhage from arteriovenous malformations or cerebral aneurysms, previous cognitive decline, or recent infectious, inflammatory or malignant disease.

The study was approved by all the clinical research ethics committee sites [HUVH PR(IR)317/2013-PR(IR)346/2016, PI16/00981/CEI:PI-17-056, Comité de Ética de Investigación de A Coruña-Ferrol 2017-125, Corporació Sanitària Parc Taulí de Sabadell 2017521, Hospital Universitario Politécnico La Fe 2016/0727, Euskadi PI2016168]. Healthy subjects (43) without known neurological, malignant, infectious, or inflammatory diseases were enrolled as the control cohort. All subjects signed informed consent following the Declaration of Helsinki.

STROBE guidelines for reporting observational studies were followed in this study (20).

All included patients followed a comprehensive rehabilitation program, including physiotherapy, occupational therapy, speech therapy, and/or neuropsychology, however, some centers offered intensive rehabilitation therapies whereas others followed more conventional programs. For the IRT program only subacute post-stroke patients with moderate/severe disabilities in two or more areas (gait, transfers, activities of daily living, swallowing and/or communication), with mRS previous to stroke ≤ 2 who can participate in a minimum of 3 h of comprehensive therapy per day (5 days/week) were included, according to institutional guidelines (21). When the clinical stroke condition was stable patients started mobilizations followed by a comprehensive rehabilitation program defined as IRT (≥15 h per week) or NO-IRT (conventional therapy with <15 h per week). For all centers, a physiatrist designed an initial rehabilitation program according to patients' impairments, and all patients were treated in inpatient rehabilitation units or day-hospital facilities. Rehabilitation continued until completion of a minimum of the proposed objectives or when functional stability was achieved. For further objectives, patients continued an outpatient rehabilitation program.

A total of 62 patients were initially included in the study. However, 3 of them voluntarily abandoned the study, and 6 were withdrawn from the study due to the following: decompressive cranial surgery (1), secondary aneurismal hemorrhagic stroke (1), a second ischemic event (1), or moving to other cities (3). This reduced the number of subjects in the follow-up analysis, as represented in the flow diagram in Figure 1. A baseline inclusion visit was conducted before the rehabilitation program started by an experienced physiatrist who collected demographic, clinical, and stroke-related data together with a battery of tests to assess motor and functional status (details in the Supplementary Methods). During the baseline interview, previous physical activity was considered as any cardiovascular exercise routine such as running, swimming, cycling, and obesity was calculated BMI > 29.

Follow-up visits were conducted by the same physiatrist at 1, 3, and 6 months after the start of rehabilitation including a battery of tests to assess neuro-functional status: the mRS (scores 0–6), the Granger modified Barthel Index (BI, scores 0–100) (22), the Fugl-Meyer Assessment score for the upper extremity (FMA, scores 0–66) (23), the Functional Ambulation Categories (FAC, scores 0–5), the Chedoke Arm and Hand Activity Inventory (CAHAI, scores 13–91) (24), the 10-m walk test (velocity is registered), the Medical Research Council scale (MRC, scores 0–5) of the upper and lower extremities at the proximal/distal level, and the Modified Asworth Scale (MAS, scores 0–4, including 1+) (25).

We also analyzed changes in the scores during follow-up visits vs. baseline scores to assess improvements in the neurological function: Rankin improvement was defined as a decrease of ≥1 point. For the FMA, the improvement was defined as an increase of ≥10 points, described previously as the minimal clinical important difference (26). For the Chedoke Arm and Hand Activity Inventory, an improvement was defined as an increase of ≥7 points (27). For the 10-m walk test, the walking velocity was calculated, and improvement was considered if walking velocity increased by >0.3 m/s. The FAC was categorized into three categories: cannot walk (score 0), dependent walk (scores 1–3), and independent walk (scores 4 and 5), and improvement was defined as a shift to a higher category (28). For the MRC scale, our analysis differentiated classification between normal (score 5) or impaired (scores 0–4) muscle strength.

At all visits blood sampling in serum-separating tubes was obtained, centrifuged at 1,500 rpm for 15 min, and serum stored at −80°C until use.

Serum levels of ANG were measured by ELISA (#DAN00; R&D Systems, USA) and analyzed together with previous-obtained results from cohort one using the same test (15). Briefly, 200 μl of diluted serum samples (1:200) were loaded per duplicate, and only values with a coefficient of variation < 20% were accepted for the statistical analysis. To verify low inter-plate variability (7 ELISA plates were analyzed in total) we included a commercial internal control from Sigma-Aldrich (Human serum type AB, male, from clotted, cat#H6914), and a coefficient of variation < 20% was accepted.

The SPSS 20.0 package was used for statistical analyses. The database entries were reviewed by an independent researcher. Descriptive statistics: categorical variables were reported as frequencies (percentages) and continuous variables as mean ± standard deviation (SD) or median [interquartile range (IQR)], as appropriate. Missing data were considered at random and no imputations were used. The normality assumption of quantitative variables was checked with the use of quantile-quantile (Q-Q) plots. Statistical significance was assessed by Pearson's chi-square or Fisher's exact test for categorical variables, Student's t-test for continuous variables, and the Mann-Whitney U test for functional scores and numerical variables without normal distribution. Temporal profile changes in normally distributed variables were analyzed with repeated measures ANOVA (Bonferroni post-hoc test) or the Friedman followed by Wilcoxon tests for non-normal distributions. Box plots were used to represent the temporal profile of non-normal distributed variables (functional scores) and bar-graphs showing mean with 95% confidence interval to represent normally distributed variables (temporal profile of ANG). Pearson (normal distribution) or Spearman (non-normal distribution) tests were used. Repeated measures in general linear models were conducted to assess changes in the ANG temporal profile between the two rehabilitation groups, controlling for other baseline characteristics (CAHAI). The temporal profile changes in the functional scales were analyzed with Mixed Models Analysis generated by the SAS System (Version W32_7 PRO) adjusting for the baseline scores. A p-value < 0.05 was considered statistically significant.

The baseline characteristics of the stroke cohort and controls are described in Table 1. Of note, our stroke cohort presented more men (79.7 vs. 20.3%, p < 0.01), more tobacco users (37.1 vs. 16.7%, p = 0.02), and less previous exercise (39 vs. 76.7%, p < 0.001) than the control. Importantly, our two rehabilitation groups presented similar baseline characteristics except for the previous statins medication which was more frequent in the IRT group (46.7 vs. 17.6%, p = 0.03) as shown in Table 2.

The two rehabilitation groups presented similar clinical characteristics at emergency admission: stroke characteristics, acute treatment, hospitalization regimen, or baseline functional scores (see Tables 3, 4). Significant differences were only found in the number of rehabilitation hours per week: IRT 15 (15–16) vs. NO-IRT 7.5 (5–9), p < 0.01, and in the baseline CAHAI score which was lower in the IRT group: 13 (13–16) vs. 18.5 (13–82), p = 0.02.

Both groups presented significant improvements over the 6 month follow-up period in functional and motor tests, but the CAHAI score and MAS scale only improved in the IRT group (Table 5). Regarding the impact of the rehabilitation intensity on motor and functional scores, Figure 2 shows improvements occurring earlier in the IRT group for the CAHAI, FMA, and FAC tests (p < 0.001 at 1 month), but not in the NO-IRT group, which achieved significance later at 3months (p < 0.05). Other tests presented similar profiles regardless of therapy intensity (Supplementary Figure 1). Importantly, at 3 months the IRT group achieved a full community ambulation level (0.89 m/s), but not the NO-IRT group (0.56 m/s) (Table 5). However we did not find differences overtime between rehabilitation groups in a mixed model: FAC (p = 0.86), Walking (p = 0.42), mRS (p = 0.82), BI (p = 0.41), FMA (p = 0.10), and CAHAI (p = 0.10). In this regard, the effect of time was independent of the type of therapy (p < 0.01) for all tests and did not change when adjusting for baseline mRS or Barthel.

Baseline ANG was significantly higher in strokes than in controls (520 ± 139 vs. 432 ± 155 ng/mL; p < 0.01, see Table 1), with no baseline differences observed between the rehabilitation groups (see Table 4). No correlation was observed between baseline ANG and time-to-start RHB or baseline sampling. Notably, the number of rehabilitation hours were reduced according to the individual patient's achievements as represented in Figure 3A, with a substantial switch from IRT to NO-IRT over time. For this reason, the ANG level was analyzed only up to the 3rd month. In all strokes the ANG temporal profile showed significant differences (p < 0.001, see Supplementary Figure 2), being elevated at 1 month vs. controls (p = 0.036). Regarding the type of therapy, we found a significant interaction with ANG levels over time (p = 0.030, see Figure 3B). The ANG temporal profile in the NO-IRT and IRT groups showed differences (p < 0.01, see Figure 3C) in opposite directions since ANG increased after 1 month of IRT (p < 0.05 vs. baseline) but it decreased in the NO-IRT group (p < 0.01 vs. 3rd month).

Finally, we examined the relationship between outcome scores improvements from admission to 6 months follow–up with ANG, but we could not confirm a predictive value of ANG at any of the tested time points (data not shown).