Pre-stroke physical activity and Val66Met (rs6265) genotype of BDNF gene correlate with the post-stroke cognitive outcome: a prospective cohort study Authors: Dariusz Kotlega, Barbara Peda, Agnieszka Zembron-Lacny, Katarzyna Baldy-Chudzik, Edyta Wawrzyniak-Gramacka, Malgorzata Szczuko Published: 2026 ABSTRACT Pre-stroke physical activity may protect against post-stroke cognitive impairment (PSCI) via neuroplastic mechanisms including BDNF signalling. However, evidence in stroke survivors and the modifying role of the BDNF Val66Met polymorphism remain limited. In this prospective single-centre cohort, 97 patients with ischaemic stroke underwent detailed neuropsychological assessment during index hospitalisation and at 6 month follow-up. Pre-stroke activity was quantified as weekly MET-minutes (IPAQ scale). Serum BDNF and Val66Met genotype were measured. Associations between METs and cognitive outcomes (baseline, follow-up) were assessed with Spearman correlations and adjusted multivariable linear regression. Higher pre-stroke METs correlated with superior baseline performance across global cognition (MoCA), verbal learning/recall (CVLT indices), attention and executive measures. After adjustment, CVLT long-delay free recall (LDFR) and TMTA remained significant (β for CVLT LDFR = 0.000087; β for TMTA = − 0.000336 per MET-min/week, p < 0.05). At 6 months several correlations indicated better performance with higher METs (VFT letter K, CVLT LDFR-sten, VFT fruits and vegetables, CVLT SDCR-sten, TMTA, TMTB), but did not remain statistically significant after FDR correction. The follow-up METs were associated with CVLT SDCR in adjusted models (β = 0.00011, p < 0.05). Mean serum BDNF did not correlate with METs (mean 27,261 ± 7967 pg/mL) and did not differ from controls. Val/Val (GG) carriers of Val66Met genotype of BDNF gene performed better cognitively than Met allele carriers, but the genotype did not correlate with serum BDNF level. No evidence of mediation, modification, or direct BDNF association was found in the tested cognition models. Greater pre-stroke physical activity is associated with better early cognitive outcomes, whereas 6 month findings remain exploratory. The BDNF Val66Met polymorphism also correlates with distinct cognitive domains. However, neither factor enhances peripheral BDNF levels suggesting the polymorphism affects BDNF activity rather than quantity. Evaluating these factors together highlights how modifiable lifestyle and non-modifiable genetics independently shape different facets of post-stroke cognitive reserve. Introduction Ischaemic stroke is a leading cause of disability in adults. It is associated with a wide range of post-stroke symptoms that adversely affect quality of life, including speech difficulties, imbalance, falls, sarcopenia, pain, fatigue and cognitive symptoms. Post-stroke cognitive impairment (PSCI) is one of the most common sequelae and occurs in up to 64% of stroke survivors (Saunders et al. 2014). Even low-intensity physical exercise has been shown to exert beneficial effects on global cognition, memory and executive function across different populations (Singh et al. 2025). Exercise after stroke primarily targets general recovery. It helps reduce cardiovascular risk and improves general mobility. Cognitive benefits are an important, yet secondary, outcome of these activities. Available data also indicate that physical activity after stroke is associated with enhanced cognitive performance (Cumming et al. 2012). Accordingly, it has been recommended that stroke survivors engage in a structured exercise programme at least three times per week for 30–60 min per session, amounting to approximately 180 min per week, to achieve better cognitive outcomes (Li et al. 2024). Early cognitive impairment after stroke contributes to reduced levels of physical exercise at six months post-event, which may reflect the fact that cognitive function is required for engagement in most activities, including physical exercise (Viktorisson et al. 2021). Each one-point decrease in Montreal Cognitive Assessment (MoCA) score after stroke has been linked to greater non-adherence to physical activity recommendations (Luzum et al. 2023). While most studies focus on the relationship between post-stroke physical activity and cognitive outcomes, there is also limited evidence indicating that the level of pre-stroke physical activity correlates with post-stroke cognitive performance (Reinholdsson et al. 2022; Wong et al. 2016; Damsbo et al. 2020; Bovim et al. 2019; Zhu et al. 2025). The mechanisms underlying the beneficial effects of physical activity on cognitive outcomes are thought to involve increased cerebral oxygen supply and cerebral perfusion, which stimulate neurogenesis, angiogenesis and synaptic plasticity. In addition, exercise is associated with a reduction in inflammatory processes, mediated by increased expression of neurotrophic factors, reduced oxidative stress and decreased β-amyloid formation (Luzum et al. 2023; De Assis et al. 2024, Singh et al. 2025). Ischemic stroke causes neuroinflammation and secondary neuronal injury. Physical activity can help reduce these harmful processes. It improves brain perfusion and stimulates neuroplasticity. These mechanisms help counteract post-stroke cognitive impairment. The pathophysiology of post-stroke cognitive impairment (PSCI) is primarily related to focal structural brain lesions caused by infarction, making stroke location a critical determinant of PSCI risk (Chen et al. 2023; Schellhorn et al. 2021). In addition to this macroscopic damage, secondary neuroinflammatory responses significantly exacerbate neuronal injury. These pro-inflammatory processes are mediated by cytokines and immune cells infiltrating the blood-brain barrier (BBB), together with increased oxidative stress (Zhao et al. 2022; Xia et al. 2023). Brain-derived neurotrophic factor (BDNF) is a member of the neurotrophin family and plays a key role in axonal regeneration, cognitive modulation, angiogenesis, synaptogenesis and neuroplasticity. It is highly expressed in the hippocampus and exerts its effects primarily through activation of the tropomyosin receptor kinase B (TrkB). BDNF has been identified as a potential mediator of exercise-induced improvements in cognitive function, as both BDNF signalling and physical activity enhance long-term potentiation and thereby support neuroplastic processes (Romero Garavito et al. 2025). Genetic variability within the BDNF gene is well established, and several common single nucleotide polymorphisms (SNPs) have demonstrated clinical relevance (Kotlega et al. 2017). Depending on the specific SNP variant, BDNF synthesis in response to physical exercise may differ (Bugge Kambestad et al. 2023), and certain variants have been shown to augment memory performance following exercise interventions (Piepmeier et al. 2020). One of the most extensively studied polymorphisms, Val66Met (rs6265), has been associated with an increased risk of conversion from mild cognitive impairment (MCI) to Alzheimer’s disease (Xie et al. 2017). However, evidence regarding the role of the Val66Met polymorphism in modulating the relationship between physical exercise and specific cognitive domains, such as attention, information processing speed, executive function, memory and global cognition, remains limited and inconsistent (Liu et al. 2020). Furthermore, the Val66Met polymorphism has been linked to cognitive performance and hippocampal volume, although most studies have been conducted in non-stroke populations, and data in stroke survivors are scarce (Dincheva et al. 2012). To our knowledge, no studies have specifically examined the impact of physical exercise on cognitive performance in stroke survivors in relation to the Val66Met SNP of the BDNF gene. The primary aim of this study was to investigate the association between pre-stroke physical activity (quantified as weekly MET-minutes) and post-stroke cognitive performance at baseline and follow-up in patients with ischemic stroke. Secondary aims were to assess the relationship between pre-stroke physical activity and serum BDNF concentration, and whether the BDNF Val66Met polymorphism modifies the associations of physical activity with both serum BDNF and cognitive outcomes. We also intended to independently assess the relationship between pre-stroke physical activity, serum BDNF concentration, and the BDNF Val66Met polymorphism in relation to cognitive outcomes. We hypothesised that higher pre-stroke physical activity would be associated with better global and domain-specific cognition after stroke, that greater activity would correlate with higher serum BDNF, and that Met-carrier status would attenuate the positive association between activity and BDNF/cognitive improvement. We also aimed at evaluating the effects of Val66Met genotype of BDNF gene on cognitive functions in stroke survivors, irrespectively from the impact on BDNF. Although physical activity and the BDNF Val66Met genotype represent distinct variables, a modifiable behavioural factor and a non-modifiable genetic factor, respectively, they conceptually converge on the BDNF signalling pathway and the broader concept of cognitive reserve. Evaluating both within a single cohort provides a comprehensive framework for understanding how different elements of neuroplasticity independently or synergistically shape post-stroke cognitive trajectories. Materials and methods Subjects In this prospective study, we enrolled 97 patients with a diagnosis of ischaemic stroke. All participants were recruited during their hospital stay in the Neurology Department of a district hospital in Poland. The primary inclusion criterion was an acute cerebral ischaemic stroke, diagnosed on the basis of clinical presentation and confirmed by brain imaging (MRI or CT). Routine laboratory investigations and standard stroke management were provided in line with current guidelines. Stroke was defined as a clinical syndrome of sudden onset with focal or global cerebral dysfunction lasting ≥ 24 h or resulting in death, attributable to a vascular cause (Sacco et al. 2013; Adams et al. 2007). Patients were excluded if neuroimaging demonstrated intracerebral haemorrhage. Additional exclusion criteria were active malignancy, autoimmune disease, clinical signs of infection, or body temperature > 37.4 °C. Individuals who were unable to provide informed consent or to perform neuropsychological tests due to impaired consciousness, severe speech disturbance or significant muscle weakness, were also excluded. At follow-up, the number of patients available for neuropsychological reassessment ranged from 28 to 43, owing to non-attendance at scheduled visits. The control group comprised 33 stroke-free, age- and sex-matched individuals recruited from local university clubs for older adults. The study covered period between March 2018 and March 2020. The breakdown of screening process is presented in Fig. 1. The protocol of the study was approved by the Ethics Committee in Zielona Gora (decision number 08/73/2017, Feb 27, 2017). The study was conducted in accordance with the Declaration of Helsinki. All patients signed a written informed consent form to participate in the study. Fig. 1Flow chart of the study population Flow chart of the study population Neuropsychological assessment Neuropsychological testing was carried out by a trained neuropsychologist experienced in the assessment of cognitive disorders in stroke. The cognitive tests were performed during the index hospitalisation, 7 days after admission, and again at the 6 month follow-up visit. The number of patients reassessed at follow-up was lower than at baseline due to non-attendance. Participants completed the 21-item Beck Depression Inventory–II (BDI-II), rating each item from 0 to 3 to indicate how they had felt over the previous two weeks. The instrument maps onto DSM-IV diagnostic criteria for major depressive disorder; total scores span 0–63, with larger values denoting greater severity of depressive symptoms (Beck et al. 1996; Reeves et al. 2012). Cognitive functioning was assessed with a neuropsychological test battery that included the California Verbal Learning Test, Second Edition (CVLT-II), a measure of verbal learning, episodic memory, recognition memory, and attention. We analysed the following CVLT-II indices: List A, List B, short-delay free recall (SDFR), short-delay cued recall (SDCR), long-delay free recall (LDFR), long-delay cued recall (LDCR), and semantic clustering (Delis et al. 2000; Yi 2011). Raw CVLT-II scores were converted into standard ten (sten) scores for all indices. The sten is a normalized psychological scale (mean = 5.5, SD = 2) divided into ten intervals; this transformation, routinely used in clinical neuropsychology, permits comparison across individuals and aids interpretation in correlational analyses. Higher sten values indicate superior cognitive performance. Short-term verbal memory, attention, and verbal working memory were evaluated using the Forward and Backward Digit Span Tests (FDST and BDST), subtests drawn from the Wechsler Adult Intelligence Scale and the Wechsler Memory Scale (Woods et al. 2011; Fink et al. 2014). The Trail Making Test (TMT), consisting of Parts A and B, was used to evaluate psychomotor processing speed, sustained attention, visuospatial scanning, working memory, cognitive flexibility, and set-shifting. Test performance was quantified as the time required to complete each part, recorded in seconds, with longer completion times reflecting poorer cognitive performance (Talarowska et al. 2012; Cahn et al. 1995). The sten scale is a normalized psychological metric. It has a mean of 5.5 and a standard deviation of 2. A shift of 1–2 units typically reflects a clinically noticeable change in cognitive capacity. The T-score (‘ten’ scale), applied to measures such as the Trail Making Test (TMTA ten, TMTB ten), has a population mean of 50 and an SD of 10. For both standard metrics, higher values indicate superior cognitive performance, facilitating a uniform interpretation across different cognitive domains. Inhibitory control and the ability to adjust behaviour in response to changing task demands were examined using the Stroop Colour and Word Test (SCWT), also known as the Stroop Dots Trial A and B (SDT A and SDT B). An experimental Polish adaptation derived from the original version of the test was administered (Scarpina and Tagini 2017). Verbal fluency was evaluated using a Verbal Fluency Test administered under two conditions: semantic (category) fluency and phonemic (letter) fluency. Semantic fluency was assessed with the Verbal Fluency Test–Animals (VFT-A) and the Verbal Fluency Test–Fruits and Vegetables (VFT-F&V), whereas phonemic fluency was measured using the Verbal Fluency Test–Letter “K” (VFT-K) and the Verbal Fluency Test–Letter “P” (VFT-P). A Polish adaptation of the original instrument was employed. This test battery was used to examine executive functions, expressive language abilities, memory processes, and processing speed (Cahn et al. 1995). Neuropsychological outcomes labelled with the suffix “follow up” denote measurements collected at the 6 month follow-up assessment. With the exception of the Trail Making Test (TMT) and Stroop Dots Test (SDT), higher scores across all instruments indicated superior performance within the respective cognitive domains. The Montreal Cognitive Assessment (MoCA) was administered as a brief and clinically practical screening instrument to assess global cognitive functioning, encompassing attention, memory, orientation, language, visuospatial skills, and abstract reasoning (Zietemann et al. 2018). Pre-stroke physical activity was assessed with the use of short form of International Physical Activity Questionnaire (IPAQ) scale. This self-reported questionnaire consists of brief items assessing the frequency and duration of walking, moderate, and vigorous physical activity, and sitting time over the past seven days, providing an overall physical activity estimate suitable for surveillance. Results are calculated into total energy expenditure per week (MET min/week) (Balboa-Castillo et al. 2023). Whenever possible, the physical activity history was corroborated by a family member or caregiver present during the hospital stay. BDNF and Val66Met polymorphism analysis The plasma BDNF concentration was evaluated by using the R&D Systems ELISA kit (USA). The intra-assay coefficient of variation (CV) was < 5%, and the detection limit was estimated at 20 pg/mL for the used kit. Genomic DNA was isolated from blood using the GeneJET Genomic DNA Purification Kit (Thermo Scientific) according to the manufacturer’s protocol for mammalian cells. DNA concentration and purity were assessed spectrophotometrically using a NanoPhotometer (Implen). Purified DNA was stored at − 80 °C until further analysis. BDNF Val66Met genotyping was performed using the ARMS (tetra primer amplified refractory mutation system) method (ARMS-PCR). The method utilizes the specificity of four primers in a single PCR reaction, which allows for the amplification of two allele-specific products (253 and 201 bp) and the entire region (401 bp) as an internal control, as previously described (Sheikh et al. 2010). The first set of primers, P1 and P2 [P1 (forward): 5′-CCT ACA GTT CCA CCA GGT GAG AAG AGT G-3′; P2 (reverse): 5′- TCA TGG ACA TGT TTG CAG CAT CTA GGT A-3′] amplifies the 401-bp rs6265 region of the BDNF gene. The pair in the second set served to identify G/A substitution, where primer P3 is specific for the G allele [P3: 5′-CTG GTC CTC ATC CAA CAG CTC TTC TAT AAC-3′] and primer P4 is specific for the A allele [P4: 5′-ATC ATT GGC TGA CAC TTT CGA ACC CA-3′]. The PCR amplification mixture in a volume of 25 µL contained: buffer solution for (Thermo Scientific, Waltham, MA, USA), 2.5 mM MgSO4 (Promega, Madison, WI, USA), 0.5 mM of each dNTP (Promega), 0.2 µM of each of four primers (Genomed, Poland), 1 U of Platinum Taq DNA Polymerase (Thermo Scientific), and 3 µL of DNA template. The PCR products were separated by electrophoresis on a 3% agarose gel and stained with ethidium bromide. Statistical methods Continuous variables were summarized as mean ± standard deviation (SD) or median (interquartile range, IQR), depending on their distribution, whereas categorical variables were expressed as counts and percentages. Normality was assessed using the Shapiro-Wilk test and visual inspection of distributions. Between-group comparisons of continuous variables were performed using Welch’s t-test when assumptions for approximate normality were met and the Mann-Whitney U test otherwise. Categorical variables were compared using the chi-square test or Fisher’s exact test, as appropriate. The BDNF Val66Met polymorphism was coded consistently across cohorts and analyzed both as a three-category variable (GG, GA, AA) and as a binary variable contrasting Val/Val homozygotes (GG) with Met-carriers (GA + AA). Genotype distributions were compared between patient and control cohorts using 3 × 2 and 2 × 2 contingency tables. Demographic characteristics, including age and sex, were also compared between cohorts. Associations between physical activity (MET minutes/week) and serum BDNF concentration were assessed using Spearman’s rank correlation, both in the overall sample and after stratification by genotype. To investigate possible effect modification by genotype, regression models included interaction terms between physical activity and BDNF genotype. Multivariable regression models were then used to evaluate the independent associations of baseline physical activity and BDNF-related variables with cognitive outcomes, adjusting for clinically relevant covariates selected on the basis of clinical judgment and preliminary univariable analyses. Neuropsychological test results were summarized separately for baseline and follow-up assessments as mean ± SD. Because follow-up availability differed across tests, analyses involving longitudinal cognitive outcomes were interpreted with consideration of variable sample size by assessment. All tests were two-tailed, and p < 0.05 was considered statistically significant. Statistical analyses were performed using Python 3.11.6. Results The demographic characteristic is presented in Table 1. Table 1Demographic characteristic of study groupVariableValue n Age (years)63.5 ± 12.197Sex (Male), n (%)43 (44.3)97BMI29 ± 5.197CRP (mg/L)3.08 ± 4.397LDL (mg/dL)111.63 ± 45.6197HDL (mg/dL)52.55 ± 15.6297Hypertension, n (%)84 (85.6%)97Diabetes/IFG, n (%)47 (48.4%)97Current Smoking, n (%)40 (41.2%)97Atrial Fibrillation, n (%)10 (10.3%)97Previous Stroke/TIA, n (%)11 (11.3%)97Baseline NIHSS score5.05 ± 2.0697 Table 1Demographic characteristic of study groupVariableValue n Age (years)63.5 ± 12.197Sex (Male), n (%)43 (44.3)97BMI29 ± 5.197CRP (mg/L)3.08 ± 4.397LDL (mg/dL)111.63 ± 45.6197HDL (mg/dL)52.55 ± 15.6297Hypertension, n (%)84 (85.6%)97Diabetes/IFG, n (%)47 (48.4%)97Current Smoking, n (%)40 (41.2%)97Atrial Fibrillation, n (%)10 (10.3%)97Previous Stroke/TIA, n (%)11 (11.3%)97Baseline NIHSS score5.05 ± 2.0697 Demographic characteristic of study group Baseline cognitive assessments Analysis of baseline cognitive performance in relation to pre-stroke physical activity, quantified as weekly MET values, demonstrated several significant associations. Higher levels of physical activity were positively correlated with performance on the CVLT List A, CVLT SDFR (sten), Forward Digit Span Test (FDST), Verbal Fluency Test (VFT) for fruits and vegetables, VFT letter K, Backward Digit Span Test (BDST), Montreal Cognitive Assessment (MoCA), and CVLT LDFR (sten). Conversely, higher physical activity levels were negatively correlated with completion times on the Trail Making Test Part B (TMT-B), Stroop Test Part 2, Trail Making Test Part A (sten), and Trail Making Test Part A completion time, indicating better performance with increasing physical activity. No statistically significant associations were observed for the remaining cognitive measures (Table 2). Table 2Association between baseline cognitive functions and MET value (min/week) in study groupOutcome n Mean ± SDSpearman’s rhop- valueFDR-adjusted pCVLT LDFR (sten)636.21 ± 2.240.4490.000220.00516MOCA9724.31 ± 3.470.3350.000780.00682TMTA time9754.99 ± 25.14− 0.3320.000890.00682BDST973.57 ± 1.510.3110.001940.01115TMTA ten9754.73 ± 12.75− 0.2930.003550.01635Stroop2 time97164.55 ± 51.82− 0.2680.007840.02829VFT-Letter K9710.04 ± 4.490.2650.008610.02829VFT-fruits and vegetables9718.72 ± 4.860.25930.010330.02971TMTB time97135.66 ± 71.64− 0.2470.014510.03707FDST975.07 ± 1.090.2150.03451NSCVLT SDFR (sten)635.87 ± 1.830.2630.03698NSCVLT List A9645.54 ± 9.610.2050.04475NSVFT-Letter P9710.30 ± 3.840.195NSNSCVLT List A (sten)635.94 ± 1.840.209NSNSCVLT LDCR (sten)625.68 ± 2.140.204NSNSCVLT SDCR (sten)635.78 ± 2.160.196NSNSStroop1 time9795.11 ± 27.53− 0.155NSNSVFT-animals9720.63 ± 5.650.145NSNSTMTB ten9761.05 ± 15.98− 0.142NSNSBeck978.96 ± 6.18− 0.108NSNSCVLT list B965.18 ± 1.980.0796NSNSCVLT list B (sten)635.86 ± 2.21− 0.0713NSNSCVLT semantic clustering (sten)965.23 ± 2.230.0106NSNS Table 2Association between baseline cognitive functions and MET value (min/week) in study groupOutcome n Mean ± SDSpearman’s rhop- valueFDR-adjusted pCVLT LDFR (sten)636.21 ± 2.240.4490.000220.00516MOCA9724.31 ± 3.470.3350.000780.00682TMTA time9754.99 ± 25.14− 0.3320.000890.00682BDST973.57 ± 1.510.3110.001940.01115TMTA ten9754.73 ± 12.75− 0.2930.003550.01635Stroop2 time97164.55 ± 51.82− 0.2680.007840.02829VFT-Letter K9710.04 ± 4.490.2650.008610.02829VFT-fruits and vegetables9718.72 ± 4.860.25930.010330.02971TMTB time97135.66 ± 71.64− 0.2470.014510.03707FDST975.07 ± 1.090.2150.03451NSCVLT SDFR (sten)635.87 ± 1.830.2630.03698NSCVLT List A9645.54 ± 9.610.2050.04475NSVFT-Letter P9710.30 ± 3.840.195NSNSCVLT List A (sten)635.94 ± 1.840.209NSNSCVLT LDCR (sten)625.68 ± 2.140.204NSNSCVLT SDCR (sten)635.78 ± 2.160.196NSNSStroop1 time9795.11 ± 27.53− 0.155NSNSVFT-animals9720.63 ± 5.650.145NSNSTMTB ten9761.05 ± 15.98− 0.142NSNSBeck978.96 ± 6.18− 0.108NSNSCVLT list B965.18 ± 1.980.0796NSNSCVLT list B (sten)635.86 ± 2.21− 0.0713NSNSCVLT semantic clustering (sten)965.23 ± 2.230.0106NSNS Association between baseline cognitive functions and MET value (min/week) in study group Most of the results remained statistically significant after FDR Benjamini-Hochberg correction. To determine appropriate covariates for the multivariable models, we conducted a preliminary univariable analysis to examine two-sided associations between demographic or clinical factors and physical activity. Factors demonstrating significant correlations with physical activity included age (rho = − 0.380, p = 0.00012), ischaemic heart disease (rho = − 0.306, p = 0.0023), congestive heart failure (rho = − 0.296, p = 0.0032), and hypertension (rho = − 0.218, p = 0.0319), indicating that older age and the presence of these cardiovascular conditions were associated with lower pre-stroke MET values. Other variables did not show significant correlations with physical activity, including sex (p = 0.309), diabetes (p = 0.497), smoking (p = 0.248), and BMI (p = 0.639). However, to ensure rigorous control for confounding, we forced fundamental demographic characteristics (age, sex, BMI) and critical vascular risk factors (diabetes, smoking) into the final multivariable models alongside the data-driven covariates, based on their well-established clinical relevance to post-stroke cognitive outcomes. After this comprehensive multivariable adjustment, CVLT LDFR (sten) remained significantly associated with weekly MET values (p < 0.05; β = 0.000087; 95% CI 0.000015 to 0.000160; R2 = 0.188). This finding indicates that each additional 1000 MET-min/week was associated with an approximate 0.087-point increase in CVLT LDFR (sten) score. Trail Making Test Part A (sten) also remained significantly associated with physical activity, showing an inverse relationship (partial p < 0.05; β = − 0.000336; 95% CI − 0.000722 to − 0.000010; R2 = 0.219). This finding indicates that each additional 1000 MET-min/week was associated with an approximate 0.366-point improvement in TMTA ten score. The corresponding regression plots are presented in Figs. 2 and 3. Fig. 2The scatter plot of association between baseline CVLT LDFR (sten) and MET value The scatter plot of association between baseline CVLT LDFR (sten) and MET value Fig. 3The scatter plot of association between baseline TMTA (ten) and MET value The scatter plot of association between baseline TMTA (ten) and MET value Follow up cognitive assessments Table 3 summarises the associations between follow-up cognitive test performance and weekly MET values. Higher levels of physical activity were positively correlated with performance on the Verbal Fluency Test (VFT) letter K, CVLT LDFR (sten), VFT fruits and vegetables, and CVLT SDCR (sten). In contrast, higher MET values were negatively correlated with completion times on the Trail Making Test Parts A and B, indicating better performance with greater physical activity. No statistically significant associations were observed for the remaining follow-up cognitive measures. Table 3Association between follow up cognitive functions and MET value (min/week) in study groupOutcome n Mean ± SDSpearman’s rhop-valueFDR-adjusted pTMTB time follow up40120.90 ± 70.75− 0.3750.0170NSTMTA time follow up4047.50 ± 24.52− 0.3630.0213NSCVLT SDCR (sten) follow up266.31 ± 2.110.4340.0267NSVFT-Fruits and Vegetables follow up3318.76 ± 3.350.3740.0319NSCVLT LDFR (sten) follow up266.46 ± 2.340.3960.0454NSVFT-letter K follow up3411.82 ± 4.280.3430.0469NSVFT-animals follow up3322.00 ± 5.840.329NSNSCVLT SDFR (sten) follow up266.04 ± 1.910.369NSNSVFT-letter P follow up3410.65 ± 4.550.286NSNSBeck follow up4010.05 ± 7.04− 0.261NSNSCVLT LDCR (sten) follow up267.88 ± 9.590.313NSNSCVLT list A follow up2650.35 ± 7.770.299NSNSCVLT list A (sten) follow up266.69 ± 1.720.297NSNSTMTA (ten) follow up4048.15 ± 15.88− 0.208NSNSStroop1 time follow up3284.84 ± 28.26− 0.199NSNSBDST follow up384.13 ± 1.460.172NSNSFDST follow up395.46 ± 1.140.153NSNSCVLT semantic clustering (sten) follow up265.85 ± 2.650.152NSNSStroop2 time follow up31139.10 ± 47.39− 0.115NSNSCVLT list B (sten) follow up266.50 ± 2.37− 0.123NSNSCVLT list B follow up265.62 ± 2.14− 0.110NSNSTMTB (ten) follow up4057.82 ± 17.41− 0.067NSNS Table 3Association between follow up cognitive functions and MET value (min/week) in study groupOutcome n Mean ± SDSpearman’s rhop-valueFDR-adjusted pTMTB time follow up40120.90 ± 70.75− 0.3750.0170NSTMTA time follow up4047.50 ± 24.52− 0.3630.0213NSCVLT SDCR (sten) follow up266.31 ± 2.110.4340.0267NSVFT-Fruits and Vegetables follow up3318.76 ± 3.350.3740.0319NSCVLT LDFR (sten) follow up266.46 ± 2.340.3960.0454NSVFT-letter K follow up3411.82 ± 4.280.3430.0469NSVFT-animals follow up3322.00 ± 5.840.329NSNSCVLT SDFR (sten) follow up266.04 ± 1.910.369NSNSVFT-letter P follow up3410.65 ± 4.550.286NSNSBeck follow up4010.05 ± 7.04− 0.261NSNSCVLT LDCR (sten) follow up267.88 ± 9.590.313NSNSCVLT list A follow up2650.35 ± 7.770.299NSNSCVLT list A (sten) follow up266.69 ± 1.720.297NSNSTMTA (ten) follow up4048.15 ± 15.88− 0.208NSNSStroop1 time follow up3284.84 ± 28.26− 0.199NSNSBDST follow up384.13 ± 1.460.172NSNSFDST follow up395.46 ± 1.140.153NSNSCVLT semantic clustering (sten) follow up265.85 ± 2.650.152NSNSStroop2 time follow up31139.10 ± 47.39− 0.115NSNSCVLT list B (sten) follow up266.50 ± 2.37− 0.123NSNSCVLT list B follow up265.62 ± 2.14− 0.110NSNSTMTB (ten) follow up4057.82 ± 17.41− 0.067NSNS Association between follow up cognitive functions and MET value (min/week) in study group None of the results remained statistically significant after FDR Benjamini-Hochberg correction. Multivariable analysis confirmed significant association between weekly MET values and follow-up performance on the CVLT subtest, such as CVLT SDCR (sten), CVLT SDFR (sten) (p < 0.05; β = 0.00011; 95% CI 0.0000082 to 0.0002011; R2 = 0.585). Associations between BDNF, Val66Met of BDNF gene, and MET values The mean serum BDNF concentration in the study group was 27,261.43 ± 7966.91 pg/mL (n = 97). No statistically significant correlation was observed between weekly MET values and serum BDNF concentrations (p > 0.05). In unadjusted analyses, baseline serum BDNF was not significantly associated with cognitive performance after correction for multiple testing. The strongest nominal correlation was observed for CVLT LDFR (sten) at follow-up (rho = 0.512, p = 0.0074), but this did not remain significant after FDR correction (q = 0.264). Other nominal associations, including Stroop1 time, VFT-Fruits and Vegetables, and VFT-Letter P, were likewise not significant after correction. The genotype distribution of the BDNF Val66Met polymorphism was as follows: GG in 69 participants (71.1%), GA in 23 participants (23.7%), and AA in 5 participants (5.1%). Serum BDNF concentrations did not differ significantly across genotype groups (Mann–Whitney U test, p > 0.05). Spearman rank correlation analyses between MET values and serum BDNF concentrations stratified by genotype revealed no significant associations in the GG (ρ = 0.078, p > 0.05) or GA groups (ρ = − 0.375, p > 0.05). A strong negative correlation was observed in the AA genotype group (ρ = − 0.90, p < 0.05); however, this finding should be interpreted with caution due to the very small number of participants with this genotype (n = 5). Serum BDNF concentrations were comparable between the study group (27,261.43 ± 7966.91 pg/mL; n = 97) and the control group (26,350 ± 4968 pg/mL; n = 33), with no statistically significant difference detected (Mann–Whitney U test, p = 0.33). Similarly, the distribution of BDNF Val66Met genotypes did not differ between the study and control groups (χ2 test, p = 0.397). A significant difference in age distribution was observed between the study and control groups (63.56 ± 12.13 years vs. 71.18 ± 4.69 years, respectively; Welch’s t-test, p < 0.01). In contrast, sex distribution did not differ between groups, with 43 males in the study group and 14 males in the control group (χ2 test, p = 1.0). The analysis of the impact of the BDNF Val66Met polymorphism on cognitive performance is presented in Table 4. Significant associations were observed for the Montreal Cognitive Assessment (MoCA), CVLT semantic clustering (sten), and CVLT List A. Comparisons between GG (Val/Val) genotype carriers and combined Met-allele carriers (GA + AA) were performed using the Mann–Whitney U test. Individuals with the GG (Val/Val) genotype demonstrated significantly better performance on these measures than Met-allele carriers. No statistically significant differences were observed for the remaining cognitive tests (p > 0.05). Table 4The comparison between Val66Met genotypes of BDNF gene in regards to cognitive functionsVariableGenotype n Mean ± SDp-valueMOCAGG6924.75 ± 3.360.04MOCAAA + GA2823.21 ± 3.54CVLT semantic clustering (sten)GG685.62 ± 2.520.0024CVLT semantic clustering (sten)AA + GA284.28 ± 1.90CVLT List AGG6846.79 ± 10.010.0296CVLT List AAA + GA2842.5 ± 7.94 Table 4The comparison between Val66Met genotypes of BDNF gene in regards to cognitive functionsVariableGenotype n Mean ± SDp-valueMOCAGG6924.75 ± 3.360.04MOCAAA + GA2823.21 ± 3.54CVLT semantic clustering (sten)GG685.62 ± 2.520.0024CVLT semantic clustering (sten)AA + GA284.28 ± 1.90CVLT List AGG6846.79 ± 10.010.0296CVLT List AAA + GA2842.5 ± 7.94 The comparison between Val66Met genotypes of BDNF gene in regards to cognitive functions In the adjusted linear regression model, only CVLT semantic clustering (sten) remained statistically significant, with AA/GA carriers scoring on average 1.32 sten units lower than GG homozygotes (β = − 1.32; 95% CI − 2.29 to − 0.34; p = 0.008). In the analysis of the impact of the BDNF Val66Met polymorphism on follow-up cognitive performance, no statistically significant associations were observed. Comparisons between GG homozygotes and combined Met-allele carriers (AA + GA) using the Mann–Whitney U test did not reveal any significant differences (p > 0.05). To assess whether BDNF mediated the association between physical activity and cognitive performance, we fitted covariate-adjusted linear models examining the relationships between MET and BDNF, BDNF and cognitive outcomes, MET and cognitive outcomes before and after inclusion of BDNF. No evidence of mediation was observed. Specifically, MET was not associated with BDNF in any of the tested models, including for CVLT LDFR (sten) (β = 0.033, p = 0.818), TMTA ten (β = − 0.009, p = 0.937), and CVLT SDCR (sten) at follow-up (β = 0.050, p = 0.833). Likewise, BDNF was not independently associated with cognitive outcomes after adjustment for MET and covariates, with p-values of 0.930, 0.568, and 0.538, respectively. The estimated indirect effects were negligible (a × b = − 0.000000103, 0.000000884 and 0.00000165, respectively). To evaluate effect modification by the BDNF Val66Met polymorphism, genotype was entered into adjusted linear models together with MET for the interaction analysis. No significant interaction was observed for CVLT LDFR (sten) (interaction β = 0.000026, p = 0.738), TMTA ten (interaction β = 0.000401, p = 0.277), or CVLT SDCR (sten) at follow-up (interaction β = − 0.000040, p = 0.737), indicating that the association between physical activity and cognition did not differ by genotype group. In additional direct analyses, serum BDNF itself was not independently associated with cognitive performance. In adjusted models, BDNF was not related to CVLT LDFR (sten) (β = − 0.0000004, 95% CI − 0.000074 to 0.000073, p = 0.991, R2 = 0.097), TMTA ten (β = − 0.000091, 95% CI − 0.000422 to 0.000240, p = 0.587, R2 = 0.184), or CVLT SDCR (sten) at follow-up (β = 0.000040, 95% CI − 0.000086 to 0.000165, p = 0.512, R2 = 0.454). Discussion In this prospective cohort study of stroke patients, we examined the associations between pre-stroke physical activity, quantified as weekly MET values, and post-stroke cognitive performance. Baseline cognitive assessments revealed positive correlations between physical activity and CVLT List A, CVLT SDFR (sten), Forward Digit Span Test (FDST), Verbal Fluency Test (VFT) for fruits and vegetables, VFT letter K, Backward Digit Span Test (BDST), Montreal Cognitive Assessment (MoCA), and CVLT LDFR (sten). Conversely, negative correlations were observed with Trail Making Test Part B (TMT-B) time, Stroop Test Part 2 time, TMTA (sten) and TMTA completion time. No statistically significant associations were detected for the remaining cognitive measures. These findings indicate that higher levels of self-reported pre-stroke physical activity are associated with better cognitive performance in the early post-stroke phase, spanning multiple domains including global cognition, memory (verbal learning and recall), attention, working memory, processing speed, and executive function. After multivariate adjustment, the associations with CVLT LDFR (sten) and TMTA (sten) remained statistically significant, highlighting their independent relationship with pre-stroke physical activity. We also examined the associations between pre-stroke physical activity and cognitive performance at 6 months post-stroke. Positive correlations were observed for VFT letter K, CVLT LDFR (sten), VFT fruits and vegetables, and CVLT SDCR (sten), whereas negative correlations were found with TMTA and TMTB completion times, indicating better performance with higher physical activity levels. No associations remained statistically significant after FDR Benjamini-Hochberg correction which indicates these results should be interpreted as exploratory, hypothesis-generating in nature. Multivariable analysis confirmed significant associations for CVLT SDCR (sten) only, reinforcing the relationship between higher pre-stroke physical activity and superior cognitive outcomes. Overall, these findings are consistent with the baseline assessments, supporting a robust trend linking greater physical activity to improved post-stroke cognitive performance. In a previous study evaluating the impact of physical activity in the 48 h preceding stroke, a positive correlation with cognitive outcome was also observed. However, cognitive performance in that study was assessed solely using the MoCA test (Reinholdsson et al. 2022). Another study assessed the impact of the total amount of physical activity during the year preceding stroke and found that higher activity levels were associated with a reduced risk of developing post-stroke dementia within 3 to 6 months after the event (Wong et al. 2016). An analysis of physical activity during the seven days preceding stroke demonstrated that more active participants exhibited better cognitive performance. Cognitive function was assessed using the Symbol Digit Modalities Test (SDMT) and the Mini-Mental State Examination (MMSE) at one- and 6 months post-stroke (Damsbo et al. 2020). Available evidence demonstrates considerable heterogeneity in methodology across studies. In the present study, we employed the validated International Physical Activity Questionnaire (IPAQ) alongside an extensive battery of neuropsychological assessments conducted by an experienced, specialised clinical psychologist. To date, this represents the most comprehensive evaluation of post-stroke cognitive performance in relation to physical activity. Our findings are further supported by recent large-scale longitudinal data demonstrating that pre-stroke physical activity is a critical determinant of post-stroke resilience. A cohort study of over 12,000 participants revealed that higher pre-stroke activity levels significantly reduce long-term functional and ADL (Activities of Daily Living) limitations, which heavily depend on preserved cognitive capacity (van Allen et al. 2024). Although no significant correlation was observed between weekly MET values and serum BDNF concentrations in our cohort, physical activity is known to attenuate post-stroke cognitive impairment through multiple broader mechanisms. These include enhancing cerebral blood flow, promoting general neuroplasticity, and reducing neuroinflammation and oxidative stress (Shui et al. 2024; Zhu et al. 2025; Souza et al. 2022; De Assis et al. 2024). Given the lack of association with BDNF in our study, it is plausible that the observed cognitive benefits of pre-stroke physical activity rely primarily on these alternative, complex neuroprotective pathways rather than strictly on peripheral BDNF elevation. In addition to cognitive assessments, we measured baseline serum BDNF levels and genotyped participants for the Val66Met single nucleotide polymorphism of the BDNF gene. No significant association was observed between pre-stroke MET values and BDNF concentrations, and BDNF levels did not differ according to genotype. Rank correlation analyses suggested that physical activity might influence BDNF in a genotype-dependent manner. However, the only statistically significant correlation was observed in participants with the AA genotype of Val66Met, a group comprising only five individuals, precluding definitive conclusions. We also found no differences in BDNF levels or genotype distribution between the study and control groups. While physical activity is frequently hypothesized to increase peripheral BDNF, recent evidence in post-stroke cohorts reveals significant variability. Importantly, recent studies demonstrate that aerobic exercise interventions may not induce significant changes in serum BDNF levels in stroke survivors, irrespective of the exercise intensity or the BDNF Val66Met genotype (De Las Heras et al. 2025). This directly aligns with our findings and is consistent with other reports showing no significant effect of exercise on circulating BDNF in chronic stroke (Maguire et al. 2023). Together, these results underscore the complexity of peripheral neurotrophic signalling and suggest that cognitive improvements may rely on other neuroplastic pathways. Other studies indicate that high-intensity physical activity can elevate BDNF levels in stroke survivors, whereas moderate-intensity exercise may not produce the same effect (Boyne et al. 2019; Hsu et al. 2021). In our study, BDNF Val/Val (GG) carriers demonstrated superior cognitive performance compared with Met-allele carriers (AA + GA), particularly in global cognition (MoCA), verbal learning (CVLT List A), and semantic organization during learning (CVLT semantic clustering). Notably, no BDNF genotype was associated with higher serum BDNF levels, suggesting that the Val66Met polymorphism may influence BDNF activity rather than its circulating concentration, as proposed by previous studies (Brown et al. 2020; De Assis et al. 2022). The precise role of the Val66Met genotype in modulating the neurotrophic effects of BDNF warrants further investigation. Evidence from other populations supports these observations. In healthy individuals, Met allele carriers exhibited reduced stability of information stored in iconic memory compared with Val/Val homozygotes, indicating that Val/Val individuals retain information in iconic memory for longer periods (Beste et al. 2011). Among patients with bipolar disorder, Met carriers performed significantly worse than Val/Val individuals on the Wisconsin Card Sorting Test (WCST) (Rybakowski et al. 2006). Similarly, in a community sample of elderly White individuals, Val/Val homozygotes outperformed both Val/Met heterozygotes and Met/Met homozygotes on delayed recall and an alphabet-coding task (ACT), a measure of processing speed (Miyajima et al. 2008). Another study reported that Met carriers performed worse than controls on delayed recall tasks, with performance declining further as recall intervals increased (Goldberg et al. 2008). In post-stroke cohorts, lower cognitive performance has been consistently associated with the presence of the Met allele of the BDNF Val66Met polymorphism, aligning with our findings (Han et al. 2020, Kim et al. 2012). A key distinction of our study is that it represents, to date, the most extensive analysis of post-stroke cognitive outcomes in relation to BDNF genotype. Notably, as expected for distinct biological modifiers, physical activity and the Val66Met genotype exhibited associations with different cognitive domains. Physical activity primarily correlated with processing speed and delayed recall (TMTA, CVLT LDFR), while the Val66Met polymorphism was more closely associated with global cognition and semantic organization (MoCA, CVLT semantic clustering). This divergence suggests that while both factors conceptually converge on the BDNF-mediated neuroplasticity pathway, their clinical manifestations in post-stroke recovery differ. Presenting these factors within a single analytical framework provides a more holistic view of post-stroke cognitive reserve, illustrating how lifestyle and genetics independently contribute to different facets of the cognitive phenotype. We also performed interactions analysis between physical activity, BDNF, BDNF genotype and cognitive outcome. We found no evidence that serum BDNF mediated the relationship between physical activity and cognition. Across the tested models, physical activity was not associated with BDNF, and BDNF was not independently associated with cognitive outcomes after adjustment for physical activity and clinical covariates. In addition, the BDNF Val66Met polymorphism, analyzed as GG versus AA + GA, did not modify the association between physical activity and cognition, as no significant MET and genotype interactions were observed. Unadjusted direct analysis between cognitive functions and BDNF showed that higher levels are associated with better cognitive outcome. Although several nominal associations were observed for these individual neuropsychological variables, none remained significant after correction for multiple comparisons. This study has several limitations that constrain the interpretation and generalisability of the findings. The observational, single-centre design precludes causal inference and is susceptible to residual confounding despite multivariable adjustment. Important variables such as detailed premorbid cognitive status and stroke lesion location were not uniformly controlled for. Pre-stroke physical activity was assessed by retrospective self-report on short-form of IPAQ scale, which is vulnerable to recall and reporting bias and lacks the objectivity of device-based measures (e.g., accelerometer). We acknowledge this potential recall bias. Moreover, patients reported their physical activity levels retrospectively and some participants had cognitive deficits at baseline, although mild. This might affect the accuracy of the self-reported data. This scale refers to general degree of physical activity, rather than structured exercise assessments. That’s why, to some extent, results reported by other authors might not be fully interchangeable. The sample size reduces statistical power for an analysis, especially the follow-up neuropsychological reassessments, which included only 28–43 subjects. The AA genotype subgroup comprised only five individuals, which was underpowered. Peripheral serum BDNF was measured at a single time point, while circulating BDNF may not reliably reflect central nervous system levels or dynamic physical activity-related changes, and single measurements increase susceptibility to biological and assay variability. Finally, potential medication effects, rehabilitation exposures and lifestyle changes after stroke were not fully captured and may have influenced follow-up cognitive trajectories. Conclusions In conclusion, this study provides evidence that higher levels of pre-stroke physical activity are associated with more favourable cognitive performance directly after the acute phase of ischaemic stroke. Associations at the 6 month follow-up should be interpreted cautiously as exploratory due to sample size constraints and the loss of statistical significance after multiple testing correction. These baseline associations persisted after adjustment for key demographic variables, suggesting that physical activity may be a potential contributor to cognitive reserve that could help mitigate post-stroke cognitive impairment. In contrast, peripheral serum BDNF levels were not related to pre-stroke activity or cognitive outcomes, underscoring the complexity of neurotrophic signalling and the limitations of single time-point peripheral measurements as proxies for central neuroplastic processes. The observed differences in selected memory measures between BDNF Val66Met genotypes suggest a potential moderating role of genetic factors in post-stroke cognitive trajectories. By evaluating both factors within a single framework, our findings highlight how modifiable lifestyle habits and non-modifiable genetics independently shape distinct facets of post-stroke cognitive reserve. These results cautiously support the potential benefit of promoting regular physical activity as a modifiable, low-cost strategy that might enhance brain resilience before cerebrovascular events, though small effect sizes and the observational design warrant confirmation in larger, longitudinal, multicentre studies incorporating objective physical activity assessment, serial biomarker profiling and detailed neuroimaging to clarify underlying mechanisms and to determine whether targeted exercise interventions can meaningfully reduce the burden of cognitive impairment after stroke.