Differential expression of serum biomarkers in hemodialysis patients with mild cognitive decline: A prospective single-center cohort study

Studies suggest that hemodialysis patients are at a higher risk for cognitive decline than healthy individuals; however, underlying mechanisms have not been fully elucidated. We aimed to investigate the roles of serum biomarkers, such as brain-derived neurotrophic factor (BDNF), inflammatory cytokines, fibroblast growth factor (FGF)-23 and its co-receptor α-klotho and platelet (PLT) count in mild cognitive decline (MCD) of patients undergoing hemodialysis in this prospective cohort study. Serum levels of BDNF, tumour necrosis factor (TNF)-α, interleukin (IL)-6 and the number of PLT were significantly altered in the MCD group compared with those in healthy controls (HCs) or those with normal cognitive function (NCF). Although serum α-klotho and FGF-23 levels were significantly altered in the MCD group, there were no statistical differences between the MCD and NCF groups. Serum BDNF levels and PLT counts were significantly correlated with cognitive test scores. Receiver operating characteristic (ROC) curves demonstrated that BDNF and PLT were potential biomarkers for improved MCD diagnosis in patients with hemodialysis. These findings suggest that hemodialysis-related MCD is associated with altered BDNF, TNF-α and IL-6 levels as well as PLT counts and that serum BDNF levels and PLT counts are potential biomarkers for hemodialysis-related MCD diagnosis.

Interestingly, we found that some non-elderly patients with normal cognitive function (NCF) before hemodialysis developed mild cognitive decline (MCD) after several years of hemodialysis. In addition, a cross-sectional cohort study including 108 maintenance hemodialysis patients reported that approximately 25% of individuals were cognitively declined 11 . These findings highlight the striking deleterious effects of hemodialysis on cognitive function of patients.
Hemodialysis is an independent risk factor for MCD and seriously affects the quality of life in patients with CKD, also leading to prolonged hospitalisation and increased risk of death 12,13 . However, the mechanisms underlying hemodialysis-related MCD have not yet been fully elucidated. A clinical study by Drew et al. 14 included 263 hemodialysis patients indicated that serum fibroblast growth factor (FGF)-23 was highly associated with poor performance on a cognition test. Other than FGF-23, a limited number of studies investigated the role of other serum biomarkers in the pathogenesis of hemodialysis-related MCH. We therefore aimed to assess whether other serum biomarkers were involved in the pathogenesis of hemodialysis-related MCH. Specifically, we compared the levels of brain-derived neurotrophic factor (BDNF), α-klotho and its coreceptor FGF23, tumour necrosis factor (TNF)-α, interleukin (IL)-6 and platelet (PLT) number among hemodialysis patients with MCD or normal cognitive function (NCF) and healthy control (HC) subjects. We also evaluated correlations between cognition test scores and serum biomarkers and performed receiving operating characteristic (ROC) curves to assess whether certain serum biomarkers could predict and diagnose hemodialysis-related MCD.

Results
General clinical characteristics of patients receiving hemodialysis. Table 1 shows the summary of the clinical characteristics of the study patients including age, sex, routine blood test values and biomedical parameters. Briefly, haemoglobin, haematocrit, red blood cell count, blood urea nitrogen, creatinine and albumin were significantly different among the MCD, NCF and HC groups, whereas age, sex, white blood cell (WBC) count, dialysis period, serum calcium, serum phosphate, parathyroid hormone, dry body weight or education history were not different among the three groups.
Cognitive function is worse in hemodialysis patients with MCD than those with NCF. CKD patients undergoing hemodialysis were divided based on their performance on two cognitive function tests: the Montreal Cognitive Assessment (MoCA) (Fig. 1A) and the Mini-Mental State Examination (MMSE) (Fig. 1B). We found that MoCA scores were significantly lower in the MCD group than in the HC and NCF groups, whereas no Serum PLT count and biomarker levels. Serum BDNF levels were significantly lower in both the MCD and NCF groups than the HC group. Intriguingly, there was a significant decrease in serum BDNF levels in the MCD group compared with the NCF group ( Fig. 2A). FGF-23 and α-klotho levels were significantly different among the three groups. Specifically, serum α-klotho levels were significantly down-regulated in both the MCD and NCF groups, compared with the HC subjects (Fig. 2B). In contrast, serum FGF-23 levels were significantly increased in both the MCD and NCF groups compared with the HC group (Fig. 2C). However, serum α-klotho and FGF-23 levels failed to show a significant change between the MCD and NCF groups. In the present study, we also measured serum levels of several inflammatory cytokines. We found that both TNF-α and IL-6 levels were significantly higher in the MCD group than the HC group, whereas they were both significantly lower in the NCF group compared to the MCD group. Interestingly, when compared with the HC group, TNF-α but not IL-6  Correlation of MoCA score with serum biomarkers. We next investigated the potential correlations among serum biomarkers and the MoCA scores. Our analysis revealed that BDNF, α-klotho and PLT count were positively correlated with the MoCA score (Figs 3A,B,F), whereas TNF-α and IL-6 were negatively correlated with the MoCA score ( Fig. 3D,E). Interestingly, several lines of evidence suggest that FGF-23 plays a key role in the molecular mechanisms of CKD and hemodialysis as well as the associated cognitive dysfunction. However, in the present study, we failed to show a significant correlation between FGF-23 and the MoCA score ( Fig. 3C).
Correlation of MMSE score with serum biomarkers. Our analysis also revealed that BDNF and PLT count were independently correlated with the MMSE score (Fig. 4A,F); however, other biomarkers including α-klotho, FGF-23, TNF-α and IL-6 did not correlate with the MMSE score (Fig 4B-E).

Discussion
Hemodialysis is of critical importance to patients with CKD 15 . However, there is limited information on its potential deleterious effects on cognitive function. In the present study, we found that hemodialysis-related MCD was likely associated with changes in levels of serum biomarkers including BDNF, TNF-α and IL-6 as well as blood PLT counts. We did not find, however, any differences in serum levels of α-klotho or FGF-23 between the NCF and MCF groups. BDNF is a protein that synthesised and widely distributed in the central nervous system (CNS) which plays an important role in the survival, differentiation and growth of neurons during CNS development 16,17 . Our and other studies suggest that patients with BDNF deficit are prone to neurological disorders and that clinical symptoms are significantly improved after amelioration of BDNF loss 18,19 . In the present study, we found that the NCF and MCD groups exhibited significantly decreased serum BDNF levels. Furthermore, a significant difference in serum BDNF levels observed between the MCD and NCF groups lend support that hemodialysis-related cognitive decline is associated with BDNF loss, consistent with a previous clinical study by Zoladz et al. 20 . Although the correlation between BDNF and MMSE score was not significant (p = 0.06), serum BDNF level was likely associated with the MoCA score. Furthermore, the ROC curve analysis revealed BDNF as a potential biomarker for the diagnosis of hemodialysis-induced MCD. In contrast, Shin et al. 21 demonstrated that plasma BDNF level was significantly higher in hemodialysis patients than that of HC individuals. The reasons underlying this discrepancy, however, are not yet clear, and further studies on the role of BDNF in hemodialysis-related cognitive decline are required.
Alpha-klotho and FGF-23 have been widely reported to play important roles in a number of complications caused by hemodialysis, particularly in disorders of calcium and phosphorus metabolism 22,23 . Alpha-klotho, an anti-ageing protein, has been widely reported to be down-regulated in Alzheimer's disease, an ageing-related  disease characterized by cognitive decline 24,25 . Conversely, therapeutic strategies aimed at increasing CNS levels of α-klotho might reverse neurodegenerative processes 26 . However, our results failed to show a significant alteration in serum α-klotho levels in MCD patients compared with the NCF patients, although α-klotho was significantly decreased in CKD patients after hemodialysis. Furthermore, we previously reported the correlation between serum FGF-23 and Pi levels and that hemodiafiltration might be an effective method for clearing serum FGF-23 in hemodialysis patients with hyperphosphatemia 27 . FGF-23 was recently shown to not only play an important role in the molecular mechanisms underlying CKD and hemodialysis 28,29 but also exert physiological functions in neuronal morphology and synaptic density 30 . In the present study, changes in FGF-23 levels were comparable to those of α-klotho, and no differences were observed between patients with cognitive decline and those with no decline. However, Drew et al. 14 demonstrated that FGF-23 was negatively correlated with cognitive scores. They showed that FGF-23 was significantly up-regulated in patients with cognitive impairment. The discrepancies between their findings and our results might be explained by two reasons. First, patient enrollment criteria of the two studies are distinct. While Drew et al. 14 included hemodialysis patients with cognitive impairment, the current study investigated hemodialysis patients with the milder form of cognitive decline. Second, detection methods used for biomarkers are different. While Drew et al. 14 measured the C-terminal fragment of FGF-23, the assay we used measured the full-length FGF-23. Based on our findings, we suggest that hemodialysis can cause abnormal expression of serum FGF-23 and α-klotho but that hemodialysis-related MCD might not be associated with the changes in levels of FGF-23 and α-klotho.
Inflammatory cytokines play key roles in the pathogenesis and therapeutic mechanisms of CKD, hemodialysis-related side effects and brain diseases [31][32][33] . Montinaro et al. 13 suggested that an increase in IL-6, an inflammatory cytokine, was linked to the presence of emotional symptoms. Furthermore, cognitive behavioural therapy significantly decreased the levels of inflammatory cytokines such as high-sensitivity C-reactive protein and IL-18 in 72 hemodialysis patients with sleep disorders 34 . In the present study, we demonstrated that TNF-α and IL-6 were significantly increased in hemodialysis patients and that their levels were significant different between the NCF and MCD groups. Moreover, MoCA but not MMSE was negatively correlated with serum levels of TNF-α and IL-6. This discrepancy might be explained by several factors. First, the MoCA has a wider range of measurements to assess cognitive function compared with the MMSE. The MMSE is suitable only for the detection of cognitive decline with impaired memory and language, and memory and language function account for only 60% of the cognitive function 35 . Second, the MoCA is more sensitive than the MMSE for assessing cognitive function. The MoCA is more likely to detect early-stage MCD and is more suitable for population screening 36 . Third, the MoCA is more rational in design and superior for the diagnosis of MCD. MCD is mainly characterised by delayed memory decline, which is detected more accurately with the MoCA 37 . Therefore, these findings demonstrate that hemodialysis-related MCD is accompanied by changes in serum levels of TNF-α and IL-6. However, the correlation and ROC analyses indicated that these two biomarkers might not be well suited for the diagnosis of MCD in CKD patients undergoing hemodialysis. Additional, detailed studies on the specific role of inflammatory cytokines in the pathogenesis of hemodialysis-related MCD are thus necessary.
PLTs are bioactive small pieces of cytoplasm that break off from the cytoplasmic lysis of mature bone marrow megakaryocytes 38 . Both mean PLT volume and PLT distribution width were reported to be decreased in patients with mild cognitive impairment and Alzheimer's disease 39 . In addition, PLTs were previously shown to be exhausted after hemodialysis because of repeated hemodialysis stimulation and recurrent release of PLT degranulation products 40 . The results of the current study showed that blood PLT counts were significantly decreased in hemodialysis patients with MCD than that of healthy control individuals and hemodialysis patients with NCF. Furthermore, PLT counts were positively correlated with the MoCA and MMSE scores. Interestingly, substantial amounts of BDNF are expressed in PLTs, which contain 50-to 100-fold higher BDNF levels compared with the brain, and BDNF is released after PLT activation [31][32][33][34][35][36][37][38][39][40][41][42][43] . It is therefore likely that BDNF levels are closely related to PLT counts and function. The findings of the current study implicate both BDNF levels and PLT counts in hemodialysis-related MCD, suggesting that serum BDNF and PLT counts should be considered as objective indicators to aid in the diagnosis of hemodialysis-related MCD.
In conclusion, the present study revealed that abnormalities in serum levels of BDNF, TNF-α and IL-6 as well as PLT counts, but not α-klotho or FGF-23, might be contributing to hemodialysis-related MCD and that serum BDNF level and PLT counts might act as biomarkers for the diagnosis of hemodialysis-related MCD. Further detailed studies on the role of BDNF and PLTs in hemodialysis and its adverse effects are necessary.  Table 2. Evaluation of α-klotho, FGF-23, IL-6, TNF-α, PLT and BDNF as potential serum biomarkers for the diagnosis of hemodialysis-induced mild cognitive decline. BDNF: brain-derived neurotrophic factor; FGF-23: fibroblast growth factor-23; IL-6: interleukin-6; MCD: mild cognitive decline; PLT: platelet; ROC: receiver operating characteristic; TNF-α: tumour necrosis factor-α. Treatment. All patients enrolled in this study underwent blood purification three times a week. Treatment regimen was hemodialysis twice a week combined with haemodiafiltration once a week or hemodialysis once a week combined with haemodiafiltration twice week. Hemodialysis was performed using the Behrin Low-Flux Poly (B. Braun Diacap LOPS15, Germany), and the haemodiafiltration was treated by a Braun high-throughput polysulfone membrane dialyser (B. Braun Diacap HIPS15). Each session was about four hours, and low-molecular-weight heparin was used for anticoagulation. Dialysis solutions were bicarbonate with an average blood flow velocity of 200-280 ml/min, and dialysis flow rate was 500 ml/min. Displacement volume using post-replacement was calculated by approximately 30% of ultrafiltration flow rate.
Cognitive function tests. MMSE is a widely used cognitive screening test. Items in the MMSE address orientation, memory, recall, attention and naming objects, following verbal and written commands, writing a sentence and copying a figure, to reach a score range from 0 to 30. An MMSE score <27 is considered to indicate MCD.
MoCA was a recently developed test to address the shortcomings of MMSE in detecting mild cognitive impairment. MoCA has additional, more complex tasks including executive function. Items address orientation, drawing figures, processing speed, naming objects, memory, recall, attention, vigilance, repetition, verbal fluency and abstraction. MoCA score range is the same as that of MMSE (0-30). A MoCA score ≥26 is considered to indicate normal cognition. Additionally, MoCA adds one point for those whose educational level is 12 or fewer years. Laboratory tests. Routine blood tests were measured using an automated five-class blood cell analyser electrical impedance device (Sysmex Corporation, Japan). Blood urea nitrogen, creatinine, calcium, phosphorus and albumin were detected by an automated biochemical analyser (Leadman Biochemical, China). Parathyroid hormone measurement was achieved using chemiluminescence (Beckman, US), whereas serum α-klotho levels were measured using an enzyme-linked immunosorbent assay (Immuno-Biological Laboratories, Japan). Serum FGF-23, IL-6 and TNF-α levels were measured by a Luminex 200 protein liquid chip platform (Millipore, US) according to the manufacturer's instructions.
Statistical analysis. Data were expressed as means ± standard error of the mean (S.E.M.). Statistical analyses were performed using SPSS 13.0 (SPSS, Chicago, IL, USA). Comparisons between groups were performed by one-way analysis of variance followed by post hoc Tukey test, Student's t test or χ 2 test. Correlation analysis was conducted using Pearson's product-moment coefficient. The diagnostic cut-off values, sensitivity, specificity and accuracy were determined by ROC curve analysis. A p value < 0.05 was considered to indicate a statistically significant difference.