Concordance of Alzheimer’s Disease-Related Biomarkers Between Intraventricular and Lumbar Cerebrospinal Fluid in Idiopathic Normal Pressure Hydrocephalus

INTRODUCTION

Idiopathic normal pressure hydrocephalus (iNPH) is characterized by a triad of gait disturbance, urinary incontinence, and progressive dementia, together with communicating hydrocephalus [1, 2]. It is observed in geriatric patients with a prevalence of 5.9–8.9% in those aged 80 years and older [3, 4]. The natural course of iNPH includes progressive worsening of the symptoms and delay in treatment leads to meager outcome after cerebrospinal fluid (CSF) shunt surgery [5, 6]. A positive clinical outcome by modified Rankin scale and by iNPH scale is achieved in 69% and 84% cases following surgery [7]. The concomitant neurodegenerative diseases are commonly comorbid to iNPH with the highest prevalence of Alzheimer’s disease (AD) [8].

The CSF based amyloid-β 1-42 (Aβ₄₂), total tau (T-tau), and phosphorylated tau at threonine 181 (P-tau₁₈₁) have found their standardized role in AD diagnostics. They illustrate the brain parenchyma neurodegenerative processes of amyloid accumulation to extracellular aggregates and intracellular neurofibrillary tangle formation caused by hyperphosphorylated tau. Within iNPH patients, the disease specific pattern of these biomarkers include lower Aβ₄₂, T-tau, and P-tau₁₈₁ concentration of CSF in comparison to healthy individuals of similar age [9–12]. Moreover, low T-tau and P-tau₁₈₁ can discriminate iNPH from AD [12, 13]. Furthermore, the increased lumbar CSF (L-CSF) T-tau and P-tau₁₈₁ are suggested for predictors of shunt-non-responsive iNPH [14, 15].

Despite keen research of CSF biomarkers, the composition of CSF throughout the circulating pathways of brain ventricles, spinal cord, and cortical subarachnoid space, as well as the effect of shunt surgery, is mostly unknown. CSF is not circulating like blood and the composition of CSF proteins is considered to depend on the surrounding tissue [16]. Furthermore, varying biomarker concentrations in CSF of iNPH patients has been reported based on both the timing and location of harvesting of the sample [17]. In addition, the presence of comorbid AD has tendency to alter the composition of CSF biomarkers [17]. The amyloid precursor protein derived proteins Aβ₃₈, Aβ₄₀, Aβ₄₂, and soluble AβPPα has been reported to be lower in ventricular CSF (V-CSF) compared to preoperative L-CSF [9, 18–20]. In contrary, the T-tau and P-tau measured higher in intraoperative V-CSF than preoperative L-CSF [9, 18–20]. With trigeminal neuralgia and tension type headache patients the similar trend for T-tau was seen; higher concentration in cisternal CSF [21]. However, the Aβ₄₂ did not differ significantly rostro-caudally [21]. When concentrations of Aβ₄₂, T-tau, and P-tau₁₈₁ were compared in post-traumatic hydrocephalus group, no significant rostro-caudal gradient were found [20]. The knowledge regarding post-shunt surgery rostro-caudal gradient with simultaneous samples of L- and V-CSF is sparse, but alterations in biomarker levels have been seen in longitudinal studies [9, 18]. These together challenged the clinical use of intraventricular and postoperativeCSF.

MATERIALS AND METHODS

Study population

In all, 138 patients from Kuopio University Hospital (KUH) region, Kuopio, Finland were diagnosed with probable iNPH by the Relkin criteria and using the KUH iNPH protocol [1, 22]. Ventriculoperitoneal CSF shunt system (Ps Medical Strata II or Miethke ProGAV) was received by all participants. The shunt surgeries were performed from 2009 until 2015 for the 41-patient cohort and from 2018 until 2021 for the 97-patient cohort (Fig. 1). The participants were evaluated at baseline and 3 months postoperatively by iNPH grading scale (Kubo scale, 0–12 points) [23]. The positive outcome was determined with 1-point or more decrease and unimproved less than 1-point decrease in the total iNPH grading scale points postoperatively. Furthermore, the 41-patient cohort was assessed repeatedly by iNPH grading scale as presented previously [17]. Prior to the CSF shunt implantation, a brain biopsy was obtained using a previously described protocol [17] and analyzed for Aβ– and tau pathology by neuropathologist.

OPEN IN VIEWER

Fig. 1

Formation of the study cohorts. Flowchart is presenting the formation of the cohorts and cerebrospinal fluid sampling. * Kuopio iNPH protocol of diagnosis is published previously [22]. iNPH, idiopathic normal pressure hydrocephalus; CSF cerebrospinal fluid.

RESULTS

Patient characteristics and biomarker concentrations of both cohorts are presented in Table 1. Longitudinal changes in CSF biomarkers of the 41-patient cohort have been reported previously [17]. Altogether, baseline NPH grading scale points were similar across the cohorts (Mean 6.0 for cohort 97 and 5.6 for cohort 41, p = 0.41). The only significant difference was seen in gait velocity as the cohort 97 had 0.2 m/s higher baseline gait velocity (p < 0.01). The male sex was more common in both cohorts (57% in cohort 97 and 61% in cohort 41) and the gender distribution was similar between the cohorts (p = 0.071). The preoperative baseline lumbar CSF Aβ₄₂, T-tau, and P-tau₁₈₁ concentrations were similar in cohorts of 41 and 97 patients (Aβ₄₂ p = 0.86, T-tau p = 0.64, and P-tau₁₈₁ p = 0.43) (data not shown).

The preoperative lumbar CSF Aβ₄₂ concentrations were 84% higher than intraoperative ventricular CSF (p < 0.0001) and the median V/L-CSF ratios (VLR) were 0.54 (Q1–Q3:0.40–0.75) (Fig. 2, Table 1). On the contrary, T-tau and P-tau₁₈₁ concentrations in preoperative lumbar CSF were 49% and 11% lower than seen in intraoperative ventricular CSF (T-tau p < 0.001, P-tau p = 0.027) and had median VLRs of 1.47 (Q1–Q3:1.14–2.68) and 1.01 (Q1–Q3:0.90–1.40) (Fig. 2, Table 1). Pooled post-shunt-surgery sample V-CSF concentrations were 12.6% (p < 0.0001), 21.4% (p < 0.0001), and 20.3% (p < 0.0001) lower than in L-CSF for Aβ₄₂, T-tau, and P-tau₁₈₁ (Table 1). The median VLRs were 0.85 for Aβ₄₂ (Q1–Q3:0.77–0.95), 0.79 for T-tau (Q1–Q3:0.72–0.92), and 0.77 for P-tau₁₈₁ (Q1–Q3:0.68–0.88) (Fig. 2).

OPEN IN VIEWER

Fig. 2A-C

Boxplots of ventricular-/lumbar CSF ratios. Box and whiskers plots are presenting ventricular-/lumbar CSF ratios of the biomarkers of Aβ₄₂ (A), T-tau (B), and P-tau₁₈₁ (C). Light gray boxplots are illustrating the ratios of intraoperative V-CSF and preoperative L-CSF. White boxplots are presenting the ratios of postoperative V- and L-CSF. Repeated V- and L-CSF samples of the follow-up were pooled per patient before the calculation of the V-/L-CSF ratios. Aβ₄₂, amyloid-β 1-42; T-tau, total tau protein; P-tau₁₈₁, phosphorylated tau at threonine 181; V-CSF, ventricular cerebrospinal fluid; L-CSF, lumbar cerebrospinal fluid.

Correlations between ventricular and lumbar CSF were examined by Spearman’s ρ (Table 2). In the cohort of 97 patients, preoperative L-CSF Aβ₄₂ (ρ= 0.54), T-tau (ρ= 0.34), and P-tau₁₈₁ (ρ= 0.55) (all p < 0.001) correlated to intraoperative V-CSF. Furthermore, strong correlations were seen for Aβ₄₂ (ρ= 0.77–0.88, mean ρ= 0.83), T-tau (ρ= 0.91–0.94, mean ρ= 0.92) and P-tau₁₈₁ (ρ= 0.94–0.96, mean ρ= 0.94) between simultaneous postoperative L- and V-CSF samples of 41-patient cohort throughout the follow-up (all p < 0.0001). In addition, no correlations were seen between the waiting time for surgery and the intraoperative ventricular CSF Aβ₄₂ (ρ= 0.01, p = 0.89), T-tau (ρ= –0.06, p = 0.60), and P-tau₁₈₁ (ρ= –0.01, p = 0.90) concentrations (data not shown).

Linear regression models were carried out to investigate the relationship for intraoperative V-CSF and preoperative L-CSF, patient age and gender (Table 3, Fig. 3A-C). Fitted model functions are yielded and presented in Tables 3 and 4 and Fig. 3. In the univariate model (p < 0.001, F = 38.8, R² = 0.29) for Aβ₄₂, L-CSF significantly predicted V-CSF (B = 0.34, C.I. 0.23–0.45, p < 0.001) (Fig. 3A). Multivariate model for V-CSF Aβ₄₂, consisted of age, gender, and L-CSF was statistically significant (p < 0.001, F = 13.1) and explained 30% of variance (R² = 0.30). L-CSF Aβ₄₂ was significant predictor (B = 0.34, C.I. 0.23–0.46, p < 0.001); however, age (B = –3.2, C.I. – 9.8–3.4, p = 0.33) and male gender (B = –0.06, C.I. –86.8–86.7, p = 0.99) were non-significant predictors of V-CSF. Due the non-normally distributed concentrations of preoperative and intraoperative CSF T-tau, logarithmic correction was carried. The univariate linear regression model (R² = 0.08, F = 7.9, p = 0.006) with Ln(T-tau L-CSF), predicted significantly (B = 0.49, C.I. 0.15–0.84, p = 0.006) Ln(T-tau V-CSF) (Fig. 3B). The multivariate regression model including age, gender, and Ln(T-tau L-CSF), was significant as well (R² = 0.09, p = 0.01). The L-CSF Ln(T-tau) was significant (B = 0.48, C.I. 0.11–0.84, p = 0.01), and both age (B = 0.00, C.I. –0.02–0.02, p = 0.96) and gender (B = 0.10, C.I. –0.13–0.34, p = 0.38) were non-significant predictors of V-CSF Ln(T-tau). The univariate model (F = 18.8, p < 0.001) of preoperative L-CSF P-tau₁₈₁ (B = 0.51, C.I. 0.27–0.74, p < 0.001), explained 17% of the V-CSF variance (R² = 0.17) (Fig. 3C). In multivariate model (R² = 0.19, F = 7.0, p < 0.001) with predicting variables of age (B = 0.08, C.I. –0.15–0.30, p = 0.51), gender (B = 2.0, C.I. –0.86–4.79, p = 0.17) and L-CSF P-tau₁₈₁ (B = 0.47, C.I. 0.23–0.71, p < 0.001) only L-CSF P-tau₁₈₁ was significant predictor for V-CSF P-tau₁₈₁.

OPEN IN VIEWER

Fig. 3A-C

Scatterplots of pre- and intraoperative Aβ₄₂, T-tau, and P-tau₁₈₁ in L- and V-CSF. Scatterplots of V- and L-CSF values of the biomarkers Aβ₄₂ (A), T-tau (B), and P-tau₁₈₁ (C) and linear trendlines are illustrating the linear dependency of intraoperative V- and preoperative L-CSF. Mean confidence intervals (95%) are drawn for linear trendlines. Regression equations of the linear univariate regression models are presented at upper right corner of the figure. T-tau values are presented at natural logarithmic scale due the non-normally distributed values. Aβ₄₂, amyloid-β 1–42; T-tau, total tau protein; P-Tau₁₈₁, hyperphosphorylated tau at threonine 181; V-CSF, ventricular cerebrospinal fluid; L-CSF lumbar cerebrospinal fluid; Ln, natural logarithm transferred variable; R², Coefficient of determination.

Furthermore, linear mixed effects modelling was performed to determine linear dependency for the postoperative V-CSF and L-CSF (Table 3, Fig. 4A-C). Fitted model equations are presented in Tables 3 and 4. For Aβ_42, L-CSF values (B = 0.75, C.I. 0.63–0.88, p < 0.001) predicted V-CSF values significantly (pseudo R² = 0.28) (Fig. 4A). In the multivariate model (pseudo R² = 0.29) including L-CSF Aβ₄₂, age, and gender, the regression equation was nearly concordant to univariate model. The L-CSF Aβ₄₂ (B = 0.71, C.I. 0.58–0.84, p < 0.001) and patient age in years (B = –9.2, C.I. –17.4––1.1, p = 0.027) were significant predictors of V-CSF Aβ₄₂ and patient gender was found to be non-significant (B = 24.3, C.I. –57.9–106.5, p = 0.55). With T-tau, postoperative V-CSF values were significantly predicted by postoperative L-CSF T-tau (B = 0.62, C.I. 0.55–0.68, p < 0.001) in the univariate model (pseudo R² = 0.32) (Fig. 4B). In the multivariate model (pseudo R² = 0.33), only L-CSF T-tau (B = 0.62, C.I. 0.55–0.69, p < 0.001) was a significant predictor of V-CSF, as the patient age (B = 0.53, C.I. –3.5–4.6, p = 0.79) and gender (B = 12.5, C.I. –28.9–53.9, p = 0.54) were non-significant. Similarly, postoperative L-CSF P-tau₁₈₁ (B = 0.74, C.I. 0.69–0.78, p < 0.001) predicted simultaneous V-CSF P-tau₁₈₁ values significantly (pseudo R² = 0.45) (Fig. 4C). Further, the age (B = –0.17, C.I. –0.50–0.16, p = 0.30) and gender (B = –2.0, C.I. –5.4–1.4, p = 0.24) were non-significant and L-CSF P-tau₁₈₁ (B = 0.74, C.I. 0.69–0.78, p < 0.001) significant predictors of V-CSF (pseudo R² = 0.46).

OPEN IN VIEWER

Fig. 4A-C

Scatterplots of postoperative Aβ₄₂, T-tau, and P-tau₁₈₁ in V- and L-CSF. Scatterplots of V- and L-CSF values of the biomarkers Aβ₄₂ (A), T-tau (B), and P-tau₁₈₁ (C) and linear trendlines are illustrating the linear dependency of postoperative V- and L-CSF. Mean confidence intervals (95%) are drawn for linear trendlines. Regression equations of the linear mixed effects models are presented at upper right corner of the figure. Aβ₄₂, amyloid-β 1-42; T-tau, total tau protein; P-Tau₁₈₁, hyperphosphorylated tau at threonine 181; V-CSF, ventricular cerebrospinal fluid; L-CSF, lumbar cerebrospinal fluid.

The in vitro experiment conducted with the protocol 1, revealed minor variation in Aβ₄₂ concentrations (Supplementary Table 1, Supplementary Figure 1A). In phases 3 and 4, the mean Aβ₄₂ values were most decreased in comparison to the baseline (Phase 3:11%, Phase 4:22%). In the protocol 2, the sample size dependent changes were seen in Aβ₄₂ concentrations. Lower concentrations (mean decrease of 2 ml samples: 13%) were seen when sample size was less than 5 ml (Supplementary Table 2, Supplementary Figure 1D). However, in larger sample sizes of 10–20 ml the difference came irrelevant in comparison to the baseline samples. In both protocols implemented, T-tau and P-tau₁₈₁ were relatively stable and showed no sample size dependent decrease (Supplementary Tables 1 and 2, Supplementary Figure 1B, C, E, F).

DISCUSSION

Here we studied the core AD biomarkers of Aβ₄₂, T-tau, and P-tau₁₈₁ in the iNPH patients CSF. This study provides a comprehensive insight to the CSF AD-core marker composition dynamics that varies by the location and harvesting moment of the sample. The key findings are the established rostro-caudal gradients and fitted linear models for Aβ₄₂, T-tau, and P-tau₁₈₁ between [1] pre- and intraoperative L- and V-CSF and [2] postoperative L-CSF and V-CSF.

We consider our results of decreased Aβ₄₂ and somewhat increased T-tau and P-tau₁₈₁ between pre- and intraoperative CSF to support the findings reported previously (Table 1) [9, 18–20]. However, the linearity between intraoperative V-CSF samples and preoperative L-CSF samples for T-tau and P-tau₁₈₁ is somewhat weaker than we expected. The reason behind the rather exponential increase of T-tau is probably a immediate trauma caused by surgical insertion of the intraventricular CSF catheter trough brain parenchyma [27, 28]. In the study obtaining brain interstitial fluid by microdialysis [29], similar pattern was seen for T-tau, as the insertion resulted high T-tau concentrations that decreased over the collection period of 24 h. Further, the studies comparing T-tau in preoperative tap-test L-CSF and in intraoperative V-CSF report 2 to 6-fold higher concentration in V-CSF [9, 19]. Other studies comparing the first and last fractions of lumbar tap-test CSF [19, 30], only found significant ratio of 1.2 between the last/first fraction of CSF T-tau [19]. However, these results do not completely exclude the chance that further draining of CSF would result similar gradients as reported in studies comparing L-CSF and intraoperative V-CSF. As expected, our P-tau₁₈₁ results had similar trend as T-tau. These findings are presenting the potential challenges in the interpretation of CSF T-tau and P-tau₁₈₁ harvested during surgical procedure.

On the other hand, we assume that a waiting time for shunt surgery can cumulate, e.g., the periventricular ischemic damage, as worse outcomes have been reported with prolonged waiting time [6], and therefore potentially cause discrepancy to interpretation of the intraoperative CSF biomarkers. However, we could not find correlation for shunt surgery waiting time and V-CSF Aβ₄₂, T-tau, or P-tau₁₈₁ measured intraoperatively. After all, this was not surprising as our median waiting time was rater short (2.0 months, interquartile range 1.1–3.5 months) in the cohort of 97 patients.

Furthermore, simultaneous postoperative V- and L-CSF biomarkers are largely unstudied scheme due to the ethically challenging study implementation. Our results suggest a transition of T-tau and P-tau₁₈₁ VLRs as the postoperative gradient is 0.77–0.79, respectively (Fig. 2). Somewhat supporting results of T-tau VLR being under 1 between postoperative shunt reservoir V-CSF and L-CSF, have been reported previously [9]. For Aβ₄₂, we observed approximating concentrations in V- and L-CSF as the VLR converted from 0.54 to 0.85, and this was mainly driven by the increased concentration of V-CSF Aβ₄₂ postoperatively. Contrary to Craven et al. [9], our Aβ₄₂ in V-CSF measured lower than in L-CSF. This difference might derive from the rather small number of patients in the postoperative CSF comparison of the previous publication. We consider this Aβ₄₂ change to represent beneficial shunt response and improved homeostasis maintenance of brain parenchyma, driven by increased Aβ₄₂ excretion to CSF. For the reason of the VLR transition to less than 1 postoperatively in T-tau and P-tau₁₈₁, we suggest the sampling modality of the shunt reservoir puncture. It can be considered as non-traumatic draining of CSF, as no direct harm is caused to brain parenchyma. Therefore, potentially more reliable results are received. Other explanations for this gradient transition seen with T-tau and P-tau₁₈₁, might be caused either by the altered CSF flow resulting from CSF shunt [31] or inhibition of the fundamental NPH pathology that is not yet completely understood. We have also reported that Aβ₄₂, T-tau, and P-tau₁₈₁ do not remain stable post-operative [17]. However, the VLRs of the parallel samples in every biomarker do maintain the ratio and rostro-caudal gradient throughout the follow-up.

Reasons behind the rostro-caudal gradient of proteins in CSF are somewhat hypothesized, and the composition of CSF is suggested to alter due the protein origin, molecular mass, and CSF-dynamic disorders. The brain parenchyma derived proteins should be enriched in ventricular, and blood derived in lumbar CSF [16]. The albumin and blood derived IgG, IgA, and IgM quantities have been reported to decrease when further draining lumbar CSF of iNPH patients [32] and in healthy controls [33]. With central nervous system specific proteoglycans, neurocan, and brevican, no significant ventriculo-lumbar gradient were seen pre- or postoperatively [34]. However, this assumption is not met completely in our results with iNPH patients, as the post-operative Aβ₄₂, T-tau, and P-tau₁₈₁ VLRs are all less than 1. Contrary, the ratios seen between preoperative lumbar and intraoperative ventricular CSF for T-tau and P-tau₁₈₁ are largely inclined towards high rostral concentration (Fig. 2B, C), which supports the traditional theory about the influence of protein origin.

The role of altered hydrodynamics is also a potential confounding factor for interpretation of biomarker ratios. Naturally, the CSF shunt surgery alters the CSF drainage as well as modifies the hydrostatic pressure affecting the natural CSF flow. Further, unoperated iNPH patients have been found to have different flow pattern of CSF, as re-directed aqueductal flow, and significant extra-cranial CSF productions have been suggested [35, 36]. In addition, the pathophysiology of iNPH itself has been suggested to originate from the malfunction of arachnoid granules, that potentially further modifies the CSF composition. Other iNPH pathological mechanism led from the hydrodynamics is the glymphatic pathway defect. Approximately 20% of CSF drainage to the systemic circulation is derived from the glymphatic system, and for iNPH patients, the glymphatic pathway has been reported to potentially be impaired by decreased aquaporin-4 density and tracer clearance in MRI imaging [37].

Furthermore, the dilution effect of increased ventricular volume or CSF production rate has been discussed as a reason for altered CSF AD biomarker compositions. In the study regarding disproportionately enlarged subarachnoid-space hydrocephalus (DESH) patients, a subset was noted to have low P-tau₁₈₁ and Aβ₄₂ and to associate for higher DESH-score [38], implying CSF-dynamics disorders to potentially dilute biomarker concentrations. However, a study conducted with healthy volunteers found no correlation between AD core biomarkers and ventricular volume nor the intracranial pressure and CSF production rate [39]. In a recent genome wide meta-analysis, a link between CSF P-tau₁₈₁, lateral ventricle volume and the genes of GMNC and C16orf95 was established, implying causative relationship for these phenomena [40].

Furthermore, to our knowledge the direct effect of the CSF shunt as such to the biomarker concentrations, has not been studied previously. It can be hypothesized that CSF shunt may affect to the biomarker concentrations, e.g., due to the absorption of CSF shunt material or the different protocol used during the harvesting. Hence, we conducted additional in vitro experiment with two protocols to evaluate these potential confounding factors affecting the usage of ventricular CSF and to fully mimic the sampling procedures of intraventricular CSF (Supplementary Tables 3 and 4). Based on our results the Aβ₄₂ has slight tendency to absorb to polypropene syringe. However, the impact of this phenomenon becomes insignificant in larger sample sizes of over 5 ml. There was also a trend for Aβ₄₂ to decrease between baseline and endpoint samples. Further, the changes of concentrations caused by the sample size and protocol became irrelevant when compared to the endpoint values rather than baseline. For T-tau and P-tau₁₈₁, no relevant changes were seen neither in shunt system protocol nor in the sample size protocol. This further strengthens the reliability of T-tau and P-tau₁₈₁ concentrations measured from V-CSF.

Previously, we found several fold increases in T-tau and P-tau₁₈₁, post-operatively, both in ventricular and lumbar CSF [17]. In Aβ₄₂ there was just a moderate continuous decrease. However, further study is needed to fully understand this longitudinal phenomenon caused by CSF shunt. Understanding more of this could also open a window to find shunt malfunction by biomarkers. In addition, the further information would be crucially important to evaluate value of AD biomarker values taken after surgery when attempt to indicate AD comorbidity. So far, we rely more on to prognostic value of brain biopsy than the post-operative follow-up CSF biomarker values.

A strength of this study was that it was possible to compare a series of parallel samples postoperatively. Additionally, our pre- to intraoperative CSF biomarker comparison had a relatively large number of samples. Furthermore, our samples obtained by shunt reservoir puncture were larger than 5 ml, corroborating the reliability of our results. A challenge, however, was the inability to rigorously confine the magnitude of the error for T-tau and P-tau₁₈₁ due to the surgical procedure in the intraoperative sampling. This should be considered when interpreting the obtained equations. This kind of sample collection provides a foundation for the subsequent calculation of similar equations for other CSF biomarkers as well.