Abstract
Background:
There is substantial variation caused by preanalytical procedures in the measurement of cerebrospinal fluid (CSF) biomarkers for Alzheimer’s disease (AD) reported in the literature.
Objective:
Determine whether the detergent Tween-20 improves diagnostic accuracy.
Methods:
CSF proteins (Aβ42, Aβ40, total tau, and phosphorylated tau) were measured by standard ELISA, in uncentrifuged CSF with or without 0.05% Tween-20 from patients with AD or amnestic mild cognitive impairment, and healthy elderly controls. In the main study, collection tubes containing Tween-20 (Sarstedt 15 mL) were filled with 5 mL CSF to ensure consistent detergent concentration across subsequent aliquots into Corning 2 mL tubes. These latter were also the primary collection vessel for samples without Tween-20. The effect of centrifugation, and extra tube transfer of samples with Tween-20 were also examined.
Results:
0.05% Tween-20 significantly increased mean measured CSF concentration of Aβ42 (30% ), Aβ40 (23% ), and total tau (4% ), but not phosphorylated tau. Generally, these increases were similar in all groups, although for Aβ42, the mean percentage increase with Tween-20 was slightly larger for AD. Areas under receiver-operator characteristic curves were similar whether Tween-20 was present or not. Centrifuged CSF without Tween-20 significantly reduced the measured concentration of Aβ42 versus non-centrifuged samples, a difference not seen when detergent was added. Similar CSF Aβ42 levels were found whether Tween-20 was added at collection in an extra tube or directly to the main collection tube.
Conclusion:
Addition of Tween-20 to CSF did not improve differentiation of patients from controls.
INTRODUCTION
Proposed research guidelines from the National Institute on Aging and the Alzheimer’s Association (NIA-AA) [1, 2] and from the International Working Group (IWG-2) [3] recommend that biomarkers should be included as part of the diagnostic assessment of Alzheimer’s disease (AD) when validated. Core biochemical markers in cerebrospinal fluid (CSF) include secreted amyloid beta 42 (Aβ42), where a reduction is considered to reflect amyloid pathology accumulation in the brain, and the cytoskeletal protein tau (total tau (t-tau)) and its hyperphosphorylated form (p-tau), where an increase in CSF is considered to reflect neuronal injury [1, 4]. Additionally, reduced levels of CSF Aβ40 correlate with rapid cognitive decline [5]. Quality control (QC) programs to standardize analytical factors in the measurement of biomarkers between and within laboratories are currently running [6–10], and efforts are being made to harmonize the diagnostic criteria [11]. Some proteins and peptides, like Aβ, possess chemical properties that foster inter-protein interactions (e.g., oligomeric species [12]), or interactions with plastic surfaces, such as polypropylene [13–17], which can interfere with their accurate measurement. The latter effect can be a challenge for monitoring biomarkers in a fluid such as CSF, which usually has a low protein concentration. Proteins can be lost through increased contact with plastic surfaces when changing tubes, during pipetting, or during the removal of cells in CSF by centrifugation. There is disagreement as to whether centrifugation of CSF results in reduced measurement of CSF Aβ42 levels [12, 19].
Addition of the mild, non-ionic detergent Tween-20 to CSF might result in measurement of more clinically-relevant levels of biomarkers [12, 20]. At concentrations comparable to those used in standard ELISA wash buffers, and significantly below micelle-forming levels, Tween-20 has been shown to increase Aβ42 measurements in CSF, perhaps due to prevention of binding to polypropylene tubes [16, 21]. Such effects could influence the early diagnosis of AD or its symptomatic predementia phase [22].
Although Tween-20 has been shown to increase measured levels of Aβ42, no comprehensive study has been conducted to determine whether the use of a detergent like Tween-20 is clinically useful; whether increased detectable levels of Aβ42 improve the sensitivity and specificity for identifying AD and/or amnestic mild cognitive impairment (amnestic MCI) that progresses (or not) to AD. The present study was conducted mainly to examine these possibilities. Smaller investigations have also been conducted to investigate whether Tween-20 has an effect on the potential confounding factors; centrifugation of CSF, and tube transfer.
MATERIALS AND METHODS
Ethics
The study was conducted according to the Helsinki Declaration. Written, informed consent was obtained from all patients or suitable proxies, and from all control individuals. The biobank is licensed by the Norwegian Directorate for Social and Health Affairs, and the research project was approved by the Regional Committee for Medical Research Ethics.
Subjects
Seventy-one patients, diagnosed by a single neurologist (SBS), were recruited through the Department of Neurology, University Hospital of Trondheim and assigned to this study from the summer of 2009. Patients were included with either early AD (n = 27, NINCDS-ADRDA criteria [23]), or amnestic MCI (n = 44, the International Working Group on Mild Cognitive Impairment Criteria [24]), where 21 patients with amnestic MCI did not change diagnosis during the next two years, 21 progressed to AD (these subgroups of amnestic MCI have been subsequently designated respectively sMCI and pMCI), and two patients with amnestic MCI were excluded from the study at follow-up due to other final diagnoses. Thus, of the 71 patients assigned, CSF samples from 69 were used in the analysis. All patients were ethnic Norwegians between 50–75 years of age, and with sufficient sight and hearing to complete the cognitive testing. Since the Department of Neurology tends to receive younger patients with dementia, older patients being referred to the Department of Geriatrics, patients with AD in this study are comparatively young on average. Exclusion criteria were a present psychiatric or malignant disease, use of anti-coagulating medication, or high alcohol consumption. Forty-six elderly volunteers recruited from societies for retired people in central Norway, or caregivers not genetically related to the patients, were recruited as healthy controls and also examined by SBS. Control individuals were healthy for their age without signs of a neurological disorder. Neurological examination was performed on both patient and control groups, including the Mini-Mental State Examination (MMSE) and the Ten-Word Test with delayed recall as taken from ADAS-Cog [25]. APOE genotyping was performed on blood samples from all patients and control individuals according to the method described elsewhere [26].
Sampling and treatment of CSF
CSF was obtained from patients or healthy elderly volunteers by lumbar puncture, usually at the level L4/L5 with the patient lying on their side, and collected directly into polypropylene cryovials immersed in ice-water. The pattern of CSF collection was the same for all study individuals, and usually performed early in the morning. The first 2.5 mL CSF was used for other purposes including routine clinical investigation. In the case of samples without Tween-20, 1 mL aliquots of CSF were collected directly into 2 mL cryovials (Corning) as the primary collection tube. For samples with Tween-20, 5 mL CSF was collected initially into a 15 mL polypropylene tube (Sarstedt) containing 25μL 10% Tween-20 in PBS, final concentration 0.05% [16, 27], and subsequently aliquoted into 2 mL cryovials (Corning). All samples were kept in ice-water while being sent for storage, and were frozen within 30 min of lumbar puncture unless centrifuged [12, 28]. All samples were stored at –80°C until analysis, and all analyzed samples had only this single freeze-thaw treatment. Five CSF samples without Tween-20 from patients, and two samples without Tween-20 from healthy controls, were centrifuged due to contamination from blood. Erythrocyte counts in all remaining samples were (mean ± SD) 1.0 ± 1.8/μL for patients and 1.2 ± 1.7/μL for controls (overall range 0–11 erythrocytes/μL).
The majority of CSF samples in the present study were not centrifuged. However, the effect of centrifugation on biomarker levels in a few parallel samples from patients or controls that were either centrifuged (2000 g for 10 min., 4°C) or non-centrifuged, with (n = 21) or without Tween-20 (n = 18) was alsoexamined.
In the main study, 0.05% Tween-20 was added at spinal tap, where collection was made in a separate tube from a different manufacturer (Sarstedt) to the main storage tube (Corning). Since different polypropylenes can have dissimilar protein retention and/or aggregation properties [14], a few samples (n = 8, from patients or controls) were used to investigate whether the extra tube transfer affected the measured concentration of the analytes in CSF. Analyte levels were compared in parallel samples where Tween-20 had either been added at spinal tap, or added at the same concentration to CSF in the original Corning tubes after thawing, just prior to biomarker determination.
ELISA assays
Analysis of CSF employed ELISA monoplex kits (Innogenetics: Aβ1-42 (Aβ42, cat. no. 80324), total tau protein (t-tau, cat. no. 80323), 181-phosphorylated tau (p-tau, cat. no. 80317), and Aβ1-40 (Aβ4 IBL International: cat. no. 81015). All samples were slowly thawed in ice-water prior to assay, and all kits were run according to the manufacturers’ instructions. No samples underwent more than one freeze-thaw cycle. Except for the Aβ1-40 assay where CSF was diluted 1:500 in phosphate-buffered saline, all samples were analyzed undiluted, in duplicate. Both control and patient samples were included on each plate. For Aβ42, Aβ40, t-tau, and p-tau, the within-plate variations without Tween-20 were respectively 5.8, 4.1, 2.8, and 2.8% (n = 6). Corresponding values with 0.05% Tween-20 were 4.5, 5.5, 7.3, and 1.7% (n = 6). The between-plate variations without Tween-20 were respectively 7.6, 20.9, 7.5, and 5.4% (n = 6), and corresponding values with Tween-20 were 10.5, 14.1, 5.1, and 2.9% (n = 6).
Statistical analyses
Statistical analysis was performed using R 2.14 [29]. Paired t-tests were applied to investigate the effect of centrifugation, and the addition of Tween-20 either prior to or after thawing of the CSF samples.
Marker concentrations were analyzed on the log (base 2) scale, and accordingly, plots of marker concentrations exhibited log spacing. For boxplots, boxes were drawn around the 25th and 75th percentile of the data. A line was drawn at the median and whiskers were drawn to the farthest point not more than 1.5 interquartile ranges from the edge of the box. Further data relating to these plots, including mean percentage differences between samples with and without Tween-20 which were estimated by calculating the mean paired difference between the log2 concentrations, and transforming these estimates to the percentage scale, are given. The F-test from a one-way analysis of variance model was used to address possible variation between mean percentage differences acrossdiagnostic groups.
The slope from a Deming regression [30] of log concentrations from samples with Tween-20 versus samples without Tween-20 with ratio of standard deviations equal to unity, was used to address the homogeneity of the mean percentage difference across the assay range.
Receiver-operator characteristic (ROC) curves [31] were used to examine the ability of the markers to discriminate the groups. Results are shown as the area under the ROC curve (AUC), and differences between samples with or without Tween-20 were determined using Delong’s test [32]. Only comparisons between patients with AD and healthy controls, and the MCI subgroups are shown in the text. Other comparisons are given in supplementary data. Threshold plots and density plots were created with nonparametric kernel density estimation [33] using the Epanechnikov kernel and bandwidth based on Altman and Leger [34]. Exploratory cut-offs were calculated by finding the values that maximized Youden’s index [35].
RESULTS
A summary of demographic measures for the individuals included in the study are shown in Table 1. CSF levels of Aβ and tau species in patient groups and healthy controls were measured in the various diagnostic groups, both with 0.05% Tween-20 added at spinal tap in an intermediate Sarstedt tube prior to aliquoting in Corning tubes, or in samples without detergent.The results are summarized in Supplementary Figure 1 and Supplementary Table 1a, b. The meanconcentration of Aβ42 was always higher in samples where Tween-20 had been added (all p < 0.001). Although the percentage increase was similar for the control group and MCI subgroups (about 29% ), it was a little higher for the group of patients with AD (35.9% ; F-test for percentage difference not the same across the groups, p = 0.022). For Aβ40, addition of Tween-20 resulted in a mean increase of about 23% , (p < 0.001) averaged over all groups, with no evidence that this difference varied across the diagnostic groups. The addition of Tween-20 to CSF had a smaller effect on the level of t-tau or p-tau measured than it had on the two Aβ species. In the case of t-tau the percentage increase in the presence of Tween-20, although statistically significant (p = 0.001), was only 4% on average for all groups, and did not vary significantly between the diagnostic groups. The estimated increase in p-tau with the addition of Tween-20 was less than 1% on average, and was not statistically different from zero in any of the groups. The t-tau/Aβ42 and p-tau/Aβ42 ratioson average demonstrated significant decreases with Tween-20 (respectively –20% and –23% ). There was no evidence for the mean percentage difference varying across the diagnostic groups.
Figure 1 shows a direct comparison of values with 0.05% Tween-20 compared to values without Tween-20 for Aβ42, Aβ40, t-tau, and t-tau/Aβ42. Overall increases in Aβ42, and Aβ4 with Tween-20 are evident by the shift above the y = x reference line, and the decrease with the t-tau/Aβ42 ratio shown by a shift below the y = x reference line. The slope wassignificantly different from unity only for Aβ42, where the percentage change was greater for lower values in the presence of Tween-20 (i.e. particularly during AD), compared to higher values (i.e., particularly in the control group).
Statistics for the comparisons of analyte concentration in each group (between-group comparisons transformed on a percentage scale) are shown in Supplementary Table 2. The addition of Tween-20 to CSF samples resulted in no further improvement to the separation of the patient groups and healthy controls.
To compare the diagnostic accuracy of Aβ42, Aβ40, and t-tau, and the t-tau/Aβ42 ratio, ROC plots were created. The results are shown in Table 2 as values for the AUC for patients with AD versus the group of controls, and for the pMCI versus sMCI subgroup. AUC values with Tween-20 were very similar to values without Tween-20 for all markers. All the CSF analytes except Aβ40 demonstrated good separation between patients with AD and the control group, with the largest AUC found for Aβ42; 96% without, and 95.6% with Tween-20. Excellent separation was also achieved with the t-tau/Aβ42 ratio (94.9% without, and 94.8% with Tween-20), with t-tau alone achieving a slightly lower separation (87.8% without, and 88.4% with Tween-20). When the subgroups of patients with amnestic MCI were compared, the best separation was achieved using the t-tau/Aβ42 ratio (76.9% without Tween-20 and 76.4% with Tween-20). Aβ42 alone was not a good marker to distinguish these latter groups with or without Tween-20. The pMCI and sMCI subgroups were also compared to healthy controls, with the largest AUC obtained for Aβ42 in both cases (Supplementary Table 3).
Since Aβ42, or the t-tau/Aβ42 ratio were best for separating patients with AD from healthy controls, threshold plots were constructed to calculate cut-off values in the absence or presence of Tween-20 (Fig. 2). The cut-off level for Aβ42 based on maximizing Youden’s index, was calculated to be 832 ng/L in samples with Tween-20, compared to 630 ng/L without Tween-20. These cut-offs yield estimates of sensitivity and specificity of around 88% and 92% respectively with Tween-20, and 86% and 92% withoutTween-20. Corresponding cut-off values for the t-tau/Aβ42 ratio, were respectively 0.49 with Tween-20 (sensitivity = 85% , specificity = 88% ) and 0.76 without Tween-20 (sensitivity = 81% , specificity = 93% ).
Centrifugation of CSF had a significant effect only on the concentration of Aβ42 in the absence of Tween-20: if values obtained in non-centrifuged samples without Tween-20 were set to 100% , corresponding values from centrifuged samples without Tween-20 were Aβ42: 88.9 ± 8.2% (p < 0.0005); Aβ40: 101.7 ± 28.2% ; t-tau: 98.9 ± 15.1% ; and p-tau: 100.7 ± 4.5% , all n = 13–18. In samples with Tween-20 where non-centrifuged samples were set to 100% , corresponding values from centrifuged samples with Tween-20 were Aβ42: 99.7 ± 4.4% ; Aβ40: 101.7 ± 14.7% ; t-tau: 102.9 ± 8.0% ; and p-tau: 99.4 ± 3.5% , all n = 8–21, no significant differences.
Addition of 0.05% Tween-20 after thawing of stored CSF immediately prior to ELISA, was not significantly different (paired t-test) to levels measured in CSF where Tween-20 had been added in a 15 mL Sarstedt tube during collection. If values obtained from adding detergent at collection were set to 100% , then the values obtained from parallel samples when Tween-20 was added directly to Corning tubes were Aβ42: 102.3 ± 7.8% ; Aβ40: 102.1 ± 16.8% ; t-tau: 99.5 ± 3.6% ; and p-tau: 101.5 ± 2.6% , all n = 8.
DISCUSSION
In the present study, using the polypropylene tubes as described, 0.05% Tween-20 substantially increased the measured level of Aβ42 in CSF across all diagnostic groups. Our calculated cut-off values separating patients with AD from healthy controls increased from 630 ng/L without Tween-20, to 832 ng/L in the presence of the detergent. However, the percentage gain in measurable Aβ42 in the presence of Tween-20 across the entire assay range was most marked for low Aβ42 values and therefore significantly higher in the AD group compared to the other groups. It might affect diagnostic cut-off values, but our results show the difference is very small compared to the difference between AD patients and controls. Certainly, the presence of the detergent in this cohort did not seem to significantly affect sensitivity, specificity or diagnostic accuracy. The most important finding in the present study is therefore that the addition of a low concentration of a detergent like Tween-20 did not change the performance of the assays to discriminate between patients with AD and healthy controls. This was also the case for distinguishing patients with amnestic MCI that progressed to AD within two years from those that did not. Although the addition of Tween-20 did not appear to provide any improvement in the differentiation of patients from healthy controls, it did not appear to be a detrimental factor either.
Neither Aβ40 nor the tau species showed any difference in percentage gain for high or low concentrations of measurable protein in the presence of Tween-20 across the entire assay range.
Previous studies of biomarker concentrations in CSF pretreated with Tween-20 have also found increased Aβ42 levels [12, 21]. Possible reasons suggested for the increase have included release of protein-bound Aβ, differences in oligomerization, or reduced adhesion to plastic vial walls. Toombs et al. [21] and Pica-Mendez et al. [16], found a similar increase in the level of Aβ42 in CSF in the presence of Tween-20, which they suggested to be due to release from surface adsorption. Pica-Mendez et al. also recommended using 0.05% Tween-20 to ensure there would be no loss of Aβ to tube surfaces. Toombs et al. recommended not only use of the same tube throughout a study, but that tubes should be filled to the same degree to reduce a potential adsorption effect. This was the case in the present study. However, if CSF samples were to be aliquoted again after primary collection, the use of Tween-20 would probably help retain the Aβ42 signal present at the time of draw.
Centrifugation of CSF is routine for many centers [10, 12]. The advantage of Tween-20 as shown in the present study, is that it seems to result in similar levels for all the core biomarkers measured by ELISA, whether samples are centrifuged or not. This is particularly important for Aβ42 which is the marker showing the greatest potential reduction in concentration following centrifugation of CSF. This latter aspect has been reported previously by others [12]. However, the reduction in Aβ42 measured in centrifuged samples compared to non-centrifuged samples in the present study is still well within the variation of CSF Aβ42 levels found between laboratories[6, 36].
Certainly, in the collection procedure for samples with Tween-20 in the current study, the CSF was first collected in a polypropylene tube (Sarstedt) containing Tween-20, before being aliquoted into smaller volumes, as this was considered more accurate than to add the detergent in a small volume to each tube separately. This could potentially result in more adsorption, both to the Sarstedt tube and via the pipetting step that followed, and/or introduce a source of error through the use of different polypropylene vials [14]. However, Aβ42 in our samples with Tween-20 showed similar levels, whether the Tween-20 was added to theSarstedt tube at sampling, or directly to the Corning tube prior to ELISA, so it seems to make no difference whether Tween-20 is added at spinal tap or just prior toanalysis.
Standardization has been hampered by (sometimes considerable) variation in the biomarker levels measured in CSF by different centers, even when using identical kits. Single-center studies with standardized protocols have shown better reproducibility[37]. The BIOMARKAPD project [9] and the Global Biomarker Standardization Consortium (GBSC) [8] have been specifically designed to address theneed for improved analytical standardization. Some of the variables that have precluded generation of CSF biomarker thresholds for AD diagnosis and prognosis include lack of universal reference and QC standards, and inconsistent sample handling and assay methodology.
Adding Tween-20 to CSF after collection would entail an extra procedure, which might be problematic in the clinical or research setting, but would be worthwhile if larger studies indicate reduced variation of confounding factors. We did not measure biomarker levels, including Aβ42, in a range of polypropylene or other types of plastic tubes. However, our limited data suggest that Tween-20 can reduce the effect of centrifugation, at least in the Corning tubes used here, and the extra tube transfer did not appear to have a negative effect. Tween-20 appeared to have no detrimental effect on measurement of Aβ42 or the other analytes, but since none of the results suggested any improvement in diagnostic accuracy for AD, it seems to provide no clinical advantage.
Footnotes
ACKNOWLEDGMENTS
The authors are grateful to the patients and their caregivers, as well as the healthy individuals, for participating in this study.
Guro Berge holds a PhD scholarship (46060903) from the Liaison Committee between the Central Norway Health Authority and NTNU. This study has received support from Dementia Disease Initiation (DDI), and from the Research Council of Norway (NASATS-NevroNor grant 217780/H10). The project is approved by The National Committees for Research Ethics (Approval 2013/150).
The authors thank Sylvia Nome Kvam for help in the development of the study.
