Abstract
The diagnostic performances of cerebrospinal fluid (CSF) biomarkers and amyloid positron emission tomography (PET) were compared by examining the association and concordance or discordance between CSF Aβ1-42 and amyloid PET, after determining our own cut-off values for CSF Alzheimer’s disease (AD) biomarkers. Furthermore, we evaluated the ability of CSF biomarkers and amyloid PET to predict clinical progression. CSF Aβ1-42, t-tau, and p-tau levels were analyzed in 203 individuals [27 normal controls, 38 mild cognitive impairment (MCI), 62 AD dementia, and 76 patients with other neurodegenerative diseases] consecutively recruited from two dementia clinics. We used both visual and standardized uptake value ratio (SUVR)-based amyloid PET assessments for analyses. The association of CSF biomarkers with amyloid PET SUVR, hippocampal atrophy, and cognitive function were investigated by linear regression analysis, and the risk of conversion from MCI to AD dementia was assessed using a Cox proportional hazards model. CSF p-tau/Aβ1-42 and t-tau/Aβ1-42 exhibited the best diagnostic accuracies among the CSF AD biomarkers examined. Correlations were observed between CSF biomarkers and global SUVR, hippocampal volume, and cognitive function. Overall concordance and discordance between CSF Aβ1-42 and amyloid PET was 77% and 23%, respectively. Baseline positive CSF Aβ1-42 for MCI demonstrated a 5.6-fold greater conversion risk than negative CSF Aβ1-42 . However, amyloid PET findings failed to exhibit significant prognostic value. Therefore, despite presence of a significant correlation between the CSF Aβ1-42 level and SUVR of amyloid PET, and a relevant concordance between CSF Aβ1-42 and amyloid PET, baseline CSF Aβ1-42 better predicted AD conversion.
Keywords
INTRODUCTION
Clinically diagnosed Alzheimer’s disease (AD) dementia does not necessarily guarantee AD pathology in postmortem examination. Therefore, several biomarkers that reflect AD-specific pathophysiological features have been incorporated into the AD diagnostic criteria by the National Institute on Aging - Alzheimer’s Association (NIA-AA) [1] and the International Work Group [2]. Of these, cerebrospinal fluid (CSF) biomarkers, which indicate neuropathological changes in the brain, are considered to be useful and informative biomarkers for AD diagnosis, specifically in the early stages of the disease [3]. Indeed, decreased CSF amyloid-beta 1–42 (Aβ1-42) levels, and increased total tau (t-tau) and phosphorylated tau (p-tau) levels suggest the formation of extracellular Aβ plaques and the deposition of intracellular neurofibrillary tangles that contain hyperphosphorylated tau, which reportedly occur 10–20 years before the onset of clinical symptoms in familial or sporadic AD [4, 5].
CSF biomarkers have also been used to predict conversion to AD dementia in patients with mild cognitive impairment (MCI) [6, 7]. In general, CSF biomarkers demonstrate little to no change while progressing from MCI to AD dementia or during the clinical phase of AD dementia [8, 9]. However, patients with subjective cognitive decline, or MCI who have AD signatures in CSF biomarkers at baseline, are more likely to show clinical progression [10].
Amyloid positron emission tomography (PET), in combination with CSF biomarkers, is a well-established AD imaging biomarker to directly visualize Aβ accumulation in vivo [11]. In addition to diagnostic value, amyloid positive baseline PET has high prognostic value with regard to the conversion of MCI to AD dementia [12].
Several studies have reported that low CSF Aβ1-42 and positive amyloid PET were highly correlated each other [13, 14] and both have been incorporated as the representative biomarkers of cortical Aβ pathology in the new diagnostic research criteria of AD [1]; however, there have been discrepancies in that these CSF and PET biomarkers can be interchangeably valid for the purpose of predicting and diagnosing AD [15].
Therefore, here, we first measured CSF Aβ1-42, t-tau, and p-tau levels in clinically relevant participants, and defined cut-off values to differentiate AD dementia from healthy controls and other neurodegenerative diseases. Numerous studies have demonstrated the potential utility of CSF biomarkers for the early detection of AD. However, the inter-laboratory variability observed in CSF biomarkers reduces the reliability of external-laboratory-determined cut-off values. Therefore, there was a need to establish our own cut-off values [16, 17]. Second, we investigated whether the CSF biomarker levels were correlated with the standard uptake value ratio (SUVR) of amyloid PET, hippocampal atrophy, and cognitive function. Third, we examined the concordance and discordance between CSF Aβ1-42 and amyloid PET positivity, to determine whether these two Aβ biomarkers could provide identical information. Lastly, we evaluated and compared the predictive capability of CSF biomarkers and amyloid PET with regard to the clinical progression from MCI to dementia.
MATERIALS AND METHODS
Participants
We recruited 216 individuals who attended the Dementia Clinic at Pusan National University Hospital (PNUH) and Pusan National University Yangsan Hospital (PNUYH) between April 2015 and October 2018. All participants were examined by neurologists specializing in neurodegenerative diseases, followed by clinical interview and neurological examination. Cognitive assessment included the Mini-Mental State Examination (MMSE) and the Seoul Neuropsychological Screening Battery (SNSB) [18], which is a detailed neuropsychological test. Complete blood count, biochemistry, thyroid function tests, lipid profile, vitamin B12, folate, and syphilis serology did not reveal any metabolic etiologies associated with the patients’ cognitive abnormalities. To exclude territorial cerebral infarction, brain tumor, extensive leukoaraiosis, and other structural lesions, all participants underwent either brain magnetic resonance imaging (MRI) (n = 211) or brain computed tomography (CT) (n = 5). Patients with MCI met both the NIA-AA core clinical criteria for MCI [19] and the following modified Petersen’s criteria [20]: 1) subjective memory complaints by patients or caregivers, 2) normal activities of daily living (ADL), based on clinical findings and the ADL scale (S-IADL) [21], 3) objective cognitive dysfunction as evidenced by a Z score of –1.5 standard deviations (SD) below those of age- and education-matched normal individuals, and 4) the absence of dementia. Patients with AD dementia satisfied the NIA-AA core clinical criteria for probable AD dementia [1]. Patients who met the international consensus criteria for behavioral variant frontotemporal dementia (bvFTD) [22], semantic variant of primary progressive aphasia (svPPA), nonfluent/agrammatic primary progressive aphasia (nfvPPA) [23], progressive supranuclear palsy-Richardson syndrome (PSP-RS) [24], dementia with Lewy bodies (DLB) [25], Parkinson’s disease (PD) [26] or normal pressure hydrocephalus (NPH) [27], or subjective memory impairment (SMI) [28] were also included.
Normal controls (NCs) included 27 participants, 19 of which were recruited upon being admitted to the hospital for surgical procedures involving non-brain diseases such as orthopedic, urologic, or gynecologic diseases, or varicose veins requiring spinal anesthesia. The other 8 were volunteers recruited by other means from the general population. No control had a previous history of neurological or psychiatric illness, any abnormality as determined by neurological examination, or a structural lesion by brain MRI. Additionally, cognition (as defined by the SNSB) was within normal limits for all NCs.
Informed consent was obtained from all participants, and study protocols were approved by the Institutional Review Board at each hospital (H-1502-008-026/H-1809-005-070 in PNUH, 05-2015-069/04-2018-026 in PNUYH).
Lumbar puncture and CSF processing
Lumbar puncture was conducted at L3–4 or L4–5 between 9 am and 2 pm using a 23G conventional cutting-edge needle. Initially, 1 mL of CSF was used for routine analysis (leukocyte count, glucose concentration, and protein), followed by 10–15 mL (4 mL for the 19 NCs whose CSF was collected during spinal anesthesia) was collected in polypropylene tubes. Within 4 h of collection, CSF samples were centrifuged at 2000–4000 g for 10 min at 4°C. Aliquots of the CSF samples were transferred to new polypropylene tubes (400 μL into a 500 μL tube, or 1 mL into a 1.5 mL tube) and immediately stored at –80°C, until further analysis. Nine participants who showed >12,519 red blood cells per CSF 1 μL in the routine CSF analysis were excluded.
CSF biomarker assessment
CSF samples were analyzed at the Research Institute for Convergence of Biomedical Science and Technology at PNUYH. Aliquots were thawed at room temperature for 1 h before measuring CSF biomarker levels. CSF Aβ1-42, t-tau, and p-tau181 levels were quantified using the INNOTEST ELISA kit (Fujirebio Diagnostics, Ghent, Belgium) by one experienced laboratory technician (KHS), who was blinded to the clinical diagnosis.
All samples and standards were measured in duplicates to monitor intra-test variabilities. Inter-test validation was ensured by reanalyzing CSF Aβ1-42, t-tau, and p-tau levels in 20 randomly selected samples at a six month after the initial evaluations, and correlations between the results were checked using Pearson’s correlation test. Intra- and inter-test variabilities were assessed using coefficient of variation % (CV% = SD/mean%).
We excluded four participants demonstrating p-tau levels below the detection limit proposed by the manufacturer.
PET image acquisition
Among 203 study subjects, 85 underwent amyloid PET scans using a Discovery STE PET/CT (GE healthcare, Milwaukee, WI, USA) or a Biograph 40 PET/CT scanners (Siemens, Knoxville, TN, USA). 18F-florbetaben (n = 73) or 18F-flutemetamol (n = 12) was injected intravenously, and PET images were acquired from 90 to 110 min after injection. Scans were reconstructed using the iterative ordered subset expectation maximization algorithm using 8 iterations and 14 subsets.
PET image analysis
PET images were visually assessed by a nuclear physician blinded to clinical data, using a previously described method [29, 30]. Images were designated either a positive or negative score by comparing their intensities in the indicated regions of cortical grey matter with those in adjacent white matter. Aβ negativity was defined as a lower tracer uptake in grey matter than in adjacent white matter, while Aβ positivity was defined as tracer uptake equal to or greater than that in white matter. Details of the method used to visually inspect PET images are provided in the Supplementary Material.
Quantitative analysis was conducted by normalizing the MRI images to a T1-weighted MRI template, and co-registering PET images with the MRI images in all cases. Volume of interest (VOI) values were obtained using the automated anatomical labeling atlas [31], and the cerebellum was used as a reference region. A VOI template was applied to the following 11 bilateral regions: frontal, parietal, lateral temporal, occipital, anterior cingulate, posterior cingulate, caudate nucleus, putamen, pallidum, thalamus, and hippocampus. Mean standardized uptake values (SUVs) were determined for each VOI, and SUVRs were calculated by dividing the mean SUV of each VOI by the SUV of the cerebellum. Similarly, global SUVRs were calculated by dividing the global SUVs by that of the cerebellum [32, 33]. Five participants demonstrating an absence of or inappropriate T1-weighted images were excluded from the analysis.
Neuropsychological assessment
One hundred and eighty participants underwent the SNSB, a standardized neuropsychological battery that tests attention, language, praxis, visuoconstructive function, verbal and visual memory, and frontal/executive functions [18]. To investigate possible correlation between CSF biomarkers and cognitive impairment, we modified several subdomains of the original SNSB. Attention was assessed using the backward digit span test; language using the Korean version of the Boston naming test (K-BNT); visuoconstructive function using the Rey-Osterrieth Complex Figure Test (RCFT); memory using the delayed recalls of the Seoul Verbal Learning Test and the RCFT; and frontal/executive function using categoric and phonemic word generation tests, the Stroop color-reading test, digit symbol coding (DSC) and the Korean-trail making test (K-TMT part B). Z-scores or average Z-scores from subtests in each domain underwent correlation analysis.
MRI assessments
One hundred and twenty-five subjects underwent T1-weighted MRI using a 3-T scanner (Verio, Skyra I or II, or MAGNETOM Trio A tim; Siemens, Germany). Detailed imaging parameters of each scanner are described in the Supplementary Material. T1-weighted MR images were processed for subcortical structure parcellation using the FreeSurfer software package (version 5.3.0). After normalizing T1 images intensities and skull stripping, subcortical region parcellation was conducted in the MNI 152 atlas space and bilateral hippocampal volumes were measured (in mm3) for each patient. We could not acquire the volume for one patient due to poor image quality.
Follow-up evaluation of participants with MCI
Participants with MCI routinely visited our memory clinic on a 3-month basis, during which we used the criteria identical to those used to determine the presence of probable AD dementia at baseline. Participants with MCI that converted to AD dementia (progressive MCI) were censored when AD dementia was diagnosed. Patients with MCI that did not progress to dementia (stable MCI) were censored at the time of their most recent follow-up.
Flow diagram demonstrating patient selection in each analysis.
Statistical analysis
The demographic characteristics and CSF biomarkers of diagnostic groups were compared using Chi-square/Fisher’s exact test for categorical variables and one-way analysis of variance (ANOVA) with Bonferroni’s correction for continuous variables. Receiver operating characteristic (ROC) curves were drawn by plotting sensitivity against 1-specificity, and areas under ROC curves (AUCs) were used to examine the diagnostic accuracies of continuous CSF variables [34]. Ideal cut-off values for AD dementia were derived using the Youden index [35].
Correlations between CSF biomarkers and global SUVR were expressed as Spearman’s rank correlation coefficients (ρ) to account for the non-normal distribution of the amyloid profiles. Multiple linear regression analysis adjusted for age, sex, education, and APOE ɛ4 status was used to investigate the influence of global and regional SUVRs on CSF Aβ1-42. Linear regression analysis was conducted using the same covariates to explore the effects of hippocampal volume on CSF biomarkers. We evaluated the associations between the CSF biomarkers and cognitive parameters using similar regression models, with APOE ɛ4 status as a covariate, since it reportedly influences CSF biomarker levels and hippocampal volumes [36]. Linear regression analyses were performed using log-transformed CSF biomarker data. Concordance between CSF Aβ (CSF Aβ1-42, t-tau/Aβ1-42 ratio, or p-tau/Aβ1-42) and amyloid PET (visual inspection) was defined as the proportion of individuals with both positive (Alzheimer-like, CSF + /PET +) or negative (not Alzheimer-like, CSF-/PET-) biomarkers. Discordance was defined as the proportion of individuals with only one positive biomarker (CSF + /PET- or CSF-/PET+). We examined the predictive values of baseline CSF amyloid biomarkers and amyloid PET with respect to MCI to AD dementia conversion using the Cox proportional hazards models that included age, sex, education, and APOE ɛ4 status as covariates.
RESULTS
Demographic and laboratory characteristics
Among the 216 original study subjects, 13 were excluded (due to >12,519 red blood cells per CSF 1 μL or erroneous values in CSF tau analysis) (Fig. 1). The remaining 203 were classified into four clinical diagnostic groups: NC (n = 27), MCI (n = 38), AD dementia (n = 62), and other neurodegenerative diseases (OND, n = 76). The fourth group included 14 frontotemporal dementia (FTD; 7 bvFTD, 5 svPPA, 1 nfvPPA, and 1 FTD with motor neuron disease), 5 PD, 27 NPH, 1 PSP, 7 DLB, 1 progressive bulbar palsy, 1 epileptic memory impairment, 2 spinocerebellar ataxia, 1 Wernicke encephalopathy, 3 unclassified neurodegenerative dementia, 4 unclassified parkinsonism, and 10 SMI patients.
The demographic, clinical, and CSF characteristics of these diagnostic groups are summarized in Table 1 and Supplementary Table 1. Mean ages of patients in the MCI, AD dementia, and OND groups were higher than those in the NC group. Proportion of women was higher in the MCI group than in the other three groups, and mean MMSE score was lower in the AD dementia group than in the other groups. No significant intergroup difference was found with regard to the APOE ɛ4 allele frequencies (p = 0.087).
Demographics, clinical characteristics, and CSF biomarkers according to clinical diagnostic groups
Data are expressed as the mean (SD) or median (interquartile range) or number of participants (%); p values were calculated by chi-squared test or Fisher’s exact test and ANOVA with Bonferroni correction as appropriate. The significant differences among NC, tMCI, AD dementia, and other diseases group were indicated as follow: *p < 0.05 versus NC; †p < 0.05 versus MCI; §p < 0.05 versus Other diseases. The significant differences among NC, sMCI, pMCI, AD dementia, and other diseases group were indicated as follow: ap < 0.05 versus NC; bp < 0.05 versus sMCI; cp < 0.05 versus AD dementia; dp < 0.05 versus Other diseases. AD, Alzheimer’s disease; APOE, apolipoprotein E; CDR, Clinical Dementia Rating; CDRSOB, Clinical Dementia Rating Sum of Boxes; CSF, cerebrospinal fluid; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; NC, normal control; pMCI, progressive MCI; P-tau, phosphorylated tau; sMCI, stable MCI; tMCI, total MCI; T-tau, total tau.
Mean CSF Aβ1-42 level was lower in the AD dementia group than in the NC, MCI, and OND groups. Mean CSF t-tau and p-tau levels were higher in the AD dementia group than in the NC, MCI, and OND groups. Similarly, t-tau/Aβ1-42, and p-tau/Aβ1-42 ratios were significantly higher in the AD dementia group than in the NC, MCI, and OND groups. Furthermore, the MCI group demonstrated significantly higher t-tau and p-tau levels than the NC group.
Mean levels of all CSF biomarkers of OND group were not significantly different from those of the NC group. However, the mean levels of CSF biomarkers of NPH or FTD group were significantly different from those in the AD dementia group. Contrarily, the DLB group had only significantly lower p-tau/Aβ1-42 ratios than the AD dementia group. Mean levels of p-tau, and t-tau/Aβ1-42, and p-tau/Aβ1-42 ratios of PD group were lower than those in the AD dementia group (Supplementary Table 1).
Reliability and consistency of CSF AD biomarkers
Mean intra-test variations (CV%) were 3.4% for Aβ1-42, 3.1% for t-tau, and 2.1% for p-tau, and were well within the 20% permissible level specified by the INNOTEST® ELISA kit. Inter-test variabilities (CV%) between ELISA measurements taken six months apart were 6.5% for Aβ1-42, 3.0% for t-tau, and 5.4% for p-tau. In addition, we noted high levels of consistency between the test values (r2 = 0.969, p < 0.001 for Aβ1-42; r2 = 0.971, p < 0.001 for t-tau; r2 = 0.981, p < 0.001 for p-tau).
Cut-off values of CSF biomarkers and diagnostic accuracy for AD
Each CSF biomarker and their ratios demonstrated the following cut-off values that differentiated AD dementia from NC: 631.8 pg/mL for Aβ1-42, 271.6 pg/mL for t-tau, 56.4 pg/mL for p-tau, 0.320 for t-tau/Aβ1-42, and 0.064 for p-tau/Aβ1-42 (Table 2). The AUC for t-tau/Aβ1-42 ratio (sensitivity, 90%; specificity, 96%; AUC, 0.967) and p-tau/Aβ1-42 ratio (sensitivity, 94%; specificity, 96%; AUC, 0.955) was the most accurate in distinguishing AD dementia from NC.
Diagnostic accuracy of CSF biomarkers and combinations discriminating patients with AD dementia from controls
AD, Alzheimer’s disease; AUC, area under the curve; CSF, cerebrospinal fluid; NC, normal control; P-tau, phosphorylated tau; T-tau, total tau.
Figure 2 summarizes the diagnostic accuracies of CSF biomarkers discriminating AD dementia from OND. With the exception of DLB group, p-tau/Aβ1-42 ratio was ideal in differentiating AD dementia from each of the OND. The cutoffs differentiating AD dementia from each OND are provided in Supplementary Table 2.
Diagnostic accuracy of CSF biomarkers with respect to the differentiation of AD dementia and other neurodegenerative diseases. ADD, Alzheimer’s disease dementia; AUC, area under the ROC curve; DLB, dementia with Lewy bodies; FTD, frontotemporal dementia; PD, Parkinson’s disease.
Correlation analyses between CSF biomarkers and amyloid PET
Seventy-four subjects (17 MCI, 32 AD dementia, and 25 OND) were available for this analysis (Fig 1). The time interval between CSF testing and PET imaging was 0.4 months (SD 0.7 months). Global SUVR for amyloid retention was negatively correlated with CSF Aβ1-42 (ρ= –0.600, p < 0.001) and positively correlated with CSF t-tau (ρ= 0.316, p = 0.006), p-tau (ρ= 0.301, p = 0.009) levels, and t-tau/Aβ1-42 (ρ= 0.532, p < 0.001), and p-tau/Aβ1-42 ratio (ρ= 0.571, p < 0.001) ratios. Global SUVRs were more closely correlated with CSF Aβ1-42 than with t-tau or p-tau (Fig 3). Correlations between global SUVR and several CSF profiles [CSF Aβ1-42 (ρ= –0.586, p = 0.013), CSF t-tau/Aβ1-42 ratio (ρ= 0.554, p = 0.021), and p-tau/Aβ1-42 ratio (ρ= 0.637, p = 0.006)] were statistically significant with regard to the MCI group. However, there were no significant correlations in the AD dementia and OND groups. Linear regression analysis adjusted for age, sex, education, and APOE ɛ4 status, showed that global SUVR was significantly correlated with the CSF Aβ1-42 (β= –0.444, p < 0.001), t-tau/Aβ1-42 ratio (β= 0.410, p = 0.001), and p-tau/Aβ1-42 ratio (β= 0.488, p < 0.001).
Relationship between CSF biomarkers and amyloid PET SUVRs. Spearman’s rank correlation on 74 participants (17 MCI, 32 AD dementia and 25 other neurodegenerative diseases) are used to determine ρ and p values. CSF, cerebrospinal fluid; PET, positron emission tomography; p-tau, phosphorylated tau; SUVR, standard uptake value ratio; t-tau, total tau.
Furthermore, regional SUVRs of frontal, parietal, temporal, and occipital cortices, anterior cingulate, precuneus, putamen, and pallidum were correlated with CSF Aβ1-42, contrary to those in the posterior cingulate, caudate, thalamus, and hippocampus (Table 3).
The association of CSF Aβ1-42 level and regional SUVR values
Significant correlations are in bold. CSF, cerebrospinal fluid; SUVR, standardized uptake value ratio.
Association between CSF biomarkers and hippocampal volume or cognitive profiles
MR images of 25 MCI and 31 AD dementia participants were available to measure hippocampal volume. Mean time interval between CSF collection and MRI was 37±57 days. After adjustment for age, sex, education, and APOE ɛ4 status, all five CSF biomarkers were significantly correlated with hippocampal volume (Table 4). Furthermore, we observed the presence of negative correlations between hippocampal volume and t-tau/Aβ1-42 and p-tau/Aβ1-42 ratios in the MCI group, but the absence of a significant correlation between hippocampal volume and any CSF biomarker in the AD group (Supplementary Table 3).
The association between CSF biomarkers and hippocampal volume
p-values are adjusted for age, sex, education, and APOE ɛ4. B, unstandardized beta; CSF, cerebrospinal fluid; P-tau, phosphorylated tau; SE, standard error; T-tau, total tau.
Neuropsychological data of 36 MCI and 45 AD dementia participants were available within one year of CSF collection and the results indicated that only memory was correlated with the five CSF biomarkers (Supplementary Table 4). There was no correlation between cognitive function and any of the CSF biomarkers in the AD dementia group.
Concordance/discordance between CSF Aβ1-42 and amyloid PET
Dichotomization of the CSF Aβ1-42 levels according to the cut-off values resulted in 61 patients (77%) with concordant markers (n = 31, CSF + /PET+; n = 30, CSF-/PET-) and 18 participants (23%) with discordant markers (n = 16, CSF + /PET-; n = 2, CSF-/PET+). The proportion of concordance was similar to that when CSF p-tau/Aβ1-42 (78%) or t-tau/Aβ1-42 (70%) was used as a CSF marker. The patient characteristics for each group are shown in Table 5. Most (83.9%) CSF + /PET+ patients were clinically diagnosed with AD dementia, whereas 63.3% of the CSF-/PET- patients were diagnosed with OND. CSF + /PET+ patients had a higher APOE ɛ4 allele frequency and a smaller mean hippocampal volume than CSF-/PET- patients. Among the 16 CSF + /PET- patients, eight were clinically diagnosed with MCI (n = 4) or AD dementia (n = 4), and eight with OND (4 FTD, 1 SMI, 1 NPH, 1 nfvPPA, and 1 PSP-RS). Seven CSF + /PET- patients with MCI or AD dementia had positive CSF tau markers for AD dementia, whereas six CSF + /PET- patients with OND had low CSF p-tau levels. Only two patients (1 clinically diagnosed AD dementia, 1 NPH) were classified as discordant with CSF-/PET+. The AD dementia patient with CSF-/PET+ had a borderline level of CSF Aβ1-42 (678 pg/mL) compared to the cut-off value (<631.8 pg/mL) and positive CSF tau for AD dementia. Application of a more lenient cut-off value (<757.5 pg/mL, a 95% confidence interval) for CSF Aβ1-42 resulted in the conversion of this discordant patient into a CSF + /PET+. The other CSF-/PET+ patient with NPH also changed to CSF + /PET+ upon the application of a more lenient cut-off value (<757.5 pg/mL). However, the CSF tau values of this patient were negative with regard to AD dementia. Clinical characteristics and the CSF profiles of the discordant participants (CSF + /PET- or CSF-/PET+) are summarized in Supplementary Table 5. We did not observe any significant difference in overall concordance/discordance frequency while analyzing for different amyloid tracers (Supplementary Tables 6 and 7).
Characteristics of concordant and discordant participants based on CSF Aβ1-42 levels and amyloid PET
Data are expressed as the mean (SD) or number of participants (%). AD, Alzheimer’s disease; APOE, apolipoprotein E; CSF, cerebrospinal fluid; PET, positron emission tomography; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; P-tau, phosphorylated tau; T-tau, total tau.
Progression to dementia
The mean follow-up duration for the 38 MCI patients was 14.5 months (SD 12.7 months). Among the 38, 10 progressed to AD dementia (progressive MCI), while 12 did not over a period of one year (stable MCI). One stable MCI participant was reclassified from MCI to normal cognition (SMI) during the follow-up. Sixteen MCI patients who did not progress to dementia, but had follow-up period of less than one year were excluded in the analysis. None of these patients was diagnosed with any other type of dementia during the follow-up. Differences in age were significant between both progressive and stable MCI and NC groups. There were no significant differences between the stable and progressive MCI groups with respect to sex, education, MMSE, and APOE ɛ4 status (Table 1).
Baseline CSF profiles of patients with progressive MCI were similar to those in the AD dementia group, but significantly different from those of the NC group, contrary to those in the stable MCI group patients. We did not note any significant differences in CSF Aβ1-42, t-tau, or p-tau levels between the patients with stable or progressive MCI. However, patients with progressive MCI demonstrated significantly higher t-tau/Aβ1-42 and p-tau/Aβ1-42 ratios than patients with stable MCI (p = 0.006, and p = 0.005, respectively; Table 1). Direct comparison of progressive and stable MCI patients revealed that the former had significantly higher CSF t-tau (p = 0.005) and p-tau (p = 0.027) levels and t-tau/Aβ1-42 (p = 0.005) and p-tau/Aβ1-42 (p = 0.032) ratios than the latter.
Cox proportional hazard model analysis showed that the baseline CSF Aβ1-42 positivity was associated with a 5.6-fold greater MCI to AD conversion risk over a mean follow up of 14.5 months than the CSF Aβ1-42 negativity, despite adjusting for age, sex, and education (HR = 16.4, p = 0.006). However, an amyloid positive baseline scan could not be considered to be a significant conversion risk than an amyloid negative baseline scan. After adjusting for age, sex, education, and APOE ɛ4 status, we observed that the CSF t-tau level was the only variable that predicted conversion (hazard ratio, HR = 94.3, p = 0.035) (Table 6).
Cox regression analysis for AD dementia conversion from MCI*
AD, Alzheimer’s disease; APOE, apolipoprotein E; CSF, cerebrospinal fluid; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; PET, positron emission tomography; P-tau, phosphorylated tau; SUVR, standardized uptake value ratio; T-tau, total tau. *Twenty-two MCI patients for CSF biomarkers and 10 MCI patients for both amyloid PET visual reading and SUVR were included in the analysis.
DISCUSSION
The main findings of the present study were 1) the cut-off values for CSF biomarkers, specifically, t-tau/Aβ1-42 and p-tau/Aβ1-42 ratio, that we established accurately differentiated AD dementia from NC, and AD dementia from OND; 2) CSF Aβ1-42 levels and p-tau/Aβ1-42 ratios were correlated with global SUVR values of amyloid PET; 3) CSF biomarkers were associated with hippocampal volume and memory function in patients with AD dementia and MCI; 4) there was relevant concordance between CSF Aβ1-42 and visual read amyloid PET (77%) and the discordant patients were predominantly positive for CSF Aβ1-42 and negative by PET, and 5) baseline CSF Aβ1-42 positivity resulted in a 5.6-fold greater MCI to AD dementia conversion hazard than the baseline CSF negativity, contrary to the baseline amyloid PET positivity. Furthermore, a baseline CSF t-tau level best predicted MCI to AD dementia conversion after adjusting for age, sex, education, and APOE ɛ4 status.
Median values for CSF Aβ1-42, t-tau, and p-tau in patients with AD dementia were comparable to those of previous studies that used the same ELISA kit and reported ranges of Aβ1-42 from 316 to 487 pg/mL, t-tau from 425 to 604 pg/mL, and p-tau from 73 to 83 pg/mL [37–41]. Cut-off values for each CSF biomarker that differentiated AD dementia and NCs were within the previously observed ranges (from 481 to 643 pg/mL for Aβ1-42, from 252 to 375 pg/mL for t-tau, and from 52 to 57 for p-tau) [37, 42]. Although studies typically vary with regard to pre-analytic or analytic differences [43], we followed previously standardized methods [5, 42] and one technician conducted all of the sample analyses using the same ELISA kits to reduce bias.
Best diagnostic accuracy in differentiating between AD dementia and NC were ensured using p-tau/Aβ1-42 and t-tau/Aβ1-42 ratios, which was responsible for the identification of more than 90% of the AD dementia patients. This result was considerably comparable to that of a recent study [44] that supported the use of a combinational approach involving two different pathological processes to reduce the variances associated with the diagnosis of AD dementia [44–46].
Additionally, we investigated the differentiation of AD dementia and non-AD neurodegenerative diseases. Despite the fact that CSF Aβ1-42 levels in NPH and AD dementia group patients were not significantly different, NPH patients demonstrated lower levels of the CSF Aβ1-42 than those in the NC, MCI, and OND groups, which was consistent with previous studies [47, 48]. The mechanism responsible for low CSF Aβ1-42 in NPH is attributed to impaired clearance due to a restricted extracellular space and not the deposition of Aβ plaques in the brain [49].
All CSF biomarkers for the FTD patients differed significantly from those in the AD dementia group, and p-tau/Aβ1-42 ratio most accurately differentiated AD dementia and FTD. Similarly, studies have previously reported that Aβ1-42/p-tau or Aβ1-42/t-tau levels can successfully differentiate between AD dementia and FTD [50, 51].
The DLB group showed mean CSF Aβ1-42 levels were between those in the NC and AD dementia groups. Previous studies have suggested that low CSF Aβ1-42 levels in DLB patients may be associated with an AD co-pathology [52, 53]. Although CSF p-tau levels have reportedly been considered to successfully differentiate AD and DLB [54], we noted that all CSF biomarkers, including CSF p-tau, could only modestly differentiate between the two.
In PD patients, all biomarker levels were similar to those of NCs and showed high specificity for the differentiation of AD dementia.
In more recent studies, CSF cut-off values were determined using amyloid-PET positivity as a reference to avoid the use of conservative CSF Aβ1-42 cut-off values [55, 56]. However, this method primarily depends on the use of another biomarker (amyloid PET), which is both inefficient and expensive. Therefore, we calculated cut-off values for CSF biomarkers based on clinical diagnoses, which is considered to be a more relevant and applicable approach in clinical settings. In addition, tau information can be obtained from CSF, while PET imaging is restricted to the visualization of amyloid pathologies.
We observed that amyloid PET global SUVR values were inversely correlated with CSF Aβ1-42 levels, which concurred with the outcomes of previous studies [13, 58]. Furthermore, CSF Aβ1-42 levels were correlated with regional SUVRs in the frontal, parietal, temporal and occipital cortices, anterior cingulate, precuneus, putamen, and pallidum, but not with those in the posterior cingulate, caudate, thalamus, and hippocampus. Although the reasons for the lack of associations in these regions remain unclear, it is imperative to note that technical errors can easily occur in small or atrophied areas [59]. In addition, few studies have directly assessed associations between subcortical amyloid accumulations and CSF Aβ1-42 levels [58].
All CSF biomarkers were distinctly correlated with the hippocampal volume in a combined MCI and AD dementia sample, although these associations did not hold after dichotomizing the patients through diagnostic category. The tau/Aβ1-42 ratios were correlated with hippocampal volume in the MCI group, contrary to that in the AD dementia group. Our results were partially concordant with those of Apostolova et al., who reported significant correlations between all CSF biomarkers and hippocampal volume in pooled samples, and strong correlations or trends between CSF tau-related biomarkers and hippocampal volume in AD dementia and MCI group, respectively [60]. Leonardo et al. reported that the CSF tau levels, but not the CSF Aβ1-42 levels, were correlated with the hippocampal volume in AD dementia. This outcome supported the hypothesis that CSF tau markers reflect neuronal injury caused by the formation of neurofibrillary tangle [61]. The discrepancy between Leonardo et al.’s and our study may have been caused by methodological differences. Leonardo et al. included AD dementia patients diagnosed based on CSF AD profiles; additionally, the mean time between CSF collection and MRI scan in their study was shorter (15±59 days) than ours (37±57 days).
Interestingly, all five CSF biomarkers were correlated with the memory domain only in the combined MCI and AD dementia group, but were not correlated with cognitive scores in the MCI or AD dementia group. These results were concordant with previous studies that reported significant correlations between cognitive scores and CSF biomarkers in a combined group (a NC, MCI, plus AD dementia group) and not in individual diagnostic groups [62]. There have been numerous inconsistencies in the results reported by previous studies on CSF biomarkers and cognitive function. Some reported CSF Aβ1-42 levels did not demonstrate significant correlations with cognitive severity, but that CSF t-tau or p-tau levels were highly correlated with cognition in AD dementia, which supported the notion that CSF tau is a direct marker of neuronal injury [63]. These findings can also be explained by a ceiling effect observed for CSF Aβ1-42 level, which reached a peak plateau before the MCI stage [64]. Furthermore, the amyloid burden has demonstrated a consistently significant association with episodic memory in cognitively healthy older adults [65]. However, other studies that used a variety of statistical methods and neuropsychological tests have shown variations in the relationships between CSF Aβ1-42 or tau levels and memory and frontal executive functions in AD dementia [66, 67].
In the present study, overall concordance between with CSF and amyloid PET markers, that is, CSF + /PET+ or CSF-/PET- was 77%. However, the reported rates were highly variable (57–92%) [15, 68–70]. We observed that all concordant positive (CSF + /PET+) subjects had clinically diagnosed MCI or AD dementia, as was expected; however, the CSF-/PET- patients also included those with MCI (n = 8) or AD dementia (n = 3), indicating that the cognitive impairment in these patients was etiologically associated with a suspected non-AD pathology (SNAP). However, after applying more lenient cut-off 95% CIs (597.6–757.5) for CSF Aβ1-42, two of the 11 patients with presumed SNAP were reclassified as CSF + /PET-, and one of the two had high CSF p-tau and t-tau levels, implying the possibility of an underlying AD pathology. With regard to the 18 discordant cases, the majority (n = 16) were CSF + /PET- and half of them (n = 8) demonstrated clinical MCI and AD dementia. Among which, seven had positive CSF tau markers for AD dementia, which supported the hypothesis that abnormal CSF Aβ1-42 (the accumulation of oligomeric forms) precedes a positive amyloid PET finding (the accumulation of fibrillary forms) [64]. Of the other 8 subjects (3 svPPA, 1 bvFTD, 1 SMI, 1 NPH, 1 nfvPPA, 1 PSP-RS) with CSF + /PET-, six had low CSF p-tau levels, which suggested the possibility of a non-AD pathology altering the process of different Aβ species [57, 71].
Only two patients were CSF-/PET+ (n = 2). One was a clinical AD dementia patient with a borderline CSF Aβ1-42 level and exceptionally high CSF t-tau and p-tau levels, which suggested that dementia in this patient was associated with an AD pathology. Therefore, there is a need for a more lenient cut-off value for CSF biomarkers during clinical diagnostic workups [69]. Another CSF-/PET+ patient with NPH had a relatively high CSF Aβ1-42 level and low CSF t-tau and p-tau levels, which may suggest methodological errors in visual reads of amyloid PET caused by ventriculomegaly [72].
In this study, baseline CSF Aβ1-42 was better at predicting AD dementia conversion than amyloid PET. Since low CSF Aβ1-42 (soluble oligomeric form) may represent an earlier signal than increased cortical amyloid deposition (fibrillar component of Aβ) detected by PET, CSF Aβ1-42 may be a more sensitive marker for disease progression in patients with MCI [64]. However, a number of studies concluded that amyloid PET better predicted progression to AD dementia than CSF Aβ1-42 [70, 73]. The difference could be explained by different number of subjects examined and the different methods used to determine CSF Aβ1-42 levels (INNOTEST versus INNO-BIA AlzBio3 kit) and PET imaging (11C-PIB versus 18F-flutemetamol versus 18F-flurbetaben versus 18F-fluorbetapir). We also found that CSF t-tau best predicted MCI progression adjusted after APOE ɛ4 status. In MCI, CSF t-tau or p-tau levels have been consistently better predictors of progression to AD dementia than CSF Aβ1-42 levels [74–77].
Our study has several limitations. First, its sample size was relatively small in comparison with those of other large-scale multicenter studies. Therefore, our findings need to be validated by testing a much larger number of patients. The second limitation is the lack of histopathologic confirmation of AD. Nevertheless, given that the CSF biomarker use is not standardized across laboratories, we considered it meaningful to establish a cut-off clinical value of CSF AD biomarkers in a well-characterized, homogeneous, and mono-center population, before universal diagnostic cut-off levels are established by fully automated laboratory units [78]. Third, we used visually rated images when dichotomizing amyloid PET images as abnormal or normal, and two amyloid tracers (18F-florbetaben and 18F-flutemetamol). It has been reported visual analysis and SUVR Aβ PET analyses are equivalent in terms of predicting MCI to AD conversion [79], and in our sample, the mean SUVR value of visually amyloid negative patients was significantly lower (n = 7, mean 1.15, SD 0.13) than that of amyloid positive patients (n = 28, mean 1.45, SD 0.17) (p < 0.001). The Centiloid scale (a standardized, quantitative amyloid imaging measurement scale) has been recently used to allow comparisons between results obtained using different tracers, but application of this scale requires 3D T1 image co-registration. Unfortunately, this scale could not be used in the present study because only 67% of the patients taking amyloid PET also underwent 3D T1 imaging. However, visual reads or even the use of SUVRs obtained using multiple tracers, are widely accepted for analyses [80, 81]. Lastly, the short MCI follow-up period used coupled with a small sample size might have affected results related to the prediction of MCI to AD conversion.
Despite these limitations, our study demonstrates good diagnostic accuracies for CSF biomarkers, relevant concordance between CSF amyloid markers and amyloid PET, and better prognostic performances of baseline CSF Aβ1-42 and t-tau than amyloid PET in MCI. In light of these results and accumulating evidence that CSF Aβ1-42 and amyloid PET represent different aspects of cerebral Aβ pathology [15, 82], we suggest the interchangeability of these two amyloid biomarkers may be viewed cautiously.
Footnotes
ACKNOWLEDGMENTS
This research was supported by a fund (2018-ER6204-00) by Research of Korea Centers for Disease Control and Prevention and Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2016R1D1A3B01007733).
