Clinical utility of Cerebrospinal Fluid Biomarkers for the Diagnosis of Alzheimer’s Disease in Korea

Cerebrospinal uid (CSF) biomarkers are increasingly used in clinical practice for the diagnosis of Alzheimer’s disease (AD). We aimed to 1) determine cutoff values of CSF biomarkers for AD, 2) investigate their clinical utility by estimating a concordance with amyloid positron emission tomography (PET), and 3) apply AT (amyloid/tau) classication based on CSF results. We performed CSF analysis in 51 normal controls (NC), 23 amnestic mild cognitive impairment (aMCI) and 65 AD dementia (ADD) patients at the Samsung Medical Center in Korea. We tried to develop cutoff of CSF biomarkers for differentiating ADD from NC using receiver operating characteristic analysis. We also investigated a concordance between CSF and amyloid PET results and applied AT classication scheme based on CSF biomarker abnormalities to characterize our participants. apolipoprotein


Introduction
Alzheimer's disease (AD), which is characterized by pathological hallmarks of β-amyloid (Aβ) and neuro brillary tau tangles, is a continuum disorder largely with three different clinical stages: normal cognition (NC, or preclinical), mild cognitive impairment (MCI, or prodromal), and dementia [1][2][3]. Since previous diagnostic criteria for AD were based mainly on clinical manifestations rather than on biomarkers, early diagnosis for preclinical or prodromal AD was di cult. However, advances in AD biomarker development using positron emission tomography (PET) or cerebrospinal uid (CSF) have enabled in vivo identi cation of pathologies, and therefore early diagnosis of patients even in the preclinical or prodromal stages has become possible. More recent research criteria for AD proposed by the National Institute on Aging and Alzheimer's Association (NIA-AA) task force incorporated both Aβ phenotyping and clinical status to characterize patients [3][4][5]. Further, a new research framework using a multiple-biomarker-based de nition of AD in living patients was recently proposed [6,7]; it included biomarkers for Aβ (A), pathologic tau (T), and neurodegenerative/neuronal injury (N), called the AT(N) system 4 . Individuals with preclinical AD or prodromal AD show changes in CSF amyloid. Speci cally, a decreased level of β-amyloid 1-42 (Aβ42) and increased level of total tau (t-tau) and phosphorylated tau (p-tau) in CSF are useful for the early diagnosis of AD. CSF biomarkers may have several advantages over amyloid PET. First, CSF tests are less expensive and more accessible and have no radiation exposure issue compared to amyloid PET. Second, prior studies showed that changes in CSF Aβ42 precede amyloid PET uptakes [8,9], suggesting that CSF biomarkers may be more sensitive in identifying individuals with preclinical AD. Third, the most noteworthy bene t of CSF biomarkers is that CSF gives more information about AD-related pathologies than does amyloid PET. That is, CSF biomarkers provide not only amyloid status (disease marker) but also information on the tau pathology biomarker (phosphorylated tau) and neurodegenerative marker (total tau) as well. Finally, as novel biomarkers such as neuro lament or neurogranin are being developed[10-13], one lumbar puncture can provide a variety of information encompassing AD pathology or other neurodegeneration in the future.
One issue with CSF biomarkers, however, is that there exists variability in cutoff values between laboratories. We need development of laboratory-speci c cutoff values to enable application of these values in clinical judgement. In line with this need, a recent study using multicenter CSF data from Korea reported new cutoff levels of CSF AD biomarkers and compared them with those from Western studies [14]. However, this study included participants based on clinical diagnosis only; investigating the concordance between CSF and PET positivity or accuracy of CSF biomarkers for detecting amyloid pathology was unavailable. Besides, the cutoff values developed in this study were intended to differentiate ADD from other neurodegenerative disorder. Considering that other neurodegenerative disorders may have concomitant amyloid pathology, developed cutoff values might not be pathology speci c. Therefore, in this study, we rst tried to investigate CSF biomarkers across three different cognitive groups (NC, MCI, and AD dementia [ADD]) in a single large center in Korea, in order to establish cutoff values of CSF biomarkers that could differentiate ADD from NC. Second, we investigated the concordance of CSF biomarkers and amyloid PET results to con rm the reliability of CSF biomarkers. Finally, we attempted to apply the AT biomarker classi cation system to our patients based on CSF biomarker cutoffs derived from our study, which would have clinical implications, in that both amyloid and tau information rather than amyloid alone can provide more accurate prediction of patients' prognoses.

Study participants
In this study, we included a total of 400 participants who had undergone CSF analysis at the Samsung Medical Center, Seoul, Korea, from February 2011 to July 2019. Participants received a detailed dementia evaluation, including clinical history, neurologic examination, neuropsychological tests, blood tests to exclude secondary causes of dementia, brain magnetic resonance imaging (MRI), and amyloid PET and APOE genotyping. Out of the initial 400 participants, we included only individuals with NC, amnestic MCI (aMCI), or ADD; we excluded 224 patients with other diagnoses, such as normal pressure hydrocephalus (NPH), subcortical vascular cognitive impairment (SVCI, cognitive impairment associated with severe deep white matter hyperintensities [WMH], which was de ned as a deep WMH lesion ≥ 25 mm and periventricular WMH lesion ≥ 10 mm), neurodegenerative disorders with parkinsonism, traumatic brain injury, frontotemporal dementias, brain tumor, neurosyphilis, CO poisoning, epilepsy, autoimmune encephalitis, or Creutzfeldt-Jakob disease. When ADD, aMCI, or NC participants showed a gait disturbance and an Evan's ratio over 3.0 on MRI, they were referred to as ADD, aMCI, or NC with "combined hydrocephalus" and excluded from this study (n = 37).
ADD was clinically diagnosed when patients met the diagnostic criteria for probable ADD according to NIA-AA criteria, with cognitive impairment, such as memory loss, word-nding di culties, or visual/spatial problems signi cant enough to impair a person's ability to function independently [15]. AMCI was diagnosed according to Peterson's MCI criteria [16], which was characterized by memory complaints usually corroborated by an informant, objective memory impairment for age, essentially preserved general cognitive function, and largely intact functional activities. We recruited NC participants from our memory or headache clinic or orthopedic surgery department, where they received spinal anesthesia. All the NC individuals met the following criteria: no history of neurological or psychiatric disorders, normal cognitive function, normal activities of daily living with or without subjective memory complaints, and normal CSF cell count.
We obtained written informed consent for obtaining CSF data from each patient, and this study was approved by the Institutional Review Board at the Samsung Medical Center.

CSF collection and AD biomarker analysis
For most participants, CSF samples were collected from a lumbar puncture done in the L3-4 or L4-5 intervertebral spaces using a 20 or 22G needle[17]. Fasting was not required. A subset of 28 AD and 12 aMCI patients had Ommaya reservoir insertion for a clinical trial of intraventricular stem-cell injection, 'Safety and Exploratory E cacy Study of NEUROSTEM® Versus Placebo in Patients with Alzheimer's Disease[18]', and the CSF obtained from the Ommaya reservoir before stem-cell injection was used for the current study. All CSF samples were collected into 15-ml polypropylene tubes at the time of the tap and were then sent to Samsung Medical Center laboratory within 30 minutes after collection. After samples were centrifuged at 2000 g for 10 minutes within 4 hours after collection, aliquots (1 ml) prepared from these samples at room temperature were immediately stored in bar-code-labeled polypropylene vials at − 70 °C. In our laboratory, we run assays for CSF biomarkers once CSF samples are collected from 30 to 40 patients, using INNOTEST enzyme-linked immunosorbent assay (ELISA) kits (Fujirebio Europe N.V.). The CSF biomarkers included levels of Aβ42 (Amyloid-β ( 1− 42 )), t-tau, and p-tau (181 phosphorylated tau).

Aβ PET acquisition and de nition of Aβ positivity
A total of 90 patients underwent Aβ PET (PiB, n = 5; orbetaben, n = 80; utemetamol, n = 5). For 11C-PiB PET, a 30-minute emission static PET scan was done 60 minutes after bolus injection of a mean dose of 420 MBq 11C-PiB into an antecubital vein. For 18F-orbetaben PET, a 20-minute emission PET scan in dynamic mode (consisting of 4 × 5 min frames) was done 90 minutes after bolus injection of a mean dose of 381 MBq 18F-orbetaben into an antecubital vein. For utemetamol PET, a 20-minute emission static PET scan in dynamic mode (consisting of 4 × 5 min frames) was done 90 minutes after bolus injection of a mean dose of 185 MBq utemetamol into an antecubital vein.
We de ned Aβ PET positivity (PET+) for the three different types of PET images as follows: (1) Global PiB SUVR (assessed from the volume-weighted average SUVR of 28 bilateral cerebral cortical VOIs) of greater than 1.5 as described in our previous study [19], (2) visual rating score on orbetaben PET of 2 or 3 on the brain Aβ plaque load scoring system[20], (3) positive visual interpretation of 18 F-utemetamol PET in any one of the ve brain regions (frontal, parietal, posterior cingulate and precuneus, striatum, and lateral temporal lobes) in either hemisphere [21].

Statistical analyses
We used ANOVA (Analysis of Variance) to compare clinical characteristics between NC, aMCI and ADD, and ANCOVA (Analysis of Covariates) to compare CSF biomarker levels with adjustment of age. We used paired t-tests to compare CSF biomarker levels between each two groups.
We did receiver operating characteristic (ROC) analyses to establish the cutoffs for each CSF biomarker (Aβ42, t-tau, p-tau) and ratio (t-tau/ Aβ42, p-tau/ Aβ42) that best differentiated clinical ADD from NC. The cutoffs for biomarkers were de ned as values that give the maximum Youden index (sensitivity + speci city − 1). We then calculated the sensitivity and speci city for each cutoff value.
We de ned CSF biomarker levels as abnormal (CSF+) when CSF Aβ4 was lower than cutoff values. Therefore, the concordance rate of amyloid PET and CSF biomarker results was calculated as the number of CSF+/PET + plus CSF-/PET-cases over the total number of participants in the analysis. All statistical tests were conducted using the SPSS version 22 (SPSS, Inc., Chicago, IL).

Concordance of CSF Aβ biomarker and amyloid PET
We further investigated the concordance of amyloid PET results and CSF Aβ biomarker cutoff-based results. Among 90 participants who underwent amyloid PET, the number of CSF + PET + and CSF-PETcases were 77 and 6, which results in a 92% (83/90) concordance rate. Among the seven discordant cases for PET and CSF Aβ42, two were PET+/CSF-and ve were PET-/CSF+. The clinical characteristics and CSF biomarker levels of the discordant cases are described in Table 3. Two PET+/CSF-cases were aMCI and ADD who showed CSF Aβ42 levels of 777.9 and 794.6, which are slightly higher than our cutoff values. Among the ve PET-/CSF + cases, two aMCI patients showed equivocal results on amyloid PET, and two were diagnosed as ADD when CSF tests were done, but turned out to have behavior variant frontotemporal dementia (bvFTD) when followed up. Another PET-/CSF + patient was diagnosed as ADD at rst, but was found to have chronic subdural hemorrhage (SDH). Abbreviations: aMCI, amnestic mild cognitive impairment; ADD, Alzheimer's disease dementia; APOE4, apolipoprotein E4 gene; PET, positron emission tomography; CSF, cerebrospinal uid; Aβ42, beta amyloid 1-42; T-tau, total tau; P-tau, phosphorylated tau; bvFTD, behavioral variant frontotemporal dementia; SDH, subdural hemorrhage AT biomarker classi cation using CSF AD biomarker cutoff We used cutoffs of CSF Aβ42 (= 667.9) and p-tau levels (= 61.82) to de ne abnormal Aβ (A+) and abnormal tau (T+) to apply AT biomarker classi cation in our study subjects. As shown in Fig. 3, a majority of NC was categorized into A-T-(72%), followed by A-T+ (16%). A + T-occupied 52% of aMCI, followed by A + T+ (30%), and most ADD patients were categorized into A + T+ (56%) and A + T-(41%).

Discussion
In this study, we recruited 139 participants and compared CSF biomarker levels across three groups with different cognition levels (NC, MCI, and ADD), and we developed CSF biomarker cutoffs to differentiate ADD from NC. We also tried to investigate the reliability and clinical utility of CSF biomarkers by investigating concordance rates between amyloid PET and CSF results. Our major ndings were as follows. First, CSF biomarkers were signi cantly different between NC, MCI and ADD groups. Second, the cutoff values we obtained were accurate in differentiating ADD from NC and were highly concordant with amyloid PET results. Third, whereas amyloid PET can only categorize patients into amyloid negative versus positive groups, CSF biomarkers can further categorize patients according to their amyloid and tau status. Taken together, we consider that CSF biomarkers are useful in clinical and research eld.
The rst major nding of our study was that CSF biomarkers were signi cantly different between the three diagnostic groups. CSF Aβ42 level was different between NC and aMCI or ADD. However, it was not different between aMCI and ADD, which may be in line with the fact that CSF Aβ42 levels change before neurodegeneration and cognitive impairment occurs, as has been explained by the sigmoid-shaped trajectory of CSF Aβ42 levels [8]. Unlike CSF Aβ42, CSF t-tau, t-tau/Aβ42, and p-tau/Aβ42 ratios were signi cantly different between aMCI and ADD. These ndings are also consistent with previous notions that CSF t-tau is a well-known marker for neurodegeneration [6]. Therefore, we considered that a combination of tau and amyloid markers (such as the ratio of CSF t-tau/Aβ42 or CSF p-tau/Aβ42) could represent disease progression better than could an amyloid marker alone.
The second major nding of our study was that we were able to obtain cutoff values that showed high accuracy (represented as AUC) in differentiating clinical ADD from NC. Speci cally, the AUC from ttau/Aβ42 was highest (0.994), followed by those from p-tau/Aβ42 (0.963) and Aβ42 (0.960). This high accuracy may be explained by the fact that we took extra caution when selecting patients by excluding patients with possible NPH and patients with severe WMH (patients with SVCI). In fact, recent studies showed that CSF Aβ42 levels can be decreased in pure NPH. [22][23][24] T-tau also showed a satisfactory AUC of 0.918, although that was not the case with p-tau. This nding was unexpected, because CSF p-tau has been considered to be more closely related to neuro brillary tangles pathologies than is CSF t-tau [25][26][27]. However, CSF p-tau levels might change even in ADD according to the speed of progression [28]. That is, CSF p-tau may not clearly correlate with disease progression, differently from tau PET imaging showing accumulating tau pathology. Further studies to investigate longitudinal CSF p-tau change in AD are required. Alternatively, the NC group in our study might have included preclinical participants who had neuro brillary tangles (restricted to Braak stage I/II or even involving III/IV stage regions), although they did not complain of any cognitive symptoms.
The third major nding was that, when we compared the results of CSF Aβ42 with those of amyloid PET, the concordance rate was as high as 92%. When we looked at discordant cases, CSF Aβ42 levels of two PET + CSF-patients were 777.9 and 794.6, which were slightly higher than the cutoff values. Therefore, a combination with the tau biomarker could be more accurate than is CSF Aβ42 alone for diagnosis of ADD. There were also ve PET-CSF + cases, among which two were aMCI cases as expected, because CSF biomarker changes occur prior to amyloid uptake on PET. One interesting case was an 82-year-old man who was diagnosed with ADD but had a markedly high t-tau level of 1529.2, but a low p-tau of 24.1.
When we reviewed his case later because of the exceptionally high t-tau levels, we found a minimal amount of chronic SDH on his CT image conducted before CSF tapping. Therefore, we need to reconsider a diagnosis in cases where t-tau levels are higher than the p-tau levels.
Nevertheless, a high concordance between amyloid PET and CSF results has a clinical implication, in that CSF biomarkers can be an alternative to amyloid PET in clinical practice, since CSF biomarkers are even less expensive and more accessible than an amyloid PET scan in most clinical settings. Most importantly, CSF study makes it possible to detect various neurodegenerative biomarkers simultaneously once CSF is sampled. As in our study, information on amyloid and tau can be obtained from one sample of CSF, unlike a PET scan, which should be done separately to detect two different biomarkers. In addition, new CSF biomarkers, such as neuro lament or neurogranin, are increasingly considered as promising biomarkers, which suggests that the potential of new CSF biomarkers for a variety of neurodegenerative diseases may increase.
Finally, when we used our CSF cutoff values to apply AT biomarker classi cation to study participants, most of the NC were A-T-, whereas most of the AD were A + T+, as expected. Interestingly, in the MCI group, the majority of patients was classi ed into A + T-(52%), followed by A + T+ (30%). The percentages of these groups are largely comparable to those from a previous study [29]. Another feature that our study and a previous study had in common was that there was an A-T + and A-T-population in the MCI group, recapitulating the heterogeneity of MCI patients. In the era of molecular targeting treatment, delineating biomarker pro les in MCI patients may help clinicians to select the potential candidates by predicting different prognoses.

Limitations
There are limitations in our study. First, most of the NC did not undergo amyloid PET scans. Second, we had 40 participants who had CSF taken from the Omaya reservoir, but others had it taken by lumbar puncture, which disturbs the homogeneity of a single-center study. Third, one ADD patient was proven to have SDH before CSF drainage, which could disturb CSF results that were shown as a PET-CSF discordant case. Despite these limitations, considering that every center and country needs their own cutoff values, we hope that the cutoff values derived from a single center of a large memory clinic may be useful in research and clinical elds. ROC curves of CSF biomarkers for discrimination between NC and ADD abbreviations: ROC, receiver operating characteristic; CSF, cerebrospinal uid; NC, normal control; ADD, Alzheimer's disease dementia; Aβ42,beta amyloid 1-42; T-tau, total tau; P-tau, phosphorylated tau