An examination of a novel multipanel of CSF biomarkers in the Alzheimer's disease clinical and pathological continuum

2.1. Participants

Using a uniform preanalytic protocol across included longitudinal studies, CSF was obtained from N = 720 adults ages 45 to 93 years (M = 63.9, standard deviation [SD] = 9.0; 51.6% female) participating in the Wisconsin Registry for Alzheimer's Prevention study (WRAP, n = 205), ¹³ the Wisconsin Alzheimer's Disease Research Center (WADRC, n = 411), or affiliated studies (Statins in Healthy, At‐Risk Adults: Impact on Amyloid and Regional Perfusion [SHARP, n = 63; NCT00939822]; the Alzheimer's Disease Connectome Project [ADCP, n = 9]; Fitness Aging in the Brain, [FAB, n = 40]). Enrollment criteria varied across studies (see supporting information). The combined sample includes cognitively unimpaired (CU) individuals, participants with mild cognitive impairment (MCI) or dementia due to suspected AD, and is enriched for parental history of AD dementia (determined through review of parental medical records, autopsy reports, results of a dementia questionnaire, or participant self‐report). All participants had decisional capacity and completed an informed consent process before undergoing study procedures. Lumbar punctures (LPs) were performed within 1 year of cognitive testing. If participants completed multiple LPs, their most recent LP was selected for analysis.

2.2. Clinical diagnosis

WRAP, WADRC, FAB, and ADCP participants’ cognitive performance and functional status were adjudicated by consensus conference at each visit. Diagnoses of MCI or dementia due to suspected AD were assigned based on National Institute on Aging‐Alzheimer's Association (NIA‐AA) criteria, ¹⁴ , ¹⁵ without reference to biomarkers; n = 50 participants were diagnosed with dementia (49 suspected AD and 1 dementia‐other cause), n = 54 had MCI (47 MCI presumed due to AD and 7 MCI‐other cause), and n = 616 CU. SHARP participants’ cognition was assessed formally at pre‐study, and those with signs of cognitive impairment were excluded from enrollment.

2.3. CSF collection

CSF samples were acquired with a uniform preanalytical protocol between 2010 and 2018. Samples were collected in the morning after an 8‐ to 12‐hour fast using a Sprotte 24‐ or 25‐gauge atraumatic spinal needle and 22 mL of fluid was collected via gentle extraction into polypropylene syringes and combined into a single 30 mL polypropylene tube. After gently mixing, samples were centrifuged to remove red blood cells or other debris; 0.5 mL CSF was aliquoted into 1.5‐mL polypropylene tubes and stored at −80^○C within 30 minutes of collection (see supporting information for details).

2.4. CSF assays

All CSF samples were re‐assayed at the Clinical Neurochemistry Laboratory, University of Gothenburg, using the same batch of reagents, under strict quality control procedures. The following immunoassays were performed on a cobas e 601 analyzer: Elecsys β‐amyloid(1‐42) CSF, Elecsys Phospho‐Tau (181P) CSF and Elecsys Total‐Tau CSF, S100 calcium binding protein B (S100B), and interleukin‐6 (IL6). The remaining NTK panel was assayed on a cobas e 411 analyzer including Aβ(1‐40) CSF, markers of synaptic damage and neuronal degeneration (neurogranin, neurofilament light protein [NfL], and α‐synuclein), and markers of glial activation (glial fibrillary acidic protein [GFAP], chitinase‐3‐like protein 1[YKL‐40], and soluble triggering receptor expressed on myeloid cells 2 [sTREM2]).

2.5. Amyloid PET imaging

A subset of 185 participants also underwent dynamic [C‐11]Pittsburgh compound B (PiB) PET imaging (0–70 minutes post‐injection) within 2 years of their most recent LP (mean time between PiB and LP was 0.35 ± 0.71 years). Imaging methods and PiB quantification have been previously described. ¹⁶ PiB(+/–) status was determined by visual inspection inter‐rater reliability = 0.95, intra‐rater reliability = 0.96. ¹⁶

2.6. Biomarker positivity

We used receiver operating characteristic (ROC) analyses to derive positivity thresholds for AD biomarkers (ADB) using PiB(+/–) as the standard of comparison. ROC analyses were conducted using the MatLab perfcurve function (The Mathworks, Natick, Massachusetts, USA). The optimal threshold for Aβ_42/40 and pTau₁₈₁/Aβ₄₂ discrimination was based on equally weighted cost functions for positive and negative agreement. ¹⁷ Due to the greater availability of Elecsys® IVD immunoassays in clinical settings, pTau₁₈₁/Aβ₄₂ was used in analyses requiring continuous measures of ADB, and pTau₁₈₁/Aβ₄₂ positivity status was used for analyses with dichotomous ADB status (ADB[+/–]).

Thresholds for pTau₁₈₁, tTau, NfL, neurogranin, and α‐synuclein status were determined by establishing a reference group of 223 CSF amyloid (Aβ_42/40) negative, cognitively unimpaired younger participants (ages 40–60 years). Biomarker positivity thresholds for these analytes were set at +2SD above the mean of this reference group. ¹⁸

2.7. Cognitive outcomes

The primary cognitive outcome was clinical diagnosis. As a secondary cognitive outcome, we examined the cross‐sectional relationship between biomarkers and cognitive performance, using a three‐test Preclinical Alzheimer Cognitive Composite (PACC3) described by Jonaitis et al. ¹⁹ and based on the work of Donohue et al. ²⁰ Due to variations in cognitive batteries across cohorts, Trail‐Making Test B replaced Digit Symbol as the executive function measure and Craft Story Delayed Recall was used to impute Logical Memory II‐A based on a published crosswalk. ²¹ Continuous cognitive outcomes were matched to the nearest LP visit. Only matches less than a year apart were included, and no cognitive visit was matched more than once.

2.8. Statistical analysis

Statistical analyses were conducted in R. ²² Sample characteristics were compared across clinical diagnosis using analysis of variance for continuous measures and chi‐square for categorical measures. Associations between CSF values and clinical severity were tested with linear regressions. The R package emmeans 1.4.3.01 ²³ was used to compare mean differences among groups defined by ADB status and clinical diagnosis.

We evaluated the potential added explanatory value of exploratory NTK biomarkers when modeling cognitive outcomes using categorical (clinical diagnosis) and continuous (PACC3) measures of cognition. Observations were excluded if they were missing ADB or NTK biomarkers, or any covariates (n = 47). SHARP participants (n = 66) received a different cognitive battery and were excluded from analyses of continuous cognitive performance. Logistic regression was used to model the relationship between clinical diagnosis (pooled MCI and dementia vs CU), continuous ADB, and additional NTK biomarkers for neurodegeneration or glial activation. Linear mixed effects regression was used to model the relationship between continuous cognitive performance (PACC3), continuous ADB, and additional NTK markers. Models were fit via maximum likelihood estimation. A likelihood ratio test (LRT) was used to compare larger models containing neurodegeneration and gliosis markers, respectively, with a reduced model including only continuous ADB and key covariates, age at LP, sex, apolipoprotein ε4 (APOE4) carrier status, and years of education. Due to the nature of the study, we did not correct p‐values for multiple testing.

3.1. Sample characteristics and CSF analytes

Sample characteristics are shown by clinical diagnosis in Table 1. Participants were aged 40 to 93 years (M = 63.9, SD = 9.0), mostly white, and highly educated. Cognitively unimpaired participants were younger, more educated, and less likely to carry the APOE4 risk allele compared to impaired groups. Performance on the Mini‐Mental State Examination (MMSE) and the Clinical Dementia Rating Scale Sum of Boxes (CDR‐SB) tracked with diagnostic category, as expected (Table S3 in supporting information).

3.2. Biomarker positivity thresholds

3.3. CSF analyte observations by ADB status

Scatterplots and correlations between CSF analytes for biomarker groups (AD, neurodegeneration, and glial activation) are shown by ADB status in Figure 1A‐C (See Figure S1 in supporting information for correlations between all CSF analytes). Correlations between Aβ₄₂, Aβ₄₀, pTau₁₈₁, and tTau were typical of those observed in AD (Figure 1A). ²⁴ Due to the high correlation between pTau₁₈₁ and tTau (r = .98), tTau was excluded from subsequent regression analyses with clinical diagnosis and cognition.

Correlation patterns for CSF analytes related to neurodegeneration and glial activation were consistent across ADB status for all NTK analytes. All neurodegeneration markers (Figure 1B) correlated highly with tTau (range r = .62–.87). Neurogranin was highly correlated with α‐synuclein (r = .81), while NfL was only moderately correlated with α‐synuclein (r = .50) and neurogranin (r = .38). Glial activation biomarkers (Figure 2C) were all modestly inter‐correlated (r = .22–.62) with S100B showing the lowest correlation with other glial activation markers. IL6 values were unrelated to the remaining analytes (Figure S1).

3.4. CSF analyte observations by clinical diagnosis and ADB status

Descriptive statistics for all CSF analytes and derived ratios stratified by clinical diagnosis and ADB status are shown in Table 2. Distributions of analytes are shown in Figure 2A‐D. Aβ₄₀, Aβ₄₂, and pTau₁₈₁ (Figure 2, panel A) exhibited the expected distributions for combinations of clinical and ADB status. Aβ₄₀ did not differ across clinical or ADB groups. Aβ₄₂ was lower for all ADB+ and did not differ between clinical groups. Phospho‐Tau₁₈₁ was low in all ADB–, was higher in unimpaired ADB+, and was highest in impaired (MCI and dementia) ADB+.

Neurodegeneration analytes (Figure 2, panel B) showed similar patterns between ADB and clinical status groups for neurogranin and α‐synuclein. These markers did not differ across clinical groups in ADB– and were higher in ADB+ compared to ADB– both within and across clinical groups (not enough ADB– dementia cases for comparison). NfL indicated stepwise increases in ADB+ with increasing clinical severity.

CSF analytes of glial activation YKL‐40, S100B, GFAP, and sTREM2 (Figure 2, panel C) exhibited similar patterns in ADB+ wherein impaired ADB+ had higher values compared to unimpaired ADB+. In general, these analyte distributions had considerable overlap between ADB+ and ADB– in the unimpaired group. YKL‐40 and GFAP were higher for ADB+ compared to ADB– in the MCI group. IL6 (Figure 2, panel D) was unrelated to ADB status or clinical group.

3.5. Relationships between cognitive outcomes and extended NTK analytes

Continuous ADB significantly predicted clinical diagnosis and cognitive performance (Ps < .001). Adding gliosis biomarkers did not improve model fit for either clinical status or PACC3 (clinical status: χ²[4] = 4.0, P = .41; PACC3: χ²[4] = 6.7, P = 0.15). For both clinical status (Table 3) and PACC3 (Table 4), adding neurodegeneration biomarkers improved the overall model fit when compared to a model that included continuous ADB and covariates (clinical status: χ²[3] = 17.3, P = .0006; PACC3: χ²[3] = 23.5, P = .000032). The regression coefficient for neurogranin was opposite in sign to our expectation. Secondary analyses of individual neurodegeneration biomarkers suggested that this was an artifact of statistical suppression. ²⁵ As individual biomarkers, NfL best predicted clinical status and PACC3 over and above continuous ADB. To visualize the latter findings, we plotted a loess curve of PACC3 against NfL grouping by ADB status (Figure S2 in supporting information).

4.1. Biomarker positivity

4.2. Interrelationship between neurodegenerative analytes and clinical diagnosis/ADB status

As noted in the NIA‐AA research framework, ⁶ neurodegeneration is a non‐specific feature of several neurodegenerative diseases. Because we ³¹ and others ³³ have observed remarkably high agreement between tTau and pTau₁₈₁, CSF tTau does not appear to be a fully independent measure of neurodegeneration in AD. The other NTK markers may serve an important need in this regard. Indeed, NfL, an indicator of axonal degradation, has been used as a useful neurodegeneration marker in multiple sclerosis, non‐AD tauopathies, synucleinopathies, ³⁴ and traumatic brain injury ³⁵ as well as AD. ³⁶ NfL is also in agreement with magnetic resonance imaging metrics in this population ³⁷ and correlates with pre‐dementia disease progression. ³¹ , ³⁸ In the present analyses, NfL, neurogranin, and α‐synuclein exhibited moderate to strong agreement with tTau and at least moderate agreement with each other, suggesting these markers of neurodegeneration are reflecting common aspects of neurodegeneration. Further, they exhibited elevation within diagnostic stage by ADB status or significant elevation differences across diagnoses (as shown in Figure 2). NfL exhibited the most characteristic stepwise increase across clinical diagnosis in AD biomarker positive subjects and this was further evident in Figure S2 in which a steeper relationship between lower cognitive scores and NfL was observed among ADB+ than ADB– participants. In contrast, neurogranin, a post‐synaptic protein marker, was significantly elevated in the cognitively unimpaired group who were ADB+ and remained elevated across clinical diagnoses consistent with our prior observations. ³¹ , ³⁸ Total α‐synuclein, a presynaptic marker, exhibited a similar pattern and was also strongly correlated with neurogranin (r = .80). CSF α‐synuclein was initially found to be slightly decreased in Parkinson's disease and Lewy body dementia, but subsequent studies showed a pronounced increase in CSF α‐synuclein in neurodegenerative disorders with marked neurodegeneration, including Creutzfeldt‐Jakob disease and AD. ³⁹ Elevation of this protein in our sample likely reflects synaptic degeneration rather than deposition of α‐synuclein in Lewy bodies. Its utility as a novel marker of neurodegeneration continues to undergo study.

Although promising as continuous markers of neurodegeneration (N), among AD and CU participants a lower proportion were identified as positive when defined by NFL, neurogranin, or α‐synuclein compared to tTau. Agreement across neurodegeneration biomarkers was moderate. The method for choosing thresholds for these analytes (2 SD above the mean of a CU Aβ_42/40 negative group) is a reasonable approach but assumes a monotonic relationship between age and biomarker concentration. Ongoing work in the field will lead to more precise methods for defining a meaningful threshold for N+/–.

4.3. Interrelationship between inflammation and gliosis analytes and clinical diagnosis/ADB status

Activation of microglia in response to amyloid plaques is a well‐known feature of AD, and inflammatory pathways have been shown to play a role in AD pathogenesis. ⁴⁰ , ⁴¹ Despite the involvement of inflammatory processes in AD pathophysiology, IL6 was unrelated to either markers of glial activation or clinical diagnosis, and may be more relevant at a more advanced disease state. ⁴² Markers of glial activation exhibited low to moderate intercorrelation, indicating potentially unique physiologic meaning of each analyte. YKL‐40, a glycoprotein expressed by microglia and astrocytes, and GFAP, an indicator of reactive astrocytes, ⁴³ were both elevated in ADB+ cognitively impaired subjects compared to their biomarker negative peers. The YKL‐40 finding replicates previous observations. ⁴⁴ From Figure 2, the effect sizes of glial and microglial markers observed here appear lower than for the neurodegeneration markers. Nevertheless, these results are promising and warrant further study, particularly in the context of co‐occurring diseases.

4.4. Core AD biomarkers and cognition

Before examining the effect of the NTK panel, we first confirmed that AD biomarkers were related to cognition defined by clinical diagnosis and global cognitive performance. Aβ₄₂ alone predicted impairment, but did not distinguish between MCI and AD, perhaps due to the well‐known observation that levels of this protein plateau by the dementia stage. ⁴⁵ , ⁴⁶ Normalizing against total amyloid production (ie, by the Aβ_42/40 ratio), led to clear differentiation by clinical diagnosis, as did pTau₁₈₁ and pTau₁₈₁/Aβ₄₂.

4.5. NTK biomarkers and cognition

Results from hierarchical regression analyses suggest that as a group, neurodegeneration biomarkers add value in predicting both clinical impairment (MCI/dementia vs CU) and global cognitive performance. The information in these markers is overlapping, as can be seen by the suppression effects observed in the full model; however, their relatively moderate concordance with one another indicates that the overlap is only partial. Of the three neurodegeneration markers, NfL appears to have the best concordance with clinical diagnosis. In contrast, adding gliosis markers to the regression did not improve model fit for either clinical status or global cognition, which suggests these markers do not explain additional variance in cognition beyond core AD markers. This conclusion must be tempered by the constraint of the study design. It is possible that glial markers may exhibit effects in certain contexts, such as the presence of vascular disease, ¹⁰ or their effects are non‐linear across disease state. ⁴⁷

4.6. Limitations

Although the development of immunoassays has the potential to greatly reduce assay variability, raw values, particularly for Aβ, may still be affected by preanalytic fluid collection protocols, which vary across studies. ⁴⁸ As such, the values and cut points described here may not generalize to studies in which preanalytic protocols differ. Our cohort is typical of dementia research (white, educated, relatively high functioning) but may not represent the typical AD patient, or individuals from higher risk populations like African Americans and Latinx. Results need to be interpreted in light of this limitation. Although the time interval between LP and PET imaging was up to 2 years, given the stability of amyloid in the brain we do not anticipate a smaller time interval would change our findings.