Supplementary figure1. Levels of Aβ42, t- and p-tau as measured by ELISA assay.

Supplementary figure 2.Boxplots showing biomarker values in neuropathological cases according to Braak stages and CERAD scores. First shows Aβ42 (a), t-tau (b) and p-tau (c) values for the different CERAD scores. Second row shows Aβ42 (d), t-tau (e) and p-tau (f) values for the different Braak stages in patients with AD pathology. Third row shows Aβ42 (g), t-tau (h) and p-tau (i) values for the different Braak stages in patients without AD pathology.

Supplementary table 1.Agreement between clinical and neuropathological diagnosis.

Clinical Diagnosis
Neuropatho-
logical
diagnosis / AD / LPA / Bv FTD / CBS / PNFA / SD / FTD-MND / PSP / DLB / MCI / PPA
AD / 55 (+) / 2 (+) / 7 (-) / 3 (-) / 1 (-) / 1 (-) / 1 (±) / 1 (±)
AD-Mixed / 18 (+) / 1 (-) / 4 (-) / 2(±) / 1 (±)
FTLD / 3 (-) / 1 (-) / 12 (+) / 4 (+) / 3 (+) / 1 (+) / 1 (+) / 1 (+) / 1 (-) / 1 (±)
FTLD-Mixed / 2 (±) / 3 (+) / 1 (+)
DLB / 1 (±) / 1 (-) / 1(-) / 4 (+)
Agreement / 96.0% / 66.7% / 65.2% / 33.3% / 75.0% / 33.0% / 100.0% / 100.0% / 80.0%

Overall agreement was 81.4%. Boxes labeled as (+) indicate agreement and boxes labeled as (-) indicate wrong clinical diagnosis. Cases whose clinical diagnosis was concordant with the secondary neuropathological diagnoses are labeled as (±) and were not used for estimation of percent of agreement.

Supplementary Table 2. Association between CERAD and Braak scores with CSF biomarkers.

Assay / Braak / CERAD
p-value analyte / p-value CSF sampling-death interval / p-value analyte / p-value CSF sampling-death interval
Aβ42 (all cases) / Luminex / . / - / <0.0001 / 0.24
T-tau in cases with AD / Luminex / 0.044 / 0.22
P-tau in cases with AD / Luminex / 0.097 / 0.25
T-tau in cases without AD / Luminex / 0.78 / 0.64
P-tau in cases without AD / Luminex / 0.055 / 0.39
Aβ42 (all cases) / ELISA / 0.003 / 0.026
T-tau in cases with AD / ELISA / 0.013 / 0.020
P-tau in cases with AD / ELISA / 0.017 / 0.12
T-tau in cases without AD / ELISA / 0.85 / 0.38
P-tau in cases without AD / ELISA / 0.39 / 0.21

Ordinal logistic regression models for the biomarker values predicting the neuropathological score in an ordinal logistic regression. All models are adjusted by CSF sampling-death interval.

Supplementary table 3. Statistically different biomarker measurements between FTLD, AD and control groups in the ELISA and Luminex assays using Nemenyi-Damico-Wolfe-Dunn pot-hoc test.

Aβ42 / T-Tau / P-Tau
ELISA / Luminex / ELISA / Luminex / ELISA / Luminex
Cognitively normal controls vs. AD / <0.0001 / <0.0001 / <0.0001 / <0.0001 / <0.0001 / 0.002
Cognitively normal controls vs. FTLD / 0.005 / 0.003 / 0.905 / 0.184 / 0.343 / 0.207
AD vs. FTLD / 0.0001 / <0.0001 / <0.0001 / <0.0001 / <0.0001 / <0.0001

A non-parametric test (Kruskall-Wallis test) was used to compare CSF tau and Aβ measurements on both platforms (ELISA and Luminex) across the different neuropathological groups. P-values were corrected for multiple comparisons using Bonferroni correction and if significant the Nemenyi-Damico-Wolfe-Dunn test was used as a post-hoc procedure[1, 2].

Supplementary table 4. Area under the curve and accuracy of the single analytes and ratios and the selected combination in the training-validation sample.

Assay / ELISA / Luminex
AD vs. FTLD & Controls / FTLD vs. controls / AD vs. FTLD & Controls / FTLD vs. controls
AUC / Accuracy / AUC / Accuracy / AUC / Accuracy / AUC / Accuracy
Aβ / 78.5% (65.4%-91.7%) / 65.5% (51.4%-77.8%) / 72.1% (55.1%-89.1%) / 69.5% (52.9%-82.4%) / 93.5% (88.3%-98.8%) / 86.4% (77.0%-93.0%) / 67.6% (49.6%-85.7%) / 63.2% (46.0%-78.2%)
t-tau / 81.8% (70.7%-92.9%) / 72.7% (59.0%-83.7%) / 51.0% (31.9%-70.0%) / 61.9% (45.6%-76.4%) / 78.8% (68.9%-88.8%) / 75.3% (64.5%-84.2%) / 54.3% (34.7%-74.0%) / 68.4% (51.4%-82.5%)
p-tau / 79.6% (67.5%-91.7%) / 65.5% (51.4%-77.8%) / 58.2% (39.5%-76.9%) / 59.5% (43.3%-74.4%) / 71.8% (60.7%-82.9%) / 67.9% (56.6%-77.9%) / 87.2% (72.5%-100.0%) / 89.5% (75.2%-97.1%)
t-tau/Aβ / 88.7% (79.8%-97.7%) / 87.5% (74.8%-95.3%) / 67.3% (49.5%-85.2%) / 61.9% (45.6%-76.4%) / 90.4% (83.5%-97.4%) / 85.2% (75.6%-92.1%) / 65.4% (47.3%-83.4%) / 68.4% (51.4%-82.5%)
p-tau/Aβ / 85.6% (75.2%-96.0%) / 87.5% (74.8%-95.3%) / 64.2% (46.7%-81.7%) / 59.5% (43.3%-74.4%) / 87.0% (78.9%-95.2%) / 77.8% (67.2%-86.3%) / 84.0% (68.8%-99.2%) / 79.0% (62.7%-90.5%)
Selected model / 93.0% (86.9%-99.1%) / 86.1% (75.9%-93.1%) / 72.1% (55.1%-89.1%) / 69.5% (52.9%-82.4%) / 93.7% (87.9%-99.5%) / 90.1% (81.5%-95.6%) / 90.7% (78.0%-100.0%) / 92.1% (78.6%-98.3%)

The values in parentheses represent the 95% CI.

Supplementary table 5. Coefficients of the obtained logistic regression models obtained based on neuropathological and clinical diagnoses.

Assay / Category 0 / Category 1 / Intercept / Age / T-Tau / P-Tau / Aβ42
ELISA / AD / FTLD & Controls / 0.10 / -0.006 / 0.006
ELISA / FTLD / Controls / -1.98 / 0.005
ELISA / AD / FTLD / 1.40 / -0.014 / 0.007
Luminex / AD / FTLD & Controls / -6.41 / -0.027 / 0.038
Luminex / FTLD / Controls / -7.88 / -0.046 / 0.23 / 0.033
Luminex / AD / FTLD / -6.20 / -0.19 / 0.046
ELISA / Clinical AD / Clin. FTLD / 3.34 / -0.012 / 0.022
Luminex / Clinical. AD / Clin. FTLD / -2.64 / 0.024 / 0.057

If not stated diagnoses are based on neuropathological diagnoses.

Supplementary table 6: Comparison of classification of neuropathological cases with a single neuropathological diagnosis based on clinical diagnosis and Luminex classification.

Clinical AD / Clinical FTD
Luminex AD / 48 AD (npath. Dx) (±) / 12 AD (npath. Dx) (+)
1 FTLD (npath. Dx) (-)
Luminex FTD / 1AD (npath. Dx) (-)
3 FTLD (npath. Dx) (+) / 10 FTLD (npath. Dx) (±)

(±): No difference in diagnostic accuracy between clinical diagnosis and Luminex classification.

(+):Luminex classification improves diagnostic accuracy.

(-):Luminex classification worsens diagnostic accuracy.

npath. Dx: neuropathological diagnosis.

References

1.Hollander M and Wolfe DA (1999) Nonparametric Statistical Methods, 2nd Edition. 1999, New York: John Wiley & Sons.

2.Hothorn T, Hornik K, van de Wiel MA and Zeileis A (2006) A Lego System for Conditional Inference. The American Statistician 60(3): 257-263.