OUP user menu

CSF phosphorylated tau protein correlates with neocortical neurofibrillary pathology in Alzheimer's disease

Katharina Buerger, Michael Ewers, Tuula Pirttilä, Raymond Zinkowski, Irina Alafuzoff, Stefan J. Teipel, John DeBernardis, Daniel Kerkman, Cheryl McCulloch, Hilkka Soininen, Harald Hampel
DOI: http://dx.doi.org/10.1093/brain/awl269 3035-3041 First published online: 29 September 2006


Hyperphosphorylated tau protein (P-tau) in CSF is a core biomarker candidate of Alzheimer's disease. Hyperphosphorylation of tau is thought to lead to neurofibrillary changes, a neuropathological hallmark of this type of dementia. Currently, the question is unresolved whether CSF levels of P-tau reflect neurofibrillary changes within the brain of a patient with the illness. Twenty-six patients were included with intra-vitam CSF as well as post-mortem neuropathological data. In the CSF, P-tau phosphorylated at threonine 231 (P-tau231P) was analysed. Post-mortem, scores of neurofibrillary tangles (NFT) and neuritic plaques (NP) were assessed in frontal, temporal, parietal and hippocampal cortical areas. In the same cortical regions, load of hyperphosphorylated tau protein (HP-tau load) was determined. Concentrations of P-tau231P were measured in frontal cortex homogenates. We found significant correlations between CSF P-tau231P concentrations and scores of NFTs and HP-tau load in all neocortical regions studied. The score of NPs was correlated with CSF P-tau231P only within the frontal cortex. There was a correlation between P-tau231P in CSF and brain homogenates. These findings indicate that CSF P-tau231P may serve as an in vivo surrogate biomarker of neurofibrillary pathology in Alzheimer's disease.

  • hyperphosphorylated tau protein
  • CSF
  • neuropathology
  • Alzheimer's disease


Major efforts are under way to define biomarkers of Alzheimer's disease. A potential approach is to study proteins in CSF which are linked to fundamental features of Alzheimer's disease neuropathology as outlined by a Consensus report (Consensus report, 1998). Abnormal hyperphosphorylation of the microtubule-associated protein tau and its incorporation into neurofibrillary changes [neurofibrillary tangles (NFT), dystrophic neurites surrounded by neuritic plaques (NP) and neuropil threads] are major components of Alzheimer's disease pathophysiology (for review see Mandelkow and Mandelkow, 1998).

Highly significant increases of hyperphosphorylated tau protein (P-tau) in CSF have recently been demonstrated in patients with Alzheimer's dementia compared with control groups by independent studies (for review see Blennow and Hampel, 2003; Buerger and Hampel, 2004). Measurement of P-tau in CSF has been determined a feasible core marker of the disease (Frank et al., 2003). We studied tau phosphorylated at threonine 231 (P-tau231P) in the CSF because this particular P-tau subtype has been shown to be strongly expressed in diseased brain compared with that of controls (Vincent et al., 1998) and has largely been investigated in clinical biomarker studies (Buerger et al., 2003; Hampel et al., 2004). We have shown a high sensitivity of P-tau231P in Alzheimer's disease performing superior when compared with tau phosphorylated at serine 199 (Hampel et al., 2004) as well as potential clinical applications in the classification of Alzheimer's disease versus frontotemporal dementia and major depression (Buerger et al., 2003; Hampel et al., 2004). When analysed serially, rate of P-tau231P change correlated inversely with cognition at baseline (Hampel et al., 2001). Moreover, high levels of P-tau231P correlated with cognitive decline in subjects with mild cognitive impairment (Buerger et al., 2002), an at-risk group for Alzheimer's dementia (Gauthier et al., 2006).

The relationship between CSF P-tau and cerebral neurofibrillary changes in this disease has not yet been investigated. In the current study, we asked whether levels of in vivo P-tau231P derived from CSF correlate with post-mortem assessed amount of NFT, NP, neuropil threads and P-tau231P concentration in tissue homogenates measured within the brain of Alzheimer's disease patients.

Material and methods


CSF was obtained from 26 clinically diagnosed Alzheimer's disease patients (24 females, 2 males; age 85.4 ± 7.3 years, MMSE at lumbar puncture 2.2 ± 3.8 points, range 0–14). CSF was stored immediately at −80°C until further examination. Study subjects underwent autopsy on average 1.6 (±1.5) years after spinal tap. Post-mortem delay was 2–12 h (mean 5.5 ± 2.6 h). The study was approved by the local Ethical Committee. Informed consent for the participation in the study was obtained from the patients' legal representatives.

Neuropathological work-up, NFT and NP scores

Following current NIA-R consensus recommendation for the post-mortem diagnosis of Alzheimer's disease (Consensus, 1997), 18 subjects had a high, 5 an intermediate and 3 a low likelihood that their dementia was due to Alzheimer's related pathology. The number of NFTs and NPs was assessed using light microscopy, in Bielschowsky silver impregnated sections in magnification ×100 as reported earlier (Moelsae et al., 1987) in frontal (Brodmann area 9), temporal (22), parietal (39) cortices and the CA1 region of the hippocampus. Five random fields were selected, each field measuring 0.92 mm2. A score of assessed lesions for each cortical region was calculated ranging from 0 to 10. The score value is a mean value obtained from 5 evaluated fields, e.g. a score of 3 indicates 5–9 lesions/field, a score of 6 indicates 20–24 lesions/field and a score of 10 indicates over 40 lesions/field.

Determination of P-tau231P in brain homogenates and CSF

Fresh frozen brain from the frontal cortex area 9 was allowed to thaw to 4°C. The cortical grey matter was dissected out and was homogenized with a Potter–Elvehjem tissue grinder using the ratio of 1 g tissue to 10 ml TBS containing 1 mM PMSF. The homogenate was centrifuged at 27 000 g for 20 min at 4°C and the supernatant was collected. The resulting pellet was rehomogenized using half of the original volume of TBS/1 mM PMSF, centrifuged for 20 min at 27 000 g, 4°C and the supernatant was collected and pooled with the previously obtained supernatant. Total protein content of the pooled supernatants was determined by a Coomassie Plus Protein Assay (Pierce). The supernatants were stored in polypropylene tubes at −80°C.

P-tau231P was measured in CSF and brain homogenates by an enzyme-linked immunosorbent assay (ELISA; Applied NeuroSolutions Inc., Vernon Hills, IL, USA) as previously described (Kohnken et al., 2000). Although the assays used to analyse the CSF and brain homogenates were identical, the standards used to construct the curve were of two different sources. The standard curve used to obtain CSF values was made from phospho-recombinant tau while the standard curve used for expression of the brain homogenates was constructed from a synthetic peptide containing a single phosphate at the amino acid corresponding to threonine 231 in full-length tau (441 amino acids). The two different standards have equivalent performance in the assay (data not shown); however, the units of expression differ as to the absolute amount of phospho-tau present.

Assay operators were blinded to neuropathological classification of the samples.


The visualization of hyperphosphorylated tau (HP-tau) in brain tissue was carried out employing immunohistochemical methodology. Shortly, the deparaffinized and rehydrated 7 μm thick sections were incubated overnight with HP-tau antibody (Innogenetics/Br-03) at a dilution 1:500. For detection, the Histostain SAP kit (Zymed) was used together with Vector-Red chromogen (Vector Laboratories). The antibody recognizes paired filament tau and does not cross-react with normal tau (Mercken et al., 1992). The epitope has been shown to contain the serine 202 (Goedert et al., 1993). For immunohistochemical labelling of abnormally phosphorylated tau seen in Alzheimer's disease this antibody (AT8) is commonly used. The antibody detects tangle and even pretangle material. A staging system of Alzheimer's-associated neurofibrillary pathology using immunohistochemistry and AT8 has been suggested (Braak et al., 2006).

The HP-tau-τ (HP-tau load) expression was quantified under light microscopy at ×40 magnification and was scored on a four-step scale from 0 to 3 (none, 0; some, 1; moderate, 2; or extensive, 3). In a case scored 1, occasional positively stained fibrils were seen, for score 2, several stained fibrils were noted with additional threads and for score 3 numerous fibrils and threads were noted (Alafuzoff et al., 1999).

ApoE genotyping

ApoE genotyping was performed as described previously (Heinonen et al., 1995). The ApoE genotype frequency was as follows: n = 2 for ApoE ɛ2/3, n = 11 for ApoE ɛ3/3, n = 8 for ApoE ɛ3/4, n = 2 for ApoE ɛ2/4 and n = 3 for ApoE ɛ4/4.


Multiple regression models were used in order to test the prediction of NFT and NP count as well as HP-tau load by P-tau231P for each of the regions of interest, controlling for the influence of MMSE, age, ApoE genotype, interval between lumbar puncture and death, and disease duration. The subjects included 13 ApoE ɛ4 carriers versus 13 ApoE ɛ4 non-carriers.

In order to control for accumulation of Type I error due to multiple testing and non-normality of data distribution, we applied the distribution free method of bootstrapping to each regression analysis, using 999 iterations (Efron and Tibshirani, 1986). The resulting bootstrapped regression coefficient (beta), standard errors (SE) and the 95% confidence intervals (95% CI) of the regression coefficients are reported here. For bootstrapping, no parametric assumption needs to be made. Rather, bootstrapping ‘reconstructs’ the data distribution by repeatedly drawing samples, with replacement, from a given dataset. In this case, 999 samples with n = 26 were drawn with replacement from the original dataset, and the bootstrapped statistics were computed on the basis of the resulting data distribution. The bootstrapped confidence intervals provide an estimate on the range of the size of the true population parameter, here the regression coefficient, and thus provide information on the accuracy of the findings.

Spearman rank correlations coefficients (rho), which are appropriate for bivariate datasets of a least ordinal scale, were computed between the concentration of P-tau231P measured in brain homogenate derived from the frontal cortex, frontal NP count, and each of the other markers measured within the frontal cortex including HP-tau load and NFT count. All analyses were conducted with the computer-based software package SPSS11.5 (SPSS Inc., Chicago, IL, USA).


Age, gender, MMSE score, disease duration, Braak NFT stage and CSF levels of P-tau231P in each patient and the respective group means are displayed in Table 1.

View this table:
Table 1

Demographic, clinical, neuropathological and CSF data for individual patients

Case/statisticAge (years)GenderMMSEDisease duration (in years)Braak stage NFTP-tau231P (pg/ml)
  • f, female; m, male.

Results of the bootstrapped regression models for each neuropathological marker and brain region are displayed in Table 2.

View this table:
Table 2

Regression coefficient, standard error and 95% CI for CSF levels of P-tau231P as a predictor of neuropathological markers specified for each brain region

Lesion assessedBrain regionBetaSE95% CI
NFT countFrontal0.0050.0020.001 to 0.009*
Temporal0.0050.0020.001 to 0.009*
Parietal0.0050.0020.001 to 0.010*
Hippocampus0.0030.004−0.005 to 0.011
NP countFrontal0.0040.0020.0005 to 0.007*
Temporal0.0030.002−0.00006 to 0.006
Parietal0.0020.002−0.003 to 0.006
Hippocampus0.0040.004−0.003 to 0.011
HP-tau loadFrontal0.0020.00060.0005 to 0.003*
Temporal0.0020.00050.001 to 0.003*
Parietal0.0020.00060.0009 to 0.004*
Hippocampus0.00090.0005−0.0001 to 0.002
  • Beta, regression coefficient; SE, standard error; 95% CI, 95% confidence interval of the regression coefficient.

  • *Statistically significant at α = 0.05.

CSF levels of P-tau231P were associated with NFT count and HP-tau load within all neocortical regions, such as frontal, temporal and parietal lobe, but not in the allocortical CA1 region of the hippocampus (Table 2, Figs 1 and 2). For NP count, a significant association with P-tau231P was observed within the frontal lobe (Table 2, Fig. 3). Among those covariates other than CSF levels of P-tau231P, only disease duration and ApoE genotype were significantly related to selected neuropathological measures in neocortical brain areas. In particular, increased disease duration was associated for the frontal cortex with increased NFT count (beta = 0.55, 95% CI = 0.23–0.86), NP count (beta = 0.41, 95% CI = 0.10–0.73) and HP-tau load (beta = 0.12, 95% CI = 0.07–0.28), and, in addition, parietal NFT count (beta = 0.46, 95% CI = 0.14–0.78). ApoE ɛ4 carriers, compared with ApoE ɛ4 non-carriers, showed an increased NP count within the temporal cortex (beta = 2.67, 95% CI = 0.51–4.84) and parietal cortex (beta = 2.34, 95% CI = 0.17–4.52). None of the other regressions became significant for any of the covariates (data not shown).

Fig. 1

Scatter plot for NFT count as a function of P-tau231P measured in CSF for each brain region studied. The regression line shows the predicted linear relationship between NFT count and P-tau231P, The Spearman correlation coefficient rho and the respective P-value are indicated for each bivariate correlation. The result pattern did not change when age, gender, time interval between lumbar puncture and death, ApoE genotype and disease duration were controlled for (see text for details on multivariable regression analysis).

Fig. 2

Scatter plot for HP-tau load as a function of P-tau231P measured in CSF for each brain region studied. The Spearman correlation coefficient rho and the respective P-value are indicated for each bivariate correlation. Significant correlations remained when age, gender, time interval between lumbar puncture and death, ApoE genotype and disease duration were controlled for. The correlation between frontal HP-tau load and CSF P-tau231P became significant when controlling for potential confounding factors (see Table 2).

Fig. 3

Scatter plot for NP count as a function of P-tau231P measured in CSF for each brain region studied. The Spearman correlation coefficient rho and the respective P-value are indicated for each bivariate correlation. NP count was correlated with CSF P-tau231P only within the frontal cortex, consistent with the findings of the multivariate analysis controlling for potential confounding variables as presented in the text.

CSF levels of P-tau231P were associated with P-tau231P when measured in brain homogenate of the frontal cortex (beta = 0.02, SE = 0.008, 95% CI = 0.008–0.04).

The concentration of P-tau231P measured in frontal brain homogenate was significantly correlated with count of NFT (correlation coefficient rho = 0.72, P < 0.001), HP-tau load (rho = 0.80, P < 0.001) and NP (rho = 0.60, P = 0.001) within the frontal cortex. The bivariate correlations between duration of disease and P-tau231P measured in frontal brain homogenate or CSF were not statistically significant (brain homogenate: rho = 0.25, P = 0.21; CSF: rho = −0.03, P = 0.90).


We report for the first time a correlation between P-tau231P measured in CSF and NFT scores as well as HP-tau load within neocortical areas, independent of age, MMSE, time interval between lumbar puncture and death, ApoE genotype and disease duration. The density of NP was correlated with P-tau231P levels in CSF only when measured within the frontal cortex. Levels of CSF P-tau231P were correlated with P-tau231P concentrations detected in brain homogenates. The mean CSF P-tau231P levels reported here are in accordance with previously published highly elevated concentrations in patients when compared with controls (Hampel et al., 2004; Buerger et al., 2005, 2006). The present data for individuals with Alzheimer's disease is significantly greater than the cut-offs reported in those studies and is in general at least seven times higher than levels found in controls that have been previously investigated.

Few previous studies have examined the relation between CSF-based biomarker candidates of Alzheimer's dementia and neuropathological scores (Tapiola et al., 1997; Strozyk et al., 2003). To our knowledge, none has addressed CSF P-tau and neuropathology so far. When total tau protein was measured within the CSF (Tapiola et al., 1997), a positive correlation was found between total tau and NFT within cortical brain areas, but not within the allocortical CA1 region of the hippocampus. NPs were not measured. These results are consistent with the current findings, suggesting that measures of total tau and P-tau derived from CSF reflect neuropathological changes within the brain. The measure of total tau used in the previous study, however, includes both phosphorylated and non-phosphorylated tau isoforms and may represent a marker of neurodegeneration rather than Alzheimer's disease-specific neurofibrillary changes (Blennow and Hampel, 2003). Results of the current study extend these findings indicating that P-tau231P levels measured in the CSF show a correlation with levels of P-tau231P measured within the brain as well as markers of neurofibrillary changes, especially HP-tau load and NFTs. In addition, P-tau231P measured within the brain (sampled from the frontal cortex) was correlated with both P-tau231P in CSF and each of neuropathological measures including NFT, NP and HP-tau load. Taken together, these findings suggest that P-tau231P measured in CSF may serve as a surrogate biomarker of neurofibrillary pathology in the brains of the mostly severely demented patients studied here. It is noteworthy that an increase in disease duration was associated with an increase in all three neuropathological markers within the frontal lobe. Other studies have inconsistently found a correlation between disease duration and cortical neuropathologies including NFT and plaques in Alzheimer's disease (Berg et al., 1993, 1998). From the perspective of this study, the relationship between CSF-levels of P-tau231P and core neuropathologies in the brain does not seem to be mediated by disease duration.

In contrast to the amount of NFT and HP-tau load, the density of NP across different brain regions was less consistently correlated with CSF P-tau231P. A potential explanation for the results pattern could be the higher proportion of P-tau included in NFTs when compared with NPs. Whereas NFTs and neuritic processes are composed of fibrils containing phosphorylated tau, plaques consist mostly of fibrillous beta-amyloid (Grundke-Iqbal et al., 1986; Lee et al., 1991; Dickson, 1997). The neuritic subtype of plaques is invaded by dystrophic neurites of the neuron, and only such neurites include P-tau (Dickson, 1997). Thus, NP may contain relatively less P-tau in comparison to NFTs and neuropil threads, and therefore influence the level of CSF P-tau to a lower extent when compared with the other types of neurofibrillary lesions. Moreover, the classification of a plaque as NP (in contrast to diffuse amyloid plaques) does not control for the extent to which a particular NP contains neuritic processes from the neuron (Murayama and Saito, 2004) thus allowing for variance in the degree of neuritic pathology between NPs. Consequently, the density of NP may be only a crude measure of the extent of neuritic pathology. Therefore, levels of NP may be correlated with P-tau231P in CSF to a lesser degree when compared with NFT and HP-tau load. However, these interpretations await testing in future studies.

It should be noted that across different indices of neurofibrillary changes, a correlation with CSF levels of P-tau was found within neocortical brain areas but not in allocortex such as the hippocampal formation. This is consistent with findings for total tau by Tapiola et al. (1997). Ceiling or floor effects of neurofibrillary changes within the hippocampus are unlikely to account for the absence of a correlation since the level of neuropathology in the hippocampus was not different from other brain areas (data on file). The volume of the hippocampus, however, is rather small compared with frontal, temporal and parietal cortical areas. Thus, neurofibrillary changes when measured within the hippocampus may not significantly influence the overall P-tau levels within the CSF due to the relatively small area of the hippocampus relative to the remaining cortical areas affected by neurofibrillary tangles. We have shown, however, a positive correlation between high P-tau231 levels in CSF at baseline and rate of hippocampal atrophy as measured with serial magnetic resonance imaging (Hampel et al., 2005) suggesting that CSF P-tau231 may correlate with allocortical changes when measured in a longitudinal design.

Our study has several advantages including well characterized and post-mortem verified cases of Alzheimer's disease, controlling for disease severity (MMSE), disease duration, age, ApoE genotype and interval between in vivo CSF collection and death. Several caveats should be taken into account for the interpretation of the study reported here. First, sample size may limit statistical validity. In order to reduce the influence of outliers that may show an effect particularly in small data samples, however, we used bootstrapping, thereby reducing the influence of any potential outlier. Another caveat is that the current study included mostly cases with clinically severe Alzheimer's dementia. Thus, a generalizability of the findings towards early stages of the disease needs to be tested in further studies. Finally, this is a correlative study. A third unknown factor may have mediated the observed correlation between P-tau231P measured in CSF and neurofibrillary changes within the brain. The mechanism of the development of neurofibrillary changes and expression of P-tau in the CSF has to be investigated in future experimental studies.


  • *These authors contributed equally to this work.


View Abstract