Reduction of SorLA/LR11, a Sorting Protein Limiting β-Amyloid Production, in Alzheimer Disease Cerebrospinal Fluid FREE

The neuronal sortilin-related receptor SorLA/LR11 (LR11), a member of the apolipoprotein E (ApoE) and low-density lipoprotein receptor family, functions as a sorting and trafficking protein, guiding β-amyloid precursor protein (APP) into the recycling endosome pathways that lead to production of β-amyloid peptide (Aβ).¹ In 2007, LR11 was shown to reduce Aβ production and was found to be a candidate genetic risk factor for late-onset Alzheimer disease (LOAD),² the most common cause of dementia in aged populations. Genetic polymorphisms associated with increased Alzheimer disease (AD) risk and with reduced LR11 expression were found in approximately 15% of AD cases in several but not all population-based investigations.³ Furthermore, neuronal LR11 expression is reduced in brain tissue from patients with LOAD^{4 - 5} but not from patients with early-onset familial AD, in which overproduction of Aβ42 is thought to be causative.⁶ In addition, subjects having mild cognitive impairment (MCI) with low brain levels of LR11 were significantly more cognitively impaired than subjects with high brain levels of LR11. Reduced LR11 levels occur early, increase with disease severity, and may predict progression to AD in a subgroup of individuals with MCI.⁷ Collectively, these data argue that, whatever their cause, LR11 deficits in LOAD occur early and are not simply secondary to pathologic changes. Furthermore, because they can be predicted to increase Aβ42 production, LR11 deficits are likely to contribute to pathogenesis and could be a useful target of a diagnostic biomarker or therapeutic intervention.

The selective reduction of Aβ42 in cerebrospinal fluid (CSF) of patients with AD has been widely used as a biomarker to increase diagnostic accuracy. However, deposition of Aβ42 in the brain or its reduction in CSF occurs in other non-AD dementias such as dementia with Lewy bodies⁸ and Parkinson disease dementia.⁹ Therefore, the differential diagnosis of dementia based on established clinical criteria and informed by reduced Aβ42 levels in CSF is still difficult to assess in standard practice. Other widely used CSF biomarkers for AD are elevated total tau (T-tau) or phosphorylated-tau (P-tau) levels. Cerebrospinal fluid levels of T-tau presumably reflect the intensity of neurodegeneration. There is no consensus whether any tau phosphorylation sites are specific for AD and not found in other tauopathies.¹⁰ Elevated CSF T-tau level is also found in vascular dementia¹¹ and in Creutzfeldt-Jakob disease.¹² Therefore, while useful, CSF Aβ and tau levels have limitations in the differential diagnosis of dementia.^{13 - 14} It remains a challenge to identify an ideal biomarker for AD that is directly linked to the primary mechanism of the disease and that can be used to monitor disease progression.

Although LR11 has a single transmembrane domain, it is endoproteolytically cleaved near the plasma membrane C-terminus, so that low levels of a large (approximately 250 kD) soluble LR11 N-terminal piece (LR11s) can be observed in conditioned media from 3 different human cell lines, including NT2 cells (neurogenic human embryonal carcinoma cells), BON cells (a pancreatic neuroendocrine tumor cell line),¹⁵ and SH-SY5Y human neuroblastoma cells.¹⁶ Therefore, we hypothesized that LR11 might be secreted into CSF and reduced in patients with AD. To test this hypothesis, we examined LR11 levels by Western blot in the lumbar CSF of patients with mild to moderate AD (mean Mini-Mental State Examination score, 24) and in ventricular CSF of patients with autopsy-confirmed AD and compared them with CSF of age-matched healthy persons and patients with non-AD dementias as neurologic control subjects.

The diagnosis of probable AD was established according to the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer Disease and Related Disorders Association criteria.¹⁷ Cognitive status was examined by multiple measures, including the Mini-Mental State Examination.¹⁸ Lumbar CSF samples were collected from 2 different research groups by lumbar puncture under standardized conditions, and aliquots were frozen at −80°C until used. The first group included 26 subjects (13 patients with AD and 13 age-comparable cognitively intact controls) enrolled at the University of California, San Diego AD research center site. The mean Mini-Mental State Examination scores were 24 (range, 17-28) for the patients with AD and 29 (range, 27-30) for the controls (Table 1). The second group included 17 patients with AD having serial CSF samples collected at the University of California, Los Angeles AD research center. In this group, the second CSF sample was collected 6 months after the first. The stability of our measures (LR11, secretase-cleaved soluble APP [sAPP], and ApoE) was investigated in these serial samples from patients with AD. To assess the usefulness of LR11 as a biomarker relative to Aβ42 and tau, we increased sample numbers in the first group (a total of 19 patients with AD and 18 age-comparable cognitively intact controls) and measured T-tau, P-tau, and Aβ42 levels. For all subjects, LR11, sAPP, and ApoE levels were measured using Western blot at the University of California, Los Angeles. Levels of CSF T-tau and Aβ42 were assessed using a commercially available system (Luminex xMAP; Invitrogen, Carlsbad, California).

Postmortem CSF samples were collected from 10 patients with autopsy-confirmed AD, 5 controls with non-AD dementias, and 5 age-matched healthy individuals based on autopsy reports from one of us (H.V.V.) at the University of California, Los Angeles AD research center. For the pathologic diagnoses of AD, plaque density was assessed using the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) criteria,¹⁹ and neurofibrillary tangles were evaluated using the criteria by Braak and Braak.²⁰ The following 3 types of plaques were identified and counted: (1) diffuse plaques (a neuropil deposition of finely granular material on Bielschowsky-stained sections), (2) dense core plaques (a neuropil deposition of compact argyrophilic material on Bielschowsky-stained sections), and (3) neuritic plaques (the presence of dystrophic neurites arranged radially and forming a discrete spherical lesion averaging about 30 mm in diameter).^{21 - 22} Neuritic plaques were counted on tau-immunostained sections. Plaque counts were normalized and expressed as the number of counts per millimeter squared. The number of each type of plaque was also characterized according to the CERAD rating scheme as none, sparse (1-5 per 3100 field), moderate (6-15 per 3100 field), or frequent (>15 per 3100 field).^{19 ,22} The score of the neocortical region with the highest count was used as the overall score for each subject (Table 2). CERAD criteria¹⁹ and Braak stage²⁰ were combined to give an estimate of the likelihood that AD pathologic changes underlie dementia using criteria of the National Institute on Aging and the Reagan Institute Working Group on Diagnostic Criteria for the Neuropathological Assessment of Alzheimer's Disease.²³ Autopsy-confirmed AD stage was typically intermediate (CERAD¹⁹ moderate plaques and Braak stage III-IV). Dementia with Lewy bodies was diagnosed according to the pathologic criteria by Kosaka et al.²⁴ Ventricular CSF was collected during autopsy, and aliquots were frozen at −80°C until used. Patients with AD were a mean (SD) of 83.7 (10.2) years old. The mean (SD) postmortem interval before autopsy was 18.9 (10.4) hours (Table 1).

Western blot was performed as described previously.²⁵ In brief, 50 μL of CSF was electrophoresed on 6% tris-glycine gels, and Western blots were stained with anti-LR11 (1:1000 dilution; BD Biosciences, San Jose, California) monoclonal antibody, followed with goat antimouse secondary antibody at 1:50 000 dilution for 1 hour and developed using chemiluminescence (Supersignal; Pierce, Rockford, Illinois) at 1:40 dilution. α-Secretase–cleaved APP (sAPPα) was detected by 6E10 antibody (1:2000 dilution; Signet Laboratories, Inc, Dedham, Massachusetts). After stripping immunoblots, β-secretase–cleaved APP (sAPPβ) was detected by sAPPβ polyclonal rabbit antibody (1:1000 dilution; Covance Development Services Company, Berkeley, California), followed with goat antirabbit secondary antibody at 1:100 000 dilution for 1 hour. Apolipoprotein E was detected by monoclonal mouse ApoE antibody (1:1000 dilution; Chemicon, Temecula, California). Band intensities were scanned and quantified using densitometric software (Molecular Analyst II; Bio-Rad, Hercules, California) and were expressed as arbitrary units. Quantification was performed blind to diagnosis.

Human SH-SY5Y neuroblastoma cells were maintained in Dulbecco modified Eagle medium supplemented with 2mM L-glutamine. Using 6-well plates, 5 × 10⁵ cells were plated and grown to 80% confluence at 37°C in an incubator with a humidified 5% carbon dioxide atmosphere. Cultured media were collected for Western blot. Cultured cells were placed on ice, washed, and scraped into a cold phosphate-buffered saline solution, and 3000-rpm microfuge pellets were dissolved in lysis buffer containing protease and phosphatase inhibitors, sonicated, incubated (4°C for 30 minutes), and centrifuged (14 000 rpm for 10 minutes). Supernatants were used for Western blot.

Stability of CSF analyte levels was assessed by Pearson product moment correlation coefficients of serial measures 6 months apart and by paired t tests on 6-month change in CSF levels. Between-group differences in postmortem CSF levels were assessed by analysis of variance, followed by Tukey-Kramer post hoc tests. Bivariate receiver operating characteristic (ROC) curves were constructed using the method by Pepe et al.²⁶ All statistical analyses were performed using commercially available software (StatView 5.0; StatView, Cary, North Carolina) except for the ROC curves and related statistics, which were calculated using a statistical package (ROCR; GNU General Public License, Boston, Massachusetts; http://rocr.bioinf.mpi-sb.mpg.de²⁷ )and a statistical programming language (R; R Development Core Team; http://www.R-project.org).

Hampe et al¹⁵ reported that a soluble form of LR11 was secreted into conditioned medium of human NT2 and BON cells. The molecular mass of soluble LR11 seemed slightly less than that of the membrane-bound approximately 250-kD full-length LR11 form, in which the LR11 fragment was about 10 kDa less than the membrane fraction form. Consistent with this study, we found that soluble LR11 is also present in cultured medium from human SH-SY5Y neuroblastoma cells (Figure 1A) and in human CSF (Figure 1B). LR11s in cultured media and in human CSF were slightly smaller than the form found in membrane fractions on 6% gels, suggesting that LR11 could be secreted into culture media or human CSF and detected by Western blot with LR11 antibody.

To determine whether CSF LR11 is stable over time in patients with AD, we examined LR11 from serial CSF samples obtained 6 months apart in a cohort of 17 patients with AD. Western blot showed no difference in the LR11 level between first and second samples in CSF from patients with AD (P =.81 [Figure 2A]). A high correlation between first and second samples of LR11 was observed (r = 0.92, P < .001 [Figure 2B]), suggesting that the level of LR11 is stable within a 6-month time frame in patients with established AD. The relative levels of sAPPα (P =.80), sAPPβ (P =.78), and ApoE (P =.65) also did not show significant differences between first and second samples after a 6-month interval. Similarly, good correlations between the first and second CSF samples were found for sAPPα (P < .001, r = 0.97 [Figure 2C]), sAPPβ (P =.001, r = 0.72 [Figure 2D]), and ApoE (P =.001, r = 0.78 [Figure 2E]).

To determine whether there was any alteration in LR11 protein levels of CSF from patients with AD, we first compared LR11 protein levels in subjects with probable AD with those in cognitively healthy controls by Western blot using a monoclonal antibody against LR11. The specificity of our anti-LR11 was well characterized in vitro by small interfering RNA to LR11² and in vivo using LR11 knockout mice.¹⁶ Our results indicated that LR11 levels were significantly decreased in patients having AD compared with healthy controls (P =.03 [Figure 3A and B]). This result is consistent with previous findings that LR11 levels were significantly decreased in brain tissue from patients with LOAD.^{1 ,6 ,28} To further confirm these results, we increased sample numbers (to 19 patients with AD and 18 controls), reran and reanalyzed Western blots for LR11, and measured CSF Aβ42, P-tau, and T-tau levels. The mean LR11 (P =.04) levels remained statistically significantly different between patients with AD and controls in the expanded sample (Table 3). The Aβ42 (P =.003), T-tau (P =.046), and P-tau (P =.04) levels also showed significant differences between patients with AD and controls, suggesting that these patients with AD were typical for the established biomarkers of mild AD.

We used ROC curves to characterize the potential discriminant usefulness of LR11, Aß42, T-tau, and P-tau lev els using the sample of 19 patients with AD and 18 controls. The Aβ42 level was the most effective univariate discriminator of AD vs control in our data, with an area under the ROC curve (AUC) of 0.789 compared with 0.661 for LR11 level. Adding T-tau or P-tau level to a bivariate discriminant model with Aβ42 level did not substantively improve the model as measured by AUC, only increasing the AUC to 0.793. In contrast, adding LR11 level increased the AUC to 0.865 (P =.01). The Aβ42/LR11 ROC curve cutoff that correctly identifies 90% of the cases (sensitivity, 0.90) also correctly identifies 81% of the controls (specificity, 0.81) (Figure 4).

Because LR11 functions as a neuronal sorting protein that binds APP and regulates APP processing to decrease Aβ production in cultured cells,^{2 ,28} we investigated whether there was any change in the sAPP level in CSF of human subjects and its correlation with LR11 level. The results showed that the sAPPβ level was significantly decreased in patients having AD compared with controls (P =.02 [Figure 3C]) and that the sAPPα level had a trend toward a decrease (P =.06, data not shown), consistent with earlier studies.^{29 - 30} The reduced sAPPα (P < .001, r = 0.64; data not shown) and sAPPβ (P < .001, r = 0.72) levels significantly correlated with levels of the APP binding partner, LR11 (Figure 3E).

Apolipoprotein E is involved in lipid transport and cholesterol homeostasis as a major component of brain lipoproteins. The presence of the ApoE4 allele is a major risk factor for LOAD,³¹ and LR11 is a member of the low-density lipoprotein receptor family reported to interact with ApoE.^{32 - 33} Therefore, we investigated CSF ApoE levels and their possible correlation with LR11 levels in these subjects. Consistent with previous findings,³⁴ ApoE levels in CSF did not show any significant difference between patients with AD and controls (P =.56 [Figure 3D]). Furthermore, no correlation between the ApoE level and the LR11 level was observed (P =.46, r = 0.15 [Figure 3F]).

To further confirm the observation that the CSF LR11 level is decreased in patients with AD, we studied postmortem CSF samples of patients with autopsy-confirmed moderate AD. The results showed that LR11 levels in postmortem CSF samples of patients with AD were significantly diminished compared with those of healthy controls or patients with non-AD dementias (P < .05 [Figure 5A and B]). Again, ApoE levels did not show any difference in these 3 groups (Figure 5A and C). These results are comparable to those obtained with lumbar CSF samples from patients with probable AD (Figure 3).

In this study, we report that LR11 protein can be detected and measured in CSF and that levels are significantly reduced in CSF of patients with mild to moderate AD compared with age-matched controls. This reduction was confirmed in postmortem CSF samples of patients with autopsy-proven AD but not in patients with non-AD dementias such as dementia with Lewy bodies, multiple system atrophy, and vascular dementia. Perhaps not surprisingly for an APP binding protein, LR11 level significantly correlated with the sAPPα and sAPPβ levels. Despite being a low-density lipoprotein receptor family member that can bind ApoE, the LR11 level did not correlate with the ApoE level. These findings are consistent with a primary role for LR11 as a neuronal sorting protein rather than as a significant CNS ApoE receptor.

Although the genetic association of LR11 with LOAD remains controversial, LR11 deficits with reductions in LR11 messenger RNA and protein have been a consistent finding in patients with LOAD but not in patients with early-onset familial AD, in which overproduction of Aβ42 is known to be caused by presenilin or APP mutations. This suggests that LR11 loss does not occur simply secondary to pathologic changes and might occur at early disease stages and even before diagnosis. Herein, we found that decreased LR11 levels in CSF occurred at an early stage (in patients with a mean Mini-Mental State Examination score of 24) and was further confirmed in patients having autopsy-proven AD with Braak stage III or IV. Because reduced LR11 levels have been observed in a subgroup of individuals with MCI,⁷ LR11 deficits in CSF may be detectable in patients with MCI or even earlier stages of AD in a subset of patients at risk for LOAD. Combining the essential function of LR11 in trafficking APP and regulating Aβ production, these data strongly suggest that CSF LR11 levels might be a useful biomarker for LOAD.

LR11 contains a vacuolar protein sorting 10-protein domain involved in protein transport between the plasma membrane, endosomes, and late Golgi compartments.^{33 ,35 - 36} Increased LR11 levels significantly alter the sorting and trafficking of APP to the recycling Golgi and early endosomal compartments, which results in a decrease in Aβ production via the amyloidogenic pathway.^{1 ,37} In addition, it has been reported that LR11 sorts APP to intracellular protein complexes (retromers) that traffic APP away from α- and β-secretase.^{38 - 39} Retromer trafficking dysfunction can lead to increased APP in the late endosome, an organelle in which BACE activity and Aβ production are maximized.^{40 - 41} Reductions of sAPP (including α- and β-sAPP) in CSF of patients with AD have been reported and were initially proposed as a diagnostic marker for AD.^{29 - 30 ,42} However, sAPP deficits have not been widely studied or used as diagnostic markers clinically. In this study, sAPP levels (including sAPPα and sAPPβ) were reduced in CSF of patients with AD. Notably, LR11 levels significantly correlated with those of sAPP. It is possible that this correlation occurs as a result of the binding of LR11 with APP and regulation of its processing. These results may provide a potential explanation for the changes of sAPP level in CSF from patients with AD.

Considering that LR11 is a member of the ApoE receptor family, LR11 could be influenced by ApoE in CSF and vice versa. However, ApoE levels in CSF did not differ between patients with AD and controls and did not correlate with LR11 levels. Any role for LR11 in CNS ApoE metabolism remains hypothetical.

In summary, this study demonstrates that soluble LR11 is significantly reduced in CSF of patients with early AD and correlates with reductions in sAPP but not the ApoE level. Reduced LR11 levels in CSF was confirmed in patients having autopsy-proven AD with Braak stage III or IV but not in patients with non-AD dementias such as dementia with Lewy bodies, multiple system atrophy, and vascular dementia. LR11 messenger RNA is downregulated in lymphoblasts, and protein levels are reduced in cortical and hippocampal neurons from patients with LOAD but not patients with early-onset familial AD.^{4 ,6} This suggests that LR11 protein deficits might be detectable in plasma, but our attempts to measure plasma LR11 levels failed, and any relationship with more accessible blood messenger RNA levels needs to be explored. Together, this study provides preliminary evidence that the LR11 level may be a potential CSF biomarker for sporadic but not familial AD. In our sample, Aβ42 level was the best single CSF predictor of AD. Adding LR11 level to a discriminant function with Aβ42 level improved sensitivity and specificity (AUC changes from 0.789 to 0.865, P =.01), while adding T-tau or P-tau level did not. These are preliminary results and are without a validation sample. Because LR11 expression is sensitive to dietary omega-3 fatty acids,¹⁶ it is not surprising that LR11 level is not as closely connected to diagnosis as CSF Aβ42 level.

The stability of LR11 in CSF suggests that it may be used to monitor therapeutic effects of AD treatment. This would be particularly true in the case of drugs expected to work by modulating CNS LR11 expression such as the ω-3 fatty acid docosahexaenoic acid.¹⁶ Because of the limited non-AD dementia sample size herein, the potential of CSF LR11 level as a diagnostic marker for AD should be further investigated and combined with other biomarkers with larger sample sets, including MCI or preclinical cases. For example, it is possible that measuring the LR11 level alone in MCI or in combination with Aβ level would help to predict which patients with MCI are at highest risk for progression to AD.

Correspondence: Greg M. Cole, PhD, Geriatric Research and Clinical Center, Greater Los Angeles Veterans Affairs (VA) Healthcare System, VA Medical Center, Research 151, Bldg 7, Room A101, 16111 Plummer St, North Hills, CA 91343 (gmcole@ucla.edu).

Accepted for Publication: June 8, 2008.

Author Contributions:Study concept and design: Ma, Frautschy, and Cole. Acquisition of data: Ma, Galasko, Ringman, Vinters, Pomakian, Ubeda, Rosario, and Cole. Analysis and interpretation of data: Ma, Galasko, Vinters, Edland, Pomakian, Rosario, Teter, and Cole. Drafting of the manuscript: Ma, Galasko, and Cole. Critical revision of the manuscript for important intellectual content: Ma, Galasko, Ringman, Vinters, Edland, Pomakian, Ubeda, Rosario, Teter, Frautschy, and Cole. Statistical analysis: Ma and Edland. Obtained funding: Cole. Administrative, technical, and material support: Ma, Galasko, Ringman, Vinters, Pomakian, Rosario, and Cole. Study supervision: Frautschy and Cole.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grants AT3008 (Dr Cole), NIA AG13471 (Dr Cole), and P50 AG005131 (Dr Galasko) from the National Center for Complementary and Alternative Medicine, P01 AG16570 (Dr Cole; project 1, UCLA, ADRC), and NIR6-07-59659 from the Alzheimer's Association (Dr Ma). Dr Vinters was supported in part by the Daljit S. and Elaine Sarkaria Chair in Diagnostic Medicine, University of California, Los Angeles.