Neurodegenerative disease biomarkers Aβ1–40, Aβ1–42, tau, and p‐tau181 in the vervet monkey cerebrospinal fluid: Relation to normal aging, genetic influences, and cerebral amyloid angiopathy

Abstract Background The Caribbean vervet monkey (Chlorocebus aethiops sabaeus) is a potentially valuable animal model of neurodegenerative disease. However, the trajectory of aging in vervets and its relationship to human disease is incompletely understood. Methods To characterize biomarkers associated with neurodegeneration, we measured cerebrospinal fluid (CSF) concentrations of Aβ1–40, Aβ1–42, total tau, and p‐tau181 in 329 members of a multigenerational pedigree. Linkage and genome‐wide association were used to elucidate a genetic contribution to these traits. Results Aβ1–40 concentrations were significantly correlated with age, brain total surface area, and gray matter thickness. Levels of p‐tau181 were associated with cerebral volume and brain total surface area. Among the measured analytes, only CSF Aβ1–40 was heritable. No significant linkage (LOD > 3.3) was found, though suggestive linkage was highlighted on chromosomes 4 and 12. Genome‐wide association identified a suggestive locus near the chromosome 4 linkage peak. Conclusions Overall, these results support the vervet as a non‐human primate model of amyloid‐related neurodegeneration, such as Alzheimer's disease and cerebral amyloid angiopathy, and highlight Aβ1–40 and p‐tau181 as potentially valuable biomarkers of these processes.

The Caribbean vervet monkey (Chlorocebus aethiops sabaeus, also known as the African green monkey), an Old World monkey, is well-suited as a model of aging and neurodegenerative disease; compared to rhesus macaques, vervets are more available and carry fewer pathogens. Vervets display AD-like pathology, including accumulation of Aβ-containing cerebral plaques associated with reactive astrocytosis, dystrophic neurites, and vascular Aβ immunoreactivity resembling cerebral amyloid angiopathy (CAA), but unlike humans have relatively preserved cognition that does not appear correlated with plaque burden and demonstrate scant phosphorylated tau aggregates (Heuer et al., 2012;Kalinin et al., 2013;Lemere et al., 2004;Toledano et al., 2014). The high incidence of Aβ immunoreactive plaques in vervets older than 20 years of age (Kalinin et al., 2013) suggests that they occur as part of the normal aging of the vervet.
In humans, cerebrospinal fluid (CSF) harbors useful biomarkers of neurodegenerative disease. CSF levels of Aβ and tau change with AD.
The Vervet Research Colony (VRC) was established from vervets captured in St. Kitts and Nevis, and maintained as a single extended pedigree (Jasinska et al., 2012), thereby providing a large sample ideal for mapping quantitative traits and study of complex neurobiologic traits. Additionally, the wide distribution of subjects with ages across the normal vervet lifespan enables the study of age-related disease; vervets typically live to 11-13 years in the wild, but up to 25 years in captivity (Magden, Mansfield, Simmons, & Abee, 2015); the oldest animal in the VRC was a female that recently died at 29.1 years of age.
Previous work in the VRC reported on the acquisition and analysis of neuroanatomic phenotypes using high-resolution structural MRI in a large number of colony members (Fears et al., 2009(Fears et al., , 2011. In the current study, we extend the neurobiologic characterization of the VRC through the acquisition and analysis of CSF biomarkers in the context of the rich genetic and phenotypic information available through the VRC. from 58 founder animals that were captured from a wild population in the islands of St. Kitts and Nevis. Vervet monkeys originally arrived in the Caribbean Islands in the 1600s on trading ships from Africa (McGuire, 1974). Since the founding of the colony, breeding has been managed to provide a species-typical social environment for developing offspring and to promote genetic diversity, while preserving each of the original matrilines to simulate the social structure of wild-living troops. Female offspring remain in the breeding groups, and males are removed to separate housing at 4 years of age as previously described (Fairbanks et al., 2004). Adult males born in different social groups are introduced into the breeding groups at 3-4 year intervals; because no new animals have been imported for several decades, extensive inbreeding occurs. The average inbreeding coefficient (probability that a subject receives two alleles identical-by-descent) in the colony was recently characterized at 0.006, with animals with non-zero inbreeding coefficient ranging from 0.004 to 0.25 (mean 0.05 ;Freimer et al., 2007). The high level of inbreeding facilitates genetic mapping in our study population, improving the effective sample size for association and providing a high informative pedigree for linkage. The genetic architecture and genetic control of traits may not reflect outbred populations.

| Vervet pedigree
Vervets were fed commercial laboratory chow ad lib, supplemented by fresh fruits and produce. Specifically, the vervet chow has primarily been LabDiet 5038 ("Monkey Diet"), with several exceptions.

| CSF biomarker measurements
CSF was collected from the cisterna magna of each animal under a protocol approved by the Animal Research Committee at the University of California, Los Angeles, as previously described (Freimer et al., 2007). CSF protein concentrations were measured at the Knight for Aβ 1-42 , AT120 for tau, and AT270 for p-tau 181 . Additional biotinylated antibodies are then used to quantify the proteins -3D6 for Aβ 1-42 , and HT7 for tau and p-tau 181 . The proteins are then detected using a phycoerythrin-labeled streptavidin readout. Calibration of the standard curve was performed as described in the manufacturer's directions using six accompanying standards provided in the kit. All assays were run in duplicate. Measurements with coefficient of variation (CV) >25% or bead count <30 were excluded in quality control.

| Neuroimaging analysis
Neuroimaging data were acquired in 2007; details of the image acquisition and pre-processing protocol have been described previously (Fears et al., 2009). To generate high signal-to-noise images, nine separate structural scans were acquired from 357 animals (256 females and 101 males) using an 8-channel high-resolution knee array coil as a receiver in a 1.5 Tesla Siemens (Erlanger) Symphony unit.
The images were acquired as axial T1-weighted volumes with a 3D magnetization prepared rapid acquisition gradient echo (MPRAGE); TR = 1,900 msec, TE = 4.38 msec, TI = 1,100 msec, flip angle = 15 degrees, and voxel resolution = 0.5 mm in all three planes. The nine separate images were aligned to each other in pair-wise rigid body registrations and averaged together prior to segmentation. The age at CSF biomarker collection and measurement (performed in 2012) was approximately 5 years following the age at MRI measurement (performed in 2007); analyses were performed with each animal from the CSF analysis paired with their own earlier scan data.
Details of the six brain phenotypes investigated in the initial association analysis have been described previously (Fears et al., 2009).
For the secondary analysis of regional cortical phenotypes, the images were segmented to generate cortical thickness and surface area measures using a combination of manual and automated methods. Thirty images were manually segmented into 29 regions based on each subject's anatomical details. The 30 manually segmented images were used as input for the Freesurfer (RRID:SCR_001847) program 'mri_ca_ train' to generate a probabilistic anatomical map (https://surfer.nmr. mgh.harvard.edu/fswiki/mri_ca_train) that was then used to segment the entire dataset. The Freesurfer segmented images were then inspected and manually adjusted to correct gross segmentation errors.

| Statistical analysis
Brain-CSF associations were tested using linear regression as implemented with the Sequential Oligogenic Linkage Analysis Routines (SOLAR) software package (Almasy & Blangero, 1998; RRID:SCR_000850). SOLAR uses the pedigree structure to adjust for the dependency structure among the subjects that would otherwise violate assumptions for standard linear regression methods. For the neuroimaging measures, age and sex were controlled as covariates in a linear regression model; the residuals were then associated with CSF traits. The procedure of Benjamini and Hochberg was used to correct for multiple testing to identify significant associations in the initial set of six neuroimaging traits with the four CSF traits (24 tests), using a false discovery rate (FDR) threshold of 0.05. In the secondary analysis of regional cortical measures of thickness and surface area with Aβ 1-40 and p-tau 181 traits, the procedure of Benjamini and Hochberg was similarly used, applied to each of the two sets of association tests independently (29 tests per set).

| Heritability calculations and linkage analysis
A whole-genome sequencing strategy was used to identify and type vervet polymorphisms, as previously described (Huang et al., 2015).
The vervet pedigree was visualized using the "kinship2" package.
Heritability analysis was performed using SOLAR. The variance component models of SOLAR assume normality in the trait distributions; therefore, in order to force the distribution of CSF biomarker phenotypes to approximately normal and correct for skew and kurtosis, the inverse normal transformation was used. For each phenotype, the correlation to age and sex covariates was evaluated. Residuals after regression for significant covariates were used in downstream analyses. Quantitative polygenic screening was performed as previously described using the "polygenic" function of SOLAR, yielding an estimate of the narrow-sense polygenic heritability (h 2 , a measure of the fraction of phenotypic variance due to additive genetic factors); a p-value for this estimate; and the proportion of variance attributable to covariates. Genetic mapping was performed using multipoint linkage analysis implemented in SOLAR. First, the multipoint identity-bydescent (IBD) relationships were computed for the vervet pedigree as previously described (Huang et al., 2015). A multipoint linkage scan was then run with an interval of 5 centimorgans. LOD scores were considered suggestive at a threshold of 1.9, and significant at a threshold of 3.3 as previously proposed (Lander & Kruglyak, 1995). Zero-in scans at finer scales were performed at peaks with LOD scores >2. The confidence region around a peak was defined as the region spanned by 1 LOD intervals to either side of the peak.

| Association testing
The EMMAX software was used to map genetic associations with CSF biomarker phenotypes in the vervet (Kang et al., 2010; RRID:SCR_008217). EMMAX uses a variance component model to account for high relatedness among the vervet subjects (estimated from pairwise kinship calculated by identity-by-state). A linear mixed model was then fit to the data and used to calculate the association.
We used standard genome-wide significant thresholds from humans (significant at p < 5 × 10 −8 , suggestive at p < 5 × 10 −6 ; Risch & Merikangas, 1996), which should approximate appropriate values in vervet given the relatively recent divergence between apes and Old World Monkeys (approximately 25 million years), and the similarity in genome size between human and vervet (Warren et al., 2015).

| Species differences and polymorphisms in the vervet β-amyloid and tau proteins
The antibody-based CSF assays were designed and validated for humans; therefore, we evaluated whether species-specific differences or polymorphisms in the vervet orthologs of APP (encoding the amyloid precursor protein) and MAPT (encoding tau) could affect the performance of the antibodies.
Species-specific differences between vervet and human in the antibody epitope regions were first identified. Across the Aβ peptide the human and vervet proteins were identical, including the 2G3, 4D7A3, and 3D6 epitope regions ( Figure S3). Differences were found between the human tau protein and the vervet ortholog, including a single amino acid substitution (threonine to alanine) in the AT120 epitope region, which could affect the results for total tau concentration.   (7) 29 (11) T A B L E 1 Characteristics of vervet subjects used in the analysis, at the time of CSF collection (Gly148Arg and Leu213Pro). None overlapped with the antibody epitope regions; furthermore, no associations between these polymorphisms and quantified CSF concentrations were observed ( Figure S4).

| Relation between CSF biomarker level and brain structure
Brain imaging phenotypes and genetic data had been collected as part of a previous study (Fears et al., 2009). In total, both CSF biomarker measurements and neuroimaging data were measured in 209 animals. In general, this subgroup had fewer animals of very young age, due to the time interval between studies ( Figure S5). To determine the relationship between the CSF concentrations of Aβ 1-40 , Aβ 1-42 , tau, and p-tau 181 with neuroimaging phenotypes, we initially calculated the correlations of CSF biomarker concentrations with six global measures (cerebral volume, cerebellar volume, hippocampal volume, cross-sectional area of the corpus callosum, total gray matter thickness, and total surface area; see Table 2 and Figure 1), after correction for age and sex. At a nominal significance threshold (p = .05), Aβ 1-40 was correlated with corpus callosum cross-sectional area (p = .024), hippocampal volume (p = .013), total gray matter thickness (p = .007), and total surface area (p = .006). The correlation with total gray matter thickness and total surface area was significant (controlling FDR at 0.05) after applying the Benjamini-Hochberg correction for multiple comparisons. The direction of each regression coefficient was positive; higher concentrations of Aβ 1-40 were associated with larger brain size. Likewise, p-tau 181 concentrations were correlated with cerebral volume (p = .001), cerebellar volume (p = .039), corpus callosum cross-sectional area (p = .018), hippocampal volume (p = .039), and total surface area (p = .008). The correlation with cerebral volume and total surface area was significant after the Benjamini-Hochberg correction. In contrast with Aβ 1-40 , the sign of each regression coefficient was negative; that is, higher concentrations of p-tau 181 tended to be associated with smaller brain sizes.
Because we observed suggestive association between Aβ 1-40 and p-tau 181 and multiple neuroimaging traits, we further tested for associations between CSF concentrations of Aβ 1-40 and p-tau 181 and regional measures of gray matter thickness and surface area across the cortex (Table S1). MRI measurements were adjusted for age and sex, and the residuals were used for downstream analyses. with those in the global brain measures; Aβ 1-40 was generally correlated with increased gray matter thickness and surface area, while p-tau 181 was generally correlated with decreased gray matter thickness and surface area of brain structures. Of these brain regions, only the gray matter thickness in the fusiform gyrus was significantly correlated with age, after correction for multiple comparisons (uncorrected p = 3.68 × 10 −14 ). Fitting a linear regression model for fusiform gyrus gray matter thickness using both CSF p-tau 181 and age at MRI scan as covariates, we found that both age (β = 0.20, p = 1.03 × 10 −6 ) and p-tau 181 (β = 0.20, p = .0015) were strong predictors. This pattern remained even after including age at CSF draw (p = .00057 for p-tau 181 ), suggesting that the contributions of CSF Aβ 1-40 and p-tau 181 on regional brain volume measurements that we have identified are independent of age.

| Heritability and genetic mapping of CSF biomarkers
We first sought to quantify the heritability of CSF Aβ 1-40 , Aβ 1-42 , tau, and p-tau 181 concentrations. Narrow-sense polygenic heritability (h 2 ) T A B L E 2 Associations of global measures of brain size with CSF biomarkers was estimated using SOLAR, taking into account age and sex as possible covariates. After correcting for relatedness, CSF Aβ 1-40 (p = .001) and p-tau 181 (p = .01) were significantly correlated with age, but not Because of the demonstrated heritability of CSF Aβ 1-40 concentration, we then attempted to map genetic loci that could play a role in the phenotypic variation using genome-wide multipoint linkage analysis. At a LOD score threshold of 3.3, no markers reached genomewide significance ( Figure S6). Two peaks were suggestive for linkage (LOD > 1.9), including a 9 cM region on vervet chromosome 4, between 67 Mb and 74 Mb (maximum LOD of 2.5) and a 6 cM region on vervet chromosome 12, between 30 Mb and 37 Mb (maximum LOD of 2.2). We additionally performed association as a complementary mapping approach, correcting for family structure using a linear mixed model approach implemented using the EMMAX software (Kang et al., 2010). No markers reached genome-wide significance, but a suggestive association was identified on chromosome 4 near the proximal end of the linkage peak (minimum p = 5.68 × 10 −8 at marker CAE4_67028748, Figure S7). Genes within 500,000 base pairs of this marker included MAP1B, MRPS27, PTCD2, ZNF366, TNPO1, FCHO2, TMEM171, and TMEM174.

| DISCUSSION
We report the measurement of CSF biomarkers in a colony of vervet monkeys, a potentially valuable nonhuman primate model for neurodegenerative diseases. This dataset spans vervet monkeys of varying age, from 1 year of age to 27 years of age. For context, vervets typically live to 11-13 years of age in the wild, and about 25 years in captivity (Magden et al., 2015). As AD and other neurodegenerative diseases are highly age-related in humans, the study of this large dataset is ideal for dissecting the applicability of vervets in human disease.
AD in humans is thought to result in decreased CSF Aβ 1-42 by accumulation of Aβ 1-42 in neuritic plaques (Fagan et al., 2006) resulting in reduced drainage through the CSF (Lautner et al., 2014;Strozyk, Blennow, White, & Launer, 2003), and in increased tau by neuronal death (Franz et al., 2003;Itoh et al., 2001) and active secretion (Chai, Dage, & Citron, 2012;Saman et al., 2012). Phospho-tau, unlike total tau, seems to have greater specificity to neurofibrillary tangle pathology, and does not drastically increase in some neurodegenerative F I G U R E 1 Associations between global measures of brain size and CSF biomarkers (Aβ 1-40 , Aβ 1-42 , tau, and p-tau 181 ) in 209 vervets with both CSF collection and MRI scan. Colored points denote the strength of association -non-significant association (p > .05, black), suggestive association (p < .05, yellow), and significant association (Benjamini-Hochberg corrected p < .05, red) diseases such as Creutzfeldt-Jakob disease (Itoh et al., 2001) and acute stroke (Hesse et al., 2001). The relationship of CSF Aβ 1-40 with AD has been less clear, though CSF Aβ 1-40 has been reported as decreased in CAA (Verbeek et al., 2009). Here, we found that reductions in global measures of brain size (such as cerebral volume and total brain surface area) were associated with reduced CSF Aβ 1-40 and elevated CSF p-tau 181 . Taken together with previous observations of age-related amyloid plaques in the vervet (Kalinin et al., 2013;Lemere et al., 2004), the relationships with Aβ 1-40 and p-tau 181 are consistent with a scenario of age-related accumulation of amyloid plaques (in parenchyma and/or vasculature) corresponding to cerebral atrophy in the vervet. Though previous studies have found limited phospho-tau immunoreactivity and few neurofibrillary tangles in the vervet brain (Lemere et al., 2004), the fact that elevations in p-tau 181 levels in CSF are also correlated with changes in brain morphometry suggest that tau pathology may also play a role.
These findings are reminiscent of the observed relationship between CSF biomarkers and neuroimaging measurements in aging and neurodegenerative disease in humans. Previous work has demonstrated associations between longitudinal changes in Aβ 1-42 and ptau 231 and hippocampal volume in AD and mild cognitive impairment (Chou et al., 2009;De Leon et al., 2006;Hampel et al., 2005;Schuff et al., 2009) as well as in healthy brain aging (Fjell et al., 2010) in humans. However, we identified significant associations of Aβ 1-40 , not Aβ 1-42 , with age, gray matter thickness, and other measures of brain size. In vervets (Lemere et al., 2004) and in humans ( (Attems et al., 2011;Vinters & Gilbert, 1983 concentrations in the vervet. Linkage analysis identified two suggestive linkage peaks: a wide 9 cM peak on chromosome 4, and a narrow 6 cM peak on chromosome 12. The lack of statistical significance in linkage analysis despite the moderate heritability of CSF Aβ 1-40 concentration may suggest a highly polygenic contribution to the phenotype. While the signal did not reach genomewide significance and spanned a wide genomic range, precluding identification of a single causal gene or variant, our work provides a basis for further fine mapping of these regions. Potential candidate genes in the region include PTCD2 and MAP1B, at the proximal end F I G U R E 2 Regional associations of cortical gray matter thickness and CSF Aβ 1-40 in the vervet brain; coloration of the region indicates suggestive (p < .05, orange), and significant association (Benjamini-Hochberg corrected p < .05, red) of the linkage peak and approximately 330 and 470 kilobases from the top associated marker. The pentatricopeptide repeat domain 2 gene, encoded by PTCD2, has been observed to be accumulated in AD brain in citrullinated form, and autoantibodies against PTCD2 have been observed in AD patient serum (Acharya et al., 2012). The microtubule-associated protein 1B, encoded by MAP1B, has been described to bind tau and Aβ (Gevorkian et al., 2008;Hasegawa, Arai, & Ihara, 1990). In humans, genome-wide association studies have demonstrated genetic loci that affect CSF biomarkers, the most robust being the region surrounding the APOE gene (Cruchaga et al., 2013). The linkage analysis here may provide a foothold to discover analogous loci for vervets, but further work remains to validate and pinpoint them.
One observed anomaly was the fact that concentration of total tau was lower than that for p-tau 181 in many CSF specimens. This was unexpected because the phosphorylated species is also reactive to the antibodies against total tau, and therefore the concentration of total tau should be an upper bound for that of the phosphorylated species. Because the calibration standards and controls were derived from human tau, we postulated that non-homology in the epitope region may confound the assay. Indeed, we found an amino acid difference in the vervet tau ortholog in the AT120 epitope region that may affect the specificity of the antibody. We speculate that this lack of homology accounts for the aberrant total tau levels that were observed and, therefore, the measurement of the total tau using the AlzBio3 assay in vervets must be interpreted with caution. The epitope regions for all of the other antibodies used to quantify levels of Aβ 1-40 , Aβ 1-42 , and p-tau 181 were identical between human and vervet.
Another limitation of the study is the fact that neuroimages and CSF samples were not acquired simultaneously; the interval between imaging and CSF collection was variable, but imaging was performed approximately 5 years before CSF collection for each sample. Somewhat mitigating the possibility for extensive confounding, we had previously demonstrated high heritability of neuroimaging phenotypes, in some instances approaching 100%; therefore, environmental influences are likely to be low (Fears et al., 2009). We were able to identify associations between CSF markers and neuroimaging despite this gap, which could introduce further variability into the analysis; therefore, the associations we positively identified are likely to reflect biology. However, the temporal relationship and relative effect sizes of the CSF marker-neuroimaging relationship are not known. In other words, differences in CSF Aβ and tau concentrations on a population level in vervet can be confidently correlated to aging and neurodegeneration (after controlling for aging), but the sensitivity and degree of correlation remains unclear. Although we cannot make a conclusion about the prognostic power of CSF markers, preliminary data suggests that CSF Aβ 1-42 is a predictor of parenchymal amyloid plaques (Suzanne Craft, personal communication). Interestingly, neither cerebral volume or p-tau 181 were associated with age across the wide range of ages in the vervet cohort; however, the strongest association was detected between p-tau 181 and cerebral volume, suggesting that at least some of the associations were not driven by dependency on age. Furthermore, frank cognitive deficits were not observed or ascertained in the vervet subjects, and the study design was cross-sectional; correlation with measures of brain structure and function in future longitudinal studies may provide additional insight into the extent of neurodegeneration.
In vervets, CSF Aβ 1-40 decreases with advancing age, correlates with global and regional measures of brain size, and has a large heritable component, suggesting its utility as a biomarker of AD-or CAA-like changes in this potentially valuable animal model. Aβ  was most closely associated with decreases in gray matter thickness in the frontal lobe and hippocampus, mirroring the regional vulnerability patterns of AD. However, structural associations were also observed in the occipital lobe and precuneus, regions that are generally spared early in Alzheimer's disease and more reminiscent of the distribution of pathology in CAA. Age-related cerebral atrophy in the vervet likely involves β-amyloid pathological processes, and Aβ 1-40 and p-tau 181 appear to be robust biomarkers of these processes, though their sensitivities and correlations to disease pa- thology remain important open questions. We have also mapped suggestive linkage peaks on vervet chromosomes 4 and 12 that correspond to CSF Aβ 1-40 concentration and may represent a modifier of neurodegeneration. The aged vervet therefore has the potential to advance our understanding of the genetics, pathogenesis, and treatment of neurodegenerative disease.