Progranulin levels in blood in Alzheimer's disease and mild cognitive impairment

Abstract

Objective

Changes in progranulin (GRN) expression have been hypothesized to alter risk for Alzheimer's disease (AD). We investigated the relationship between GRN expression in peripheral blood and clinical diagnosis of AD and mild cognitive impairment (MCI).

Methods

Peripheral blood progranulin gene expression was measured, using microarrays from Alzheimer's (n = 186), MCI (n = 118), and control (n = 204) subjects from the University of California San Francisco Memory and Aging Center (UCSF‐MAC) and two independent published series (AddNeuroMed and ADNI). GRN gene expression was correlated with clinical, demographic, and genetic data, including APOE haplotype and the GRN rs5848 single‐nucleotide polymorphism. Finally, we assessed progranulin protein levels, using enzyme‐linked immunosorbent assay, and methylation status using methylation microarrays.

Results

We observed an increase in blood progranulin gene expression and a decrease in GRN promoter methylation in males (P = 0.007). Progranulin expression was 13% higher in AD and MCI patients compared with controls in the UCSF‐MAC cohort (F _2,505 = 10.41, P = 3.72*10⁻⁵). This finding was replicated in the AddNeuroMed (F _2,271 = 17.9, P = 4.83*10⁻⁸) but not the ADNI series. The rs5848 SNP (T‐allele) predicted decreased blood progranulin gene expression (P = 0.03). The APOE4 haplotype was positively associated with progranulin expression independent of diagnosis (P = 0.04). Finally, we did not identify differences in plasma progranulin protein levels or gene methylation between diagnostic categories.

Interpretation

Progranulin mRNA is elevated in peripheral blood of patients with AD and MCI and its expression is associated with numerous genetic and demographic factors. These data suggest a role in the pathogenesis of neurodegenerative dementias besides frontotemporal dementia.

Introduction

The 88 kDa progranulin and its 6 kDa processed products (granulins) represent a class of secreted proteins with diverse functions peripherally and in the brain. Granulins and progranulin have been implicated as potent immunomodulators and cell‐cycle regulators. In the CNS, progranulin expression increases with age in both neurons and microglia, and plays a role in neurite outgrowth, synapse modification, and the prevention of neuronal apoptosis.1, 2, 3, 4 This neuroprotective function is highlighted by the relationship between progranulin and neurodegenerative disease; heterozygous loss‐of‐function mutations in the gene encoding progranulin (GRN) cause frontotemporal dementia (FTD),5, 6 and a common rs5848 allele in the 3′UTR of GRN has been associated with both decreased serum and brain progranulin expression levels and increased risk of developing Alzheimer's disease (AD).7, 8, 9 Additionally, misregulation of progranulin expression has been implicated in parkinsonism, neuronal ceroid lipofuscinosis, and other neuropsychiatric disorders.10, 11

Mutations in the GRN gene may contribute to the risk of developing AD.8, 12 Additionally, progranulin localizes at the margins of amyloid plaques in both mouse and human postmortem brain tissue, and increased progranulin mRNA levels have been reported in the brains of multiple AD‐mouse models.13 Overexpression of progranulin in these models has been shown to slow plaque deposition and cognitive decline.14, 15 Taken together, these data suggest a direct role of progranulin in AD pathogenesis. Previous studies examining protein levels in peripheral blood failed to detect a relationship between peripheral progranulin protein levels and AD status.16 We first observed a relationship between GRN mRNA levels and AD status in a small patient series,17 but no study to date has conclusively shown a connection between progranulin expression and sporadic AD.

Early detection of AD has increasingly become a focus of the biomedical community, as future treatment modalities will likely hinge on slowing or preventing neurodegeneration before it has occurred. As such, mild cognitive impairment (MCI)–defined by focal memory or executive function deficits not explained by age, without dementia or loss of day‐to‐day function – has become a major focus of study.18 MCI patients convert to AD at a rate of 10–15% per year.19, 20 However, the majority of MCI patients will not transition to dementia, and some may display spontaneous improvement.19 Currently, estimates of hippocampal or entorhinal cortex volume coupled with cognitive function testing are the best predictors of disease transition.20, 21 However, reconstructive MRI imaging is expensive and not easily transferable to community hospitals. Thus, further work is needed to uncover peripheral blood markers of MCI.

We studied GRN expression levels in peripheral blood in a large patient series with AD and MCI, and asymptomatic controls, as well as in multiple datasets from the literature. We correlated GRN mRNA levels with demographic characteristics, disease status, genetic risk factors, methylation at the GRN locus, and progranulin protein levels as assayed by ELISA.

Material and Methods

Subjects and samples

This study received prior approval from the Institutional Review Board at the University of California San Francisco, and informed consent was obtained from subjects prior to study enrollment and sample collection.

UCSF‐MAC cohort

In this study, 530 patients clinically diagnosed as either AD, MCI, or unaffected controls were enrolled at the UCSF‐MAC between 2006 and 2016. MRI and Amyloid PET imaging were not uniformly performed to support diagnosis.

Peripheral blood from each subject was collected in Paxgene tubes and kept in ice prior to total RNA isolation. RNA extraction was performed, using the RNeasy QIAcube extraction kit (Qiagen) and RNA quantity was determined, using a Nanodrop instrument (Nanodrop Technologies). RNA quality was assessed with the Agilent Bioanalyzer (Agilent Technology) and samples with an RNA Integrity Number (RIN) <7 were excluded. RNA libraries were hybridized to Illumina HumanHT‐12 V4.0 microarrays at the UCLA Neuroscience Genomics Core. Microarray slides were scanned and signal processed using Illumina BeadStation and the BeadStudio software package in preparation for subsequent analysis. Here, 22 samples were ultimately excluded due to poor RIN or were detected as outliers (described below) for a final cohort size of n = 508.

AddNeuroMed cohort

The AddNeuroMed cohort 1 is publicly available in the Gene Expression Omnibus (GEO) repository (Accession: GSE63060).22 Briefly, this patient cohort is composed of 329 samples diagnosed as AD, MCI, or control with RNA hybridized to Illumina HumanHT‐12 V3.0 microarrays. Ultimately 21 outliers were removed and an additional 34 samples were dropped to correct for an age confound (described below) for a final cohort size of n = 274.

ADNI

Data used in this manuscript were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 under the direction of Principal Investigator Dr. Michael W. Weiner, MD to interrogate whether biological markers and clinical assessment can be combined to measure the progression of MCI and early AD.

The ADNI_Gene_Expression_Profile dataset is made available from the Alzheimer's Disease Neuroimaging Initiative (ADNI) repository. Samples were prepared as described (http://adni.loni.usc.edu/wp-content/uploads/2008/07/ADNI_GO_Procedures_Manual_06102011.pdf). Briefly, 811 patients in the ADNI cohort were categorized as Control, Early MCI, Late MCI, or AD using published clinical criteria. Total RNA was extracted and hybridized to an Affymetrix Human Genome U219 array and scanning and signal extraction were performed, using the Affymetrix GeneTitan system. After quality control, outlier removal, and stratification for an age confound (described below), the final processed cohort was n = 617.

San Antonio family heart study

The San Antonio Family Heart Study (SAFHS) dataset23 is publicly available through ArrayExpress (http://www.ebi.ac.uk/arrayexpress) under the accession E‐TABM‐305. This cohort is composed of 1240 peripheral blood lymphocyte samples hybridized to Illumina Human‐6 v1 Expression BeadChip microarrays. Following outlier removal, we stratified this dataset to remove collinearity between sex and smoking status (described below). We then excluded all samples below 30 years old to match the age of our other datasets. The final processed cohort was n = 543.

Genotyping

Subjects in the UCSF‐MAC cohort had genomic DNA isolated from peripheral blood following standard procedures. APOE and GRN rs5848 genotyping were carried out by real‐time PCR on an Applied Biosystems 7900HT Real Time PCR machine (Applied Biosystems, Foster City, CA), using Taqman SNP Genotyping Assays (#C___7452046_20, C___3084793_20, and C____904973_10 for rs5848, rs429358, and rs7412, respectively). Assays were run in triplicate. The SDS version 2.3 software was used to analyze the raw data and to call the genotypes.

Array processing

Microarray raw signal processing for the UCSF‐MAC cohort was performed using the lumi package.24 First, within‐sample raw gene expression intensities were normalized using variance‐stabilized transformation (VST)25 and interarray normalization was performed with robust spline normalization. Probes with a detection score below standard threshold (P = 0.01) for all samples were dropped along with probes not annotated within the lumiHumanAll.db database. Next, ComBat from the sva package26 was used to perform batch correction. Outliers were removed using a connectivity Z‐score (threshold > 2) calculated using the fundamentalNetworkConcepts function from the WGCNA package.27 The AddNeuroMed cohort was processed using the same pipeline except that log2 normalization was used instead of VST because some quality control information was missing from the raw data.

The raw data for the ADNI dataset is not publicly available. Therefore, we performed analysis on preprocessed array data which were normalized, using standard Robust‐Multi‐Array Averaging from the affy package.28 We subsequently excluded 33 outliers using connectivity Z‐scores (threshold > 2) and 11 samples with a RIN < 7, and performed batch correction using ComBat. Probes were annotated using the Affymetrix hgu219.db database from BioConductor. The SAFHS cohort was analyzed using the same pipeline as the UCSF‐MAC cohort except for batch correction, which was not performed as batch information was not available.

Removal of confounding covariates

We used linear modeling and G‐tests29 to determine if age or sex were collinear with diagnosis using a significance threshold of P < 0.05. When significant collinearity was observed, samples were stratified, using a randomized nonbiased approach to drop samples until collinearity was no longer observed. Thus, final analysis was run on cohorts with age, sex, and diagnosis verified as independent variables. For the SAFHS dataset, sample stratification was used to remove collinearity between sex and smoking status.

Statistical analysis

Intensity values for probes querying GRN expression were collected from each platform. Each GRN probe was verified for every cohort by ensuring that average expression intensity was greater than the 60th percentile of all detected probes – no probes were dropped. For Illumina platforms, two progranulin probes were identified: AAGGCTCGATCCTGCGAGAAGGAAGTGGTCTCTGCCCAGCCTGCCACCTT (Probe 1, mapping to exon 11) and GGCCTTCCCTGTCAGAAGGGGGTTGTGGCAAAAGCCACATTACAAGCTGC (Probe 2, mapping to the 3′UTR). Neither location harbors known SNPs or InDels at a high population frequency (>1:1000). As probes were correlated (r ² = 0.75), we reported only probe 1, which also had higher mean expression values. Linear models were used to assess significance of diagnosis or genotype with gender, age, and interactions (when appropriate) included as covariates. An F‐test was used to assess significance of categorical predictors with more than two groups, and Welch's T‐test was used for continuous predictors or categorical predictors with two groups. Post hoc pairwise testing was done using Tukey's test with a significance threshold of P < 0.05. The Wilcoxon rank sum test was used in a case of small, nonnormal sample.

Quantification of progranulin protein by ELISA

Progranulin protein in human plasma and cerebrospinal fluid (CSF) was quantified by A&G Pharmaceutical Inc. (Columbia, MD) using their Progranulin (GP88/PGRN) ELISA, which detects full‐length progranulin in both biofluids. The detection limit of this assay is 100 pg/mL with a working range up to 20 ng/mL. Assay details have been described previously.30, 31, 32 Briefly, a subset of patients enrolled in the UCSF‐MAC cohort underwent peripheral blood draws (n = 266) and/or lumbar puncture (n = 80) during multiple follow‐up visits. Each sample was run in duplicate and normalized against a recombinant progranulin standard and two reference serum samples. A coefficient of variation score (SD/mean*100) was calculated for every sample. The 11 plasma samples with a coefficient of variation >15% were dropped as a quality control measure. Patients who underwent multiple draws had their plasma or CSF PGRN values averaged. Outliers were removed using the R boxplot function. Statistical analysis was done using the same models and tests as described for gene expression.

Methylation

DNA was extracted from peripheral blood using standard methods. DNA methylation was quantified using the Illumina Human Methylation 450K microarray. Preprocessing was run with the RnBeads package33, using the default options for quality control. Background correction was performed with the normal exponential convolution, using out‐of‐band probes (noob) method,34 and arrays were normalized using beta mixture quantile dilation (BMIQ).35 The normalized data was corrected for batch effect using the parametric empirical Bayes method from ComBat.26 The RnBeads package was also used to extract GRN promoter methylation β‐values, which were computed by averaging the β‐values of all 10 probes from 1.5 kb upstream to 0.5 kb downstream of the transcription start site. RNBeads was similarly used to extract gene body methylation, again by averaging the β‐values of all 10 probes from the transcription start site to the end of the gene.

Results

Demographic and diagnostic determinants of GRN expression

General characteristics of all processed cohorts (postquality control and confounder stratification) are described in Table 1. We first analyzed the UCSF‐MAC cohort (n = 508).

Table 1.

Summary characteristics of four patient cohorts analyzed

Cohort	Diagnosis	Sample (n)	Male (n)	Female (n)	Mean Age	SD Age
UCSF‐MAC	Combined	508	240	268	69.8	10.4
	Control	204	95	109	70.9	11.2
	MCI	118	61	57	69.7	10.1
	AD	186	84	102	68.8	9.5
AddNeuroMedd	Combined	274	107	167	72.9	5.7
	Control	95	38	57	71.8	5.8
	MCI	66	32	34	73.1	4.9
	AD	113	37	76	73.7	5.9
ADNI	Combined	617	326	291	73.8	6.5
	Control	240	114	126	74.3	5.6
	eMCI	159	91	68	73.7	5.8
	lMCI	178	96	82	73.0	7.2
	AD	40	25	15	75.6	9.8
SAFHS	Combined	543	252	291	49.7	13.5
	Smoker	158	85	73	48.1	12.0
	Nonsmoker	385	167	218	50.2	13.9

First, we assessed the relationship between gene expression, sex, and age, after ensuring no confounding between our predictors. We verified no collinearity between sex and diagnosis in our dataset (G‐test, G‐statistic₍₂₎ = 1.3, P = 0.52) nor between age and diagnosis (F _2,505 = 2.04, P = 0.13). We next fit a linear model for our covariates. We observed that GRN mRNA expression levels were significantly higher in males than in females (log₂ fold‐change [log₂FC] = 0.09, corresponding to a 6% increase, T‐statistic = 6.26, P = 0.013, Fig. 1A) across diagnostic categories, and that age was significantly but trivially correlated with progranulin mRNA levels (Pearson's r ² = 0.008, P = 0.04; Fig. 1B) consistent with previous reports.11 We next compared GRN levels between diagnostic categories (F _2,505 = 10.41, P = 3.7*10⁻⁵). Patients diagnosed with AD had a statistically significant increase (log₂FC = 0.17, corresponding to +13%, Tukey's test P = 0.00019) in GRN mRNA expression compared with controls. We also observed a similar increase (log₂FC = 0.17, P = 0.0012) in the MCI group compared with controls (Fig. 1C). We observed no significant differences in GRN mRNA levels between AD and MCI. When we modeled interactions between sex and diagnosis, as well as age and diagnosis, we did not observe significant effects, arguing against a synergistic relationship between these covariates.

Figure 1.

Effect of demographic characteristics and disease status on peripheral blood GRN gene expression. (A) Progranulin gene expression is higher in males than females. UCSF‐MAC cohort, Welch's t‐test. (B) Progranulin gene expression increases with age across diagnostic categories. UCSF‐MAC cohort, linear regression. (C) Progranulin expression is higher in AD and MCI patients compared with controls. UCSF‐MAC cohort, F‐test and analysis of variance (AOV) with Tukey's post hoc test. (D) same as (C) in the AddNeuroMed cohort. (E) lack of difference in progranulin expression in AD, early MCI (EMCI), and later MCI (LMCI) patients compared with controls. ADNI cohort, F‐test and aov with Tukey's post hoc test. General cohort characteristics are described in Table 1. (A, C, D, E) standard boxplot representing median and interquartile range (IQR), whiskers represent 1.5 IQR greater or less than the upper and lower quartile. (B) scatterplot with line of best‐fit. *P < 0.05, ***P < 0.001, n.s. P > 0.05.

We next sought to replicate our findings. We first analyzed the AddNeuroMed Cohort (n = 274; Table 1). We initially observed collinearity between age and diagnosis (F _2,305 = 5.81, P = 0.003) and therefore stratified and discarded 34 samples until collinearity was no longer significant (F _2,271 = 3.0, P = 0.051). Sex remained independent after stratification (G₍₂₎ = 4.38, P = 0.11). We again fit a linear model with all covariates and relevant interaction terms. Progranulin mRNA expression again was higher in males compared with females (log₂FC = 0.12, +9%, T‐statistic = 4.4, P = 0.037), but age was no longer predictive (T‐statistic = 2.3, P = 0.13). We observed a significant effect based on diagnosis (F _2,271 = 17.9, P = 4.8*10⁻⁸) but no interaction effects. Similar to the UCSF‐MAC cohort, we observed a significant increase (log₂FC = 0.28, corresponding to +21%, TukeyHSD; P = 5.6*10⁻⁶) in progranulin mRNA expression for AD patients compared with controls, and an increase (log₂FC = 0.35, +27%, P = 6.4*10⁻⁷) for MCI patients compared to controls (Fig. 1D).

We next validated these sex and age findings in a third dataset, the San Antonio Family Heart Study cohort (diagnosis was not relevant for this dataset). After stratifying to correct for collinearity between sex and smoking status, we excluded all samples under 30 years of age to more appropriately match mean cohort age with our other cohorts while still maintaining statistical power (n = 543, mean age = 49.7 years). We ensured no collinearity between age (T‐statistic = 1.06, P = 0.3) or smoking status (G₍₂₎ = 4.48, P = 0.08) and sex and fit a linear model. We again found that progranulin was higher in males than females (log₂FC = 0.07, +5%, T‐statistic = 4.8, P = 0.03) and found no correlation with age (T‐statistic = 0.04, P = 0.8). Interestingly, smoking status was an independent predictor of blood progranulin mRNA expression in this cohort, with a 6% GRN increase in smokers (log₂FC = 0.08, T‐statistic = 4.5, P = 0.03).

Finally, we analyzed the ADNI patient cohort (n = 617), which is divided into asymptomatic controls, early MCI (eMCI), late MCI (lMCI, with regard to clinical disease progression), and AD. This cohort was initially confounded with significant collinearity between age (F _3,696 = 10.9, P = 5*10‐7), sex (G₍₃₎ = 11.1, P = 0.01) and diagnosis. We stratified and excluded samples (described in methods) until age (F _3,613 = 2.49, P = 0.06) and sex (G₍₃₎ = 5.58, P = 0.13) were no longer collinear with diagnosis. However, a plate batch effect remained (G₍₂₄₎ = 38.9, P = 0.03), which we corrected with the ComBat package. In this cohort, neither age nor sex were predictive of progranulin expression, and we did not observe differences in progranulin expression levels between diagnostic categories (F _3,611 = 0.31, P = 0.82, TukeyHSD post hoc P > 0.05 for all comparisons; Fig. 1E).

Effects of genotype on GRN expression

Next, we genotyped samples from the UCSF‐MAC cohort for the risk‐associated rs5848 GRN variant. Homozygotes for the risk‐associated allele (T:T, n = 25) had a 12.3% decrease in progranulin mRNA expression compared with C:C carriers (log₂FC = −0.19, Wilcoxon Rank sum test, P = 0.02) supporting an earlier report of decreased GRN mRNA levels in postmortem brain tissue of T:T genotype carriers.36 We also found that T‐allele carriers had significantly lower progranulin levels than C:C homozygotes (log₂FC = −0.1, Welch, T₅₀₆ = 2.24, P = 0.03; Fig. 2A). After ensuring there was no collinearity between the rs5848 genotype and sex (G‐statistic₁ = 0.21, P = 0.65), we fit a linear model using sex and haplotype as an interaction term, but did not observe any significant interaction in our data. Additionally, using a Fisher's exact test, we observed an enrichment of T‐genotype carriers in the AD population compared with controls, which did not reach statistical significance (OR = 1.46, 95% CI = 0.95–2.38, P = 0.09). Thus, it is noteworthy that AD patients on average still have elevated progranulin gene expression despite the overrepresentation of rs5848 T‐allele carriers (associated with lower progranulin levels) in the AD patient population.

Figure 2.

Common AD‐associated genetic variants and GRN mRNA expression in peripheral blood. (A) Progranulin gene expression is lower in rs5848 risk allele carriers (C:T or T:T genotypes) compared with low‐risk genotype (C:C). UCSF‐MAC cohort, Welch's t‐test. (B) Progranulin gene expression is higher in APOE risk allele (E4) carriers (homozygote or heterozygote) compared with low‐risk haplotypes (E2 or E3 carriers). UCSF‐MAC cohort, Welch's t‐test. (A‐B) standard boxplot representing median and IQR, whiskers represent 1.5 IQR greater or less than the upper and lower quartile. *P < 0.05, **P < 0.01.

We also studied the relationship between GRN gene expression levels and the AD risk‐associated APOE haplotype.37 We observed a significant effect of the APOE haplotype on progranulin levels in the UCSF‐MAC cohort across diagnostic categories, with progranulin mRNA levels significantly higher in E4 risk allele carriers compared with E2/E3 haplotypes (log₂FC = 0.12, +9%, Welch, T ₍₄₄₁₎ = 2.82 P = 0.005, Fig. 2B). However, there was also significant collinearity between APOE haplotype and disease status; the E4 haplotype was significantly enriched in AD patients (Fisher's exact test, OR = 4.83, P = 6.3*10⁻¹¹) compared with controls, confounding our results.

Therefore, we attempted to estimate the independent relationship between APOE haplotype and GRN expression. After removing the effects of diagnosis as a confound using a parametric empirical Bayesian estimator (ComBat), we still observed significantly higher progranulin gene expression levels in ApoE4 haplotype carriers (log₂FC = 0.08, Welch T‐test, P = 0.04), suggesting that APOE4 genotype might function as a trans expression quantitative trait locus (eQTL) controlling GRN expression. We validated this finding in the Genotype‐Tissue Expression consortium dataset (GTEx).38 Analysis of rs429358 SNP (minor C‐allele tags the APOE4 haplotype) and RNA‐seq data from 328 samples revealed a significant association between expression and genotype (effect size = 0.12, P = 0.004) in whole blood. We checked the relationship between APOE4 carrier status and GRN expression in the ADNI dataset but failed to detect a difference in GRN expression between APOE4 carriers and controls (T ₍₆₁₃₎ = 1.43, P = 0.15). We also performed the converse analysis in the UCSF‐MAC cohort. Using ComBat to remove the effect of APOE haplotype, we then modeled the effect of diagnosis on progranulin expression. We found that AD samples still had significantly higher progranulin expression (log₂FC = 0.17, T ₍₄₅₀₎ = 3.9, P = 0.0001) compared with controls suggesting that diagnosis predicts progranulin expression independent of APOE4 sample overrepresentation.

GRN protein levels in CSF and plasma

We next utilized ELISA to quantify progranulin protein levels in plasma (n = 266) and CSF (n = 80) in a subset of patients from the UCSF‐MAC cohort. After ensuring no collinearity with age (F _2,263 = 1.84, P = 0.16) and sex (G‐statistic₂ = 2.32, P=0.31), we observed no correlation between plasma progranulin protein and gene expression in peripheral blood (Pearson's r = 0.065, df = 206, P = 0.35; Fig. 3A). However, we observed a sex dimorphism in our plasma data, with females across diagnostic categories having significantly higher progranulin protein levels than males (fold‐change = 0.07, Welch's T ₂₆₄ = 2.14, P = 0.03, Fig. 3B) which recapitulates an earlier study,16 and is the opposite finding from our gene expression data, where males had higher GRN mRNA expression. We identified no correlation between age and plasma GRN protein levels (Pearson's r = 0.09, df = 264, P = 0.16). We next assessed the relationship between disease status and progranulin protein levels. In plasma, we found no difference between GRN protein levels across all disease categories (F _2,263 = 1.72, P = 0.18, Tukey's post hoc P > 0.05 all comparisons; Fig. 3C) again replicating previous findings.16 We observed no correlation between progranulin levels in the CSF and in plasma (Pearson's r = 0.13, df = 63, P = 0.31; Fig. 3E). AD patients had statistically lower CSF progranulin levels than control subjects (F _2,66 = 3.45, P = 0.04, TukeyHSD P = 0.03 for AD vs. Control, Fig. 3D), confirming an earlier report.39 We observed no relationship between age or sex and progranulin levels in CSF.

Figure 3.

Progranulin protein levels by ELISA in plasma and cerebrospinal fluid (CSF). (A) Plasma progranulin protein levels are not correlated with gene expression within patients. n = 266, linear regression. (B) Females have significantly higher plasma GRN protein levels than males. Welch's t‐test. (C) Lack of difference in plasma GRN protein levels between AD, MCI, and control patients. Analysis of variance (AOV), Tukey's post hoc test. (D) AD patients have significantly lower CSF GRN protein levels than MCI and control patients. n = 80, AOV, Tukey's post hoc test. (E) Lack of correlation between plasma and CSF GRN protein levels. n = 80, linear regression. All analysis performed with the UCSF‐MAC cohort. (B–D) standard boxplot representing median and IQR, whiskers represent 1.5 IQR greater or less than the upper and lower quartile. (A,E) scatterplot with line of best‐fit. *P < 0.05, n.s. P > 0.05.

Analysis of GRN methylation status

Finally, we analyzed DNA methylation at the GRN locus in a subset of subjects (AD patients [n=128] and controls [n=227]) from the UCSF‐MAC cohort. We found no correlation between GRN promoter CpG methylation and GRN expression (T ₍₉₁₎ = 0.16, r = 0.02, P = 0.8) nor GRN gene body methylation and GRN gene expression (T ₍₉₁₎ = 1.52, r = −0.16, P = 0.13). We next fit a linear model predicting methylation status, using all relevant predictors. Males had lower GRN promoter CpG methylation than females (Log₂FC = 0.004, T ₍₃₅₁₎ = 7.32, P = 0.007; Fig. 4C) but no differences in gene body methylation T ₍₃₅₁₎ = 1.06, P = 0.3). Conversely, there was a negative correlation between age and GRN gene body methylation; methylation decreased significantly with age (T ₍₃₅₁₎ = 62.8, r = ‐0.39, P = 3*10⁻¹⁴; Fig. 4D). We identified no correlation between GRN promoter methylation and age (T ₍₃₅₁₎ = 1.11, r = 0.05, P = 0.29) in agreement with Galimberti and colleagues.40 We identified no significant difference in methylation β‐values for either the GRN promoter (T ₍₃₅₁₎ = 0.07, P = 0.79) or gene body (T ₍₃₅₁₎ = 0.17, P = 0.67) between AD and control patients (Fig. 4A,B).

Figure 4.

Analysis of GRN promoter and gene body DNA (CpG) methylation in DNA from peripheral blood. (A–B) lack of difference in promoter (A) or gene body (B) methylation between AD and control patients. n = 355, Welch's t‐test. (C) GRN promoter methylation is significantly lower in males compared with females across diagnostics categories. Welch's t‐test. (D) GRN gene body methylation decreases with age. Linear regression. All analysis performed with the UCSF‐MAC cohort. (A‐C) standard boxplot representing median and IQR, whiskers represent 1.5 IQR greater or less than the upper and lower quartile. (D) scatterplot with line of best‐fit. **P < 0.01, n.s. P>0.05.

Discussion

We present here data indicating that patients with sporadic Alzheimer's disease have significantly increased progranulin mRNA in peripheral blood compared with controls. Additionally, we find that progranulin expression is significantly increased in patients with mild cognitive impairment. While we have replicated these findings in an independent patient cohort (AddNeuroMed), we failed to replicate them in the ADNI cohort. Of note, both our cohort and the AddNeuroMed cohort used Illumina microarray platforms while the ADNI gene expression data was run on an Affymetrix platform. We therefore hypothesize that this discrepancy may be a feature of Affymetrix probes, which may not sensitively or reliably detect changes in progranulin expression levels. This is supported by our additional failure to detect any correlation between progranulin expression and age, sex, or genotype in the ADNI dataset.

While we identified differences in mean progranulin expression between groups, our results indicate that progranulin cannot be used on its own as a sensitive or specific biomarker of disease. However, our results suggest that it should be possible to identify further differentially expressed genes that could perhaps then be incorporated into a diagnostic panel. Because peripheral blood can be drawn from living patients in the most basic clinical settings, this has broader clinical utility. Most saliently, this study represents the first association between peripheral progranulin expression and MCI to our knowledge, suggesting that progranulin may play an early role in AD pathogenesis.

Previous studies in mouse models of AD have found that the artificial increase in progranulin levels can slow disease progression including disrupting Aβ plaque deposition and neurotoxicity.14 Furthermore, loss‐of‐function mutations in the progranulin gene are risk factors for developing AD as well as other neurodegenerative disorders.4, 6, 8 Together these findings suggest that progranulin plays a primarily neuroprotective role, reactively modifying or guarding against neurodegenerative processes. Our data support this hypothesis. The small subset of patients with the rs5848 (T:T) haplotype had significantly lower progranulin levels and were also more likely to have AD, as expected.7, 9 However, AD patients in general had higher progranulin levels, especially when excluding haplotypes that directly lower progranulin expression and cause disease. This suggests that progranulin expression might increase as an effect rather than a cause of disease. It is also conceivable that undiscovered cis or trans eQTLs that control progranulin expression exist, and therefore modify risk for AD. For example, we identify the APOE haplotype as a possible trans eQTL that influences progranulin expression, independent of AD diagnosis.

If progranulin plays a functional role in AD pathogenesis, we would expect to ultimately observe changes in protein levels. However, analysis of our ELISA data failed to show a difference in progranulin protein levels between AD, MCI, and control patients, consistent with earlier reports.16 There was also no correlation between progranulin protein levels in plasma and CSF and no correlation between protein and gene expression.39 Our ELISA findings are entirely consistent with and replicate previous independent reports, indicating that our results are not the result of operational error or a technical artifact. One likely interpretation is that the current ELISA methodology fails to accurately detect subtle changes in plasma progranulin levels. The progranulin protein has complex posttranslational regulation, including glycosylation and variable cleavage into a variety of intermediaries as well as any of 8 final granulin products. These various protein configurations have myriad and often contradictory functions.41 In fact, the current ELISA used is specific for unprocessed progranulin.30 Given high antibody specificity and the diversity of progranulin end products, it is unlikely that ELISA can sensitively detect overall changes in specific forms of progranulin being expressed in AD, MCI, and controls; rather it is likely reflective of a particular subset of progranulin products.

Additionally, progranulin and granulins undergo complex spatial regulation with progranulin being shuttled to both the lysosome or excreted into the extracellular space. As such, the biological fluid and cellular components assayed have a large impact on progranulin measurements.42 This might explain the sex dimorphism discrepancy between our gene expression and ELISA data in addition to differences in progranulin cleavage and processing. ELISA, which measures extracellular fluid (plasma), may not be directly comparable to intracellular gene expression profiles. Nevertheless, the functional significance of possibly increased progranulin secretion in females versus relative intracellular retention in males requires further study.

Finally, our data suggest that blood progranulin gene expression is higher overall in males than in females. Our GRN methylation data also supports this finding, as males have reduced promoter methylation, suggesting a derepressed state primed for transcriptional activation. These data are intriguing considering that progranulin expression in rat hippocampus was found to be under the control of estrogen.43 Although there have been numerous studies describing sex dimorphisms in gene methylation in peripheral blood, these studies have failed to specifically identify progranulin.44, 45 Nevertheless, it is especially interesting to consider this sex dimorphism as women are twice as likely to develop AD as men. If progranulin indeed plays a protective role, increased endogenous progranulin expression in males might partly explain this phenomenon, and this might be mechanistically mediated by differences in progranulin promoter methylation.

Author Contributions

Study design and conception: YC and GC. Patient recruitment and trial design: AK, AB, and BM. Sample and data acquisition: AB, BM, DD, YZ, GS, BY. Data analysis: YC, DN, and GC. Manuscript preparation: YC and GC. All authors provided critical feedback.

Conflicts of Interest

Ginette Serrero and Binbin Yue are employees of A&G Pharmaceutical Inc., which has a proprietary PGRN antibody used for the ELISA portion of this work. Ginette Serrero holds issued patents related to the measurement of Progranulin in biological fluids. No other authors report conflicts of interest or financial stakes in this work.

Funding Statement

This work was funded by NIH grants RC1 AG035610, R01 AG26938, P50 AG023501, P01 AG019724, and U01 AG024904; John Douglas French Alzheimer's Foundation grant ; Tau Consortium grant ; National Cell Repository for Alzheimer's Disease (NCRAD) grant ; National Institute on Aging (NIA) grant U24 AG21886; NINDS Informatics Center for Neurogenetics and Neurogenomics grant P30 NS062691; Common Fund of the Office of the Director of the National Institutes of Health grant ; NCI grant ; NHGRI grant ; NHLBI grant ; NIDA grant ; NIMH grant ; DOD ADNI grant W81XWH‐12‐2‐0012.