Elucidating Molecular Phenotypes Caused by the SORL1 Alzheimer’s Disease Genetic Risk Factor Using Human Induced Pluripotent Stem Cells

Basal Levels of SORL1 Expression Are Variable in Human NSCs and Neurons Derived from hIPSCs

Reduced SORL1 expression has been reported in SAD patient lymphoblasts and cerebrospinal fluid and in post-mortem analyses (Dodson et al., 2006; Ma et al., 2009; Scherzer et al., 2004). We had access to 22 fibroblast samples (10 non-demented controls [NDCs] and 12 probable SAD cases clinically diagnosed by National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) (McKhann et al., 1984)) from the UCSD ADRC (Figure 1A and Table 1). Our laboratory has also reprogrammed fibroblasts derived from J. Craig Venter (J.C.V.) (Gore et al., 2011; Woodruff et al., 2013), whose diploid genome sequence has been published and is known to carry genomic variants in SORL1 associated with SAD (Levy et al., 2007). Cellular phenotypes of genomic risk variants may be difficult to detect in post-mortem tissue, which is often a mix of cell types taken at the end stage of a diseased brain. We hypothesized that a culture of purified human neurons containing the endogenous genome of an individual might provide a more sensitive system to detect potentially subtle phenotypes. Therefore, to ask whether SORL1 genomic variants affect basal expression at levels detectable in hIPSC-derived neurons, we first genotyped our small fibroblast cohort for variants in SORL1 documented previously to be associated with risk (R) or protection (P) from AD (Rogaeva et al., 2007). Next we reprogrammed a random subset of these fibroblasts, established hIPSC lines, differentiated neural stem cell (NSC) lines, and purified neurons from these individuals following standard protocols (Israel et al., 2012; Yuan et al., 2011; Table 1; Figure S1A). All of the hIPSC lines used in this study are euploid and can differentiate into the three germ layers: endoderm, mesoderm, and ectoderm (Table S1). Human neurons differentiated and purified using this protocol are a mix of glutamatergic, GABAergic, and cholinergic subtypes and are electrophysiologically active (Israel et al., 2012).

Two common haplotypes in the SORL1 locus have been associated with SAD risk in multiple studies, one defined by a set of SNPs in strong linkage disequilibrium (LD) at the 5′ end of the gene and one defined by a set of SNPs in LD at the 3′ end of the gene (Rogaeva et al., 2007). Multiple studies have confirmed a strong linkage disequilibrium in these regions using various population datasets (Bettens et al., 2008; Rogaeva et al., 2007; Wen et al., 2013). We genotyped our fibroblast lines for eight representative SORL1 variants, four SNPs in the 5′ region and four SNPs in the 3′ region. For clarity, we utilized the names of these SNPs as assigned by Rogaeva et al. (2007): SNPs 4, 8, 9, and 10 at the 5′ end and SNPs 22, 23, 24, and 25 at the 3′ end of SORL1 (Figure 1B; Figure S1). We also queried publicly available data from phase I of the 1000 Genomes Project (1000 Genomes Project Consortium et al., 2012) focusing on the Centre d’Etude du Polymorphisme Humain population, which closely resembles the population of fibroblast donors from the ADRC (northern and western European populations) to confirm the extent of disequilibrium at these sites (Figure 1B). Three of the SNPs at the 5′ end of SORL1 (SNPs 8, 9, and 10) in our 23 fibroblast samples showed apparent complete LD (the SNPs were found together in all unambiguous homozygous samples, and the estimated pairwise R² was 1.0; Figure 1C; Figure S1C). The same pattern of LD was observed in our analysis of the 1000 Genomes data (pairwise R² = 1.0; Figure 1B). LD was not as strong at the 3′ end of SORL1 in our sample (Figure 1D; Figure S1C) or in the public data, where the pairwise R² ranged from 0.64–0.88 (Figure 1B). There was little association between the 3′ haplotypes and cellular phenotypes in preliminary experiments (Figure S3), and, therefore, we focused our attention on the 5′ haplotype. We designated the haplotype at the 5′ end comprised by SNPs 8, 9, and 10 as either R (risk) or P (protective) based on previous association studies (Reitz et al., 2011; Rogaeva et al., 2007) and classified our patient samples as P/P, R/P, or R/R. That is, R/R patients are homozygous for the variants at SNPs 8, 9, and 10, which have been reported to be associated with an increased risk of developing SAD; P/P patients are homozygous for alleles at SNPs 8, 9, and 10, associated with a reduced SAD risk; and R/P patients are heterozygous for risk and protective SNPs. Interestingly, previous work using postmortem brain tissue has implicated the 5′ region of SORL1 in alternative splicing and reduced SORL1 expression in the temporal cortex, although the functional variants have not been identified (Grear et al., 2009; McCarthy et al., 2012). Therefore, we hypothesized that hIPSC-derived neurons with risk haplotypes in this 5′ region may show differential SORL1 expression.

We analyzed the levels of SORL1 mRNA in NSC lines and neurons differentiated and purified from NSC lines (six NDC genomes, including J.C.V., and seven SAD genomes). At the basal level, we found a highly variable expression of SORL1 in all cell types and genotypes (P/P, R/P, and R/R) with no detectable differences in SORL1 expression because of disease state (NDC versus AD) or SORL1 haplotype (Figures 2A–2D). We also analyzed the parent fibroblast lines for basal levels of SORL1 gene expression and, similarly, found no differences between genotypes or disease states in these cells (Figures S2A and S2B) We did see a significant increase in SORL1 expression upon differentiation to NSCs and neurons (Figure S2C), consistent with previous reports showing that SORL1 is highly expressed in the CNS (Mörwald et al., 1997; Motoi et al., 1999).

A SORL1 Haplotype Associated with Protection from AD May Regulate the Induction of SORL1 Expression by Certain Growth and Neurotrophic Factors

Our neuronal differentiation protocol contains a mix of growth and neurotrophic factors (Yuan et al., 2011) including brain-derived neurotrophic factor (BDNF), which can induce the murine homolog of SORL1 (LR11) (Rohe et al., 2009), and cyclic AMP (cAMP), which can be elevated by BDNF treatment (Cheng et al., 2011). BDNF has been hypothesized to be protective in various animal models of AD, and loss of BDNF has been implicated in AD (Nagahara et al., 2009). Treatments that increase intracellular cAMP levels, such as liraglutide, have also been shown to be protective in AD models (Hunter and Hölscher, 2012; Parthsarathy and Hölscher, 2013). Therefore, we tested the acute effects of these two factors, BDNF and cAMP, on SORL1 expression in purified human neurons. Although several growth and neurotrophic factors were present in our neuronal media (Israel et al., 2012; Yuan et al., 2011), we induced SORL1 expression above the baseline in our system by acutely treating neurons with additional, exogenous BDNF or cAMP for 24 hr.

We measured SORL1 expression induction by quantitative RT-PCR (qRT-PCR) and found that, averaging across neurons derived from our 13 cell lines, the absolute relative expression level for BDNF-treated neurons was not significantly different from that of untreated neurons (Figure 3A). However, we consistently noticed that neurons from 9 of the 13 patients exhibited a significant induction of SORL1 expression above the baseline in response to BDNF, whereas neurons from 4 of the 13 patients did not (Figure 3B; Table 2). When we examined the cell line response to BDNF by genotype, we found that all four of the lines in which SORL1 was not induced by BDNF were homozygous R/R, whereas all lines with a P allele did respond to BDNF treatment (Figure 3C, post hoc Fisher’s exact test, p = 0.001; Table 2). This suggests that the P haplotype is a marker for a dominant protective genetic variant. Available SAD case control data are highly consistent with a dominant protective variant (Table S2). A pooled analysis of public genotype data of 1,640 SAD cases and 1,728 controls with a similar ethnic background as our series suggests a consistent protective effect of the R/P and P/P haplotypes relative to the R/R haplotype, with an odds ratio of 0.77 (95% confidence interval [CI], 0.66–0.90) for R/P versus R/R haplotypes, an identical odds ratio (OR) of 0.77 (95% CI, 0.63–0.93) for P/P versus R/R haplotypes, and an OR of 0.99 (95% CI, 0.82–1.20) for R/P versus P/P haplotypes. The sample size was not sufficient to formally compare SORL1 expression in response to BDNF across all three haplotypes in our cell lines. However, induction in R/P cell lines was similar to induction in P/P cell lines, and mean induction in the pooled P/P and R/P cell lines was significantly higher than that observed in the R/R cell lines (Figure 3C).

Conversely, the BDNF inducibility phenotype did not correlate with the presence of putative risk variants at SNP4 or SNPs 22–25 at the 3′ end of SORL1 (Figures S3A and S3B). Two different SORL1 primer sets and two different housekeeping genes confirmed our mRNA expression results (Figure S3C). Because of the small amount of material we can routinely obtain from purified neurons, we characterized this phenotype primarily at the mRNA level. However, we analyzed SORL1 protein levels from representative fibroblast, NSC, and purified neuron genotypes and found that SORL1 protein induction from fibroblast to NSC was blunted in R/R cells compared with P/P and R/P cells, and, in purified neurons treated with BDNF, there was no apparent increase in SORL1 protein upon treatment in R/R cells (Figures S3D and S3E).

Interestingly, we also found that cAMP strongly induced SORL1 expression in neurons, and the average absolute relative expression level increase for cAMP-treated neurons was significant (Figure 3D). When each individual was examined, we found that cAMP treatment induced SORL1 expression above the baseline from most, but not all, cell lines (Figure 3E; Table 2).

When we examined cell lines treated with cAMP, we found that two R/R cell lines did not respond to treatment, whereas two R/R cell lines did (Figure 3E). cAMP treatment consistently and robustly induced SORL1 expression in all other cell lines, suggesting heterogeneity in this signaling pathway (Figure 3F; Table 2). We did not find a significant difference in induction of SORL1 expression between the P/P, R/P, or P/P genotypes after cAMP treatment (Figure 3F; Table 2).

To rule out the possibility that this induction phenotype in response to BDNF may be due to decreased pluripotency or differentiation efficiency, we compared Tra 1–81 and Tra 1–60 in hIPSC and CD24 labeling after neuronal differentiation and detected no differences in R/R cells (Figures S4A–S4C). Purified neurons of all genotypes showed significantly increased expression of Tau mRNA and exhibited an obvious neuronal morphology compared with NSC lines (Figures S1A and S4D). Therefore, we concluded that there are no differences in pluripotency, differentiation capability, or apparent neuronal phenotype in cells of different SORL1 genotypes. We also tested purified neurons for expression of the BDNF receptor TrkB and found no differences in TrkB expression in R/R cells (Figures S4E and S4F). Taken together, our data suggest that, rather than controlling basal expression levels, naturally occurring variants in humans in the 5′ region of SORL1 modulate expression induction in response to BDNF exposure.

Induction of SORL1 Expression Correlates with Decreased Aβ Peptide Production

APP undergoes two cleavage events, by β and γ secretases, to generate the potentially neurotoxic Aβ peptide and other fragments. Previous work suggests that SORL1 serves a protective function by binding full-length APP and either sorting it away from amyloidogenic β and γ secretase cleavage pathways or masking the secretase binding site (Andersen et al., 2005; Hermey, 2009; Schmidt et al., 2012). In particular, reducing SORL1 expression in mice and non-neuronal cell lines increases the production of Aβ peptides, whereas induction of SORL1 expression lowers the production of Aβ (Dodson et al., 2008; Rogaeva et al., 2007; Rohe et al., 2009). Because the amount of Aβ peptide produced in the human brain is an important SAD phenotype that is potentially related to the cause or progression of SAD, we tested whether induction of SORL1 with BDNF or cAMP in purified human neurons had an effect on Aβ production and secretion. Measurement of the average absolute levels of Aβ_1–40 peptides in all cell lines revealed no significant differences between cells treated with BDNF or untreated cells (Figure 4A). However, a clear trend emerged when we evaluated the fold reduction in Aβ_1–40 peptides in each individual line in response to BDNF treatment. The neurons from the four hIPSC lines carrying the R/R genomes that did not induce SORL1 expression also did not significantly reduce Aβ peptides in response to BDNF treatment, whereas all neurons from hIPSC lines carrying R/P or P/P genomes significantly reduced Aβ levels, with the exception of one line, SAD4 (Figure 4B; Table 2). When analyzed together, the reduction of Aβ_1–40 in all genomes with a P allele versus R/R genomes was significant (Figure 4C). As published previously (Israel et al., 2012), in some cell lines Aβ_1–42 is below the detection range of our assay because, in our system, endogenous levels of this analyte are approximately five times lower than Aβ_1–40 in neuronal media. However, for some lines where we were able to consistently detect Aβ_1–42 peptides, we observed a similar trend as with Aβ_1–40 (Figure S5A). Although the decrease in Aβ production induced by BDNF in P/P and R/P genomes is apparently modest (20%–30%), it is in line with reasonable expectations for changes that might influence risk but not direct induction of disease. For example, aggressive, early-onset FAD can be caused by a highly penetrant APP duplication that increases APP expression and Aβ production by 50% (Israel et al., 2012; Rovelet-Lecrux et al., 2006; Shi et al., 2012). Therefore, the average 20% change in APP processing we observed in these in vitro experiments is significant in the context of SAD risk.

In contrast to BDNF, cAMP treatment significantly and robustly decreased absolute Aβ levels by an average of 50%–60% in purified neurons of all genotypes (Figures 4D and 4E; Table 2). This decrease was similar among all genotypes tested, and there was no significant difference in Aβ reduction between P/P and R/P compared with R/R genomes (Figures 4E and 4F; Table 2). Taken together with our expression data, these results suggest that SAD risk associated with SORL1 genomic variants may be generated via specific signaling pathways within neurons and that the impact of SORL1 induction may vary in different genomic backgrounds.

Decreased Aβ Peptides Caused by BDNF, but Not cAMP, Requires SORL1

To determine whether the reduction of Aβ in response to BDNF or cAMP treatment requires SORL1 expression, we designed an shRNA against the human SORL1 mRNA and documented a robust and consistent knockdown of SORL1 mRNA and protein in transduced purified neurons of all genotypes (Figures S5B and S5C). We then treated transduced neurons with BDNF and observed that the decrease in Aβ peptides in neurons with P haplotypes at the SORL1 locus was almost completely abrogated in neurons expressing the SORL1 small hairpin RNA (shRNA), whereas knockdown of SORL1 had little effect in cells with R/R genotypes (Figure 5A; Table 2), suggesting that the decrease in APP processing after BDNF treatment is dependent on SORL1. In contrast, SORL1 knockdown neurons treated with cAMP still exhibited a reduction of Aβ peptides (Figure 5B), indicating that, in purified neurons, cAMP regulates pathways in addition to SORL1 that can impact Aβ production independently.

Overexpression of a SORL1 cDNA in BDNF Non-Responsive Cells Reduces Aβ Peptides

Our data corroborate most other work in cell and animal models, demonstrating that increased SORL1 expression reduces Aβ peptides and raises the possibility of targeting this pathway for new AD therapies (Caglayan et al., 2014). However, because SORL1 genetic risk variants may affect pathways that regulate SORL1 expression, individual patient genotypes may respond differently to various agents designed to induce SORL1. To test whether SORL1 activation can be beneficial in neurons that do not induce expression after BDNF treatment, we stably overexpressed a SORL1 cDNA in R/R neurons (derived from J.C.V. hIPSCs). We observed that, after 5 days in culture, the R/R neurons from J.C.V. containing the SORL1 cDNA showed a 4-fold increase in SORL1 mRNA, approximately the same fold induction as the highest cell line induced with cAMP (Figure 5C). These SORL1-overexpressing neurons had significantly reduced levels of both Aβ40 and 42 peptides when the cells were harvested after 5 days in culture (Figure 5D).

hIPSC, NSCs, and Neuron Generation and Characterization

All hIPSC lines were generated by four-factor reprogramming (Gore et al., 2011; Israel et al., 2012). hIPSC lines were maintained on a mouse embryonic fibroblast (MEF) feeder layer in HUES medium (Israel et al., 2012) and were routinely tested for mycoplasma. For RNA/DNA analysis, cells were transferred from MEFs to Matrigel, cultured in MEF-conditioned HUES medium, and maintained for several passages before harvesting. For germ layer differentiation analyses, embryoid bodies (EBs) were generated by growing hIPSC in low-adherence plates for 14 days without fibroblast growth factor (FGF) with a medium change every other day. Previously published hIPSC lines were cytogenetically karyotyped (Cell Line Genetics), whereas unpublished hIPSC lines were karyotyped digitally by hybridization to the Illumina Infinium Human-CoreExomeBeadChip module version 1.9.4. Pluripotency gene expression in published lines has been described previously (Gore et al., 2011; Israel et al., 2012). Unpublished lines were analyzed for pluripotency by fluorescence-activated cell sorting (FACS) for the cell surface antigens Tra 1–81 and Tra 1–60 (BD Biosciences). The differentiation capability of published lines has been described previously (Israel et al., 2012). For unpublished lines, RNA from EBs and parent hIPSC lines was harvested and converted to cDNA (Superscript, Invitrogen). qRT-PCR was performed with a primer specific for endoderm (AFP), mesoderm (DCN), and ectoderm/neuroectoderm (COL1A1, NEST). NSCs and neurons were generated using protocols described previously (Israel et al., 2012; Yuan et al., 2011) and purified by FACS. NSCs were stained for Nestin (Millipore), and neurons were stained for MAP2 (Millipore). For each genotype, neurons derived from at least two independent hIPSC lines were examined.

Cell Culture, Experimental Design, and Sample Collection

NSCs were maintained in DMEMF12 medium supplemented with B27, N2, and Fgf (20 ng/ml) and were routinely tested for mycoplasma. NSCs were differentiated to neurons by Fgf withdrawal and cultured in the presence of BDNF, glial cell line-derived neurotrophic factor, and dibutyryl cAMP (dbcAMP) for 3 weeks as described previously (Israel et al., 2012; Yuan et al., 2011). Purified neurons were generated by FACS using a published cell surface signature (Yuan et al., 2011). Neurons were plated in biological replicates treated as described in the figure legends and harvested for RNA or protein. All qRT-PCR samples were run in triplicate. SORL1 expression experiments and analyses in purified neurons are reported in the figure legends in terms of the 13 individual patient genomes analyzed. However, for each genotype (P/P, R/P, or R/R), neurons derived from at least two independent hIPSC lines were examined. SORL1 expression experiments and analyses were repeated in all genotypes for a minimum of two independent differentiations and, for some genotype representative cell lines (SAD1 [P/P], NDC1 [R/P], and CV [R/R]), four or more times.

SORL1 SNP Genotyping

Genotyped SNPs were identified from Rogaeva et al. (2007) and are listed in Figure 1.

Fibroblast cultures from each patient were lysed, and genomic DNA was isolated using the DNeasy kit (QIAGEN). Allele-specific primers for the SNPs analyzed were designed using the Web-based allele-specific primer-designing tool (http://bioinfo.biotec.or.th/WASP). PCR were run using Hot-Star Polymerase (QIAGEN) and analyzed by agarose gel electrophoresis. A list of all allele-specific primers used in this study is provided in the Supplemental Experimental Procedures.

Expression Analyses

SORL1 expression was analyzed by measuring mRNA and protein. For mRNA, total RNA from triplicate cell culture experiments was isolated using the RNeasy kit (QIAGEN) and DNase-treated (Ambion). cDNA was transcribed from total RNA using the Superscript first-strand system (Invitrogen). qRT-PCR was performed using FastStart SyBr Green (Roche) and samples were run on an Applied Biosystems 7300 real-time PCR system. RT-PCR primers for all genes in this study were designed using Primer3 and spanned intron-exon boundaries. A list of all primers is provided in the Supplemental Experimental Procedures. qRT-PCR data were analyzed using the ΔΔCt method and normalized to housekeeping genes TBP or RPL27, as noted in the figure legends. For protein, cells were lysed in 1% NP40 or radioimmunoprecipitation assay (RIPA) buffer. Protein concentration was determined using either Bradford (Bio-Rad) or bicinchoninic acid (BCA) (Pierce) assays. Protein lysates were run on NuPAGE 4%–12% Bis-Tris gels (Invitrogen), transferred to nitrocellulose membranes, and probed with antibodies to either SORL1 (BD Biosciences, clone 48/LR11) or Actin (Millipore). Quantification of the western blots was performed using ImageJ software.

BDNF and cAMP Treatment

For purified neurons, triplicate cell culture experiments were treated with BDNF (PeproTech, 50 ng/ml) or dbcAMP (Sigma, 250 μg/ml) for 24 hr. After 24 hr, cells were harvested for RNA using the RNeasy kit (QIAGEN) or protein using either 1% NP40 or RIPA buffer. Data represent at least two independent differentiation experiments.

Aβ Measurements

Purified neurons were seeded at 200,000 cells/well of a 96-well plate. After 5 days in culture, medium was harvested from triplicate wells and run on an Aβ Triplex ELISA plate (Meso Scale Discovery). Corresponding wells were lysed with RIPA buffer, and protein concentration was determined by BCA assay (Pierce). Aβ levels were normalized to the amount of total protein.

Statistics

Based on examination of distribution assumptions, t tests were used for within-cell line comparisons of expression levels before and after neurotrophin treatment, but otherwise all analyses were performed using non-parametric statistical tests. These tests are more conservative and less powerful that parametric ANOVA and t tests but are robust to normality assumptions and, therefore, not prone to spurious type I errors. Between-genotype differences in continuous variables were analyzed by Kruskal-Wallis test, as stated in the figure legends. Two group comparisons of pooled P/P and R/P cells lines to R/R cell lines were tested using non-parametric Mann-Whitney tests, as stated in the figure legends. The distribution of categorical outcomes across genotype groups was compared using Fisher’s exact test. All testing was two-sided and was performed using the R statistical programming language (R Development Core Team, 2008). All statistical results were verified independently using the VassarStats computational website (http://vassarstats.net/).