Self-Organizing 3D Human Neural Tissue Derived from Induced Pluripotent Stem Cells Recapitulate Alzheimer’s Disease Phenotypes

The dismal success rate of clinical trials for Alzheimer’s disease (AD) motivates us to develop model systems of AD pathology that have higher predictive validity. The advent of induced pluripotent stem cells (iPSCs) allows us to model pathology and study disease mechanisms directly in human neural cells from healthy individual as well as AD patients. However, two-dimensional culture systems do not recapitulate the complexity of neural tissue, and phenotypes such as extracellular protein aggregation are difficult to observe. We report brain organoids that use pluripotent stem cells derived from AD patients and recapitulate AD-like pathologies such as amyloid aggregation, hyperphosphorylated tau protein, and endosome abnormalities. These pathologies are observed in an age-dependent manner in organoids derived from multiple familial AD (fAD) patients harboring amyloid precursor protein (APP) duplication or presenilin1 (PSEN1) mutation, compared to controls. The incidence of AD pathology was consistent amongst several fAD lines, which carried different mutations. Although these are complex assemblies of neural tissue, they are also highly amenable to experimental manipulation. We find that treatment of patient-derived organoids with β- and γ-secretase inhibitors significantly reduces amyloid and tau pathology. Moreover, these results show the potential of this model system to greatly increase the translatability of pre-clinical drug discovery in AD.


Generation of organoids and analysis
To investigate the utility of such a system, we created scaffold free three-dimensional (3D) human neural organoids, from human iPSCs derived from AD patients and healthy controls and test each line for pluripotency (S1 Fig). Several protocols have been developed to create neural organoids from human pluripotent stem cells [53,[64][65][66][67]. We followed the protocol published by Kadoshima et al. [64] with minor modifications (see Methods) and successfully created complex dense 3D neural tissues from a number of human iPSC lines from AD patients and healthy control (S2A Fig and S1 Table). We subjected the 3D cultures, heretofore referred to as organoids, to immunohistochemistry with antibodies against a neuronal protein (MAP2) as well as SOX2, a marker for neural progenitor cells, to determine the presence of neural cell types. After one month of culture, we observed the emergence of translucent regions of neuroectoderm and immunolabeling for neuronal MAP2 as well as SOX2-positive neural progenitor cells, as previously described [64]  Since we are interested in age-related neural pathology, we found that the large regions of neuron-rich tissue produced by the modified Kadoshima et al protocol [64] best suited our modeling needs.
At 60d and 90d of culture, we subjected organoids from a healthy control (Ctrl) and a familial (fAD) patient (APP duplication, APP Dp 1-1, reported in Israel et al., 2012) to a number of assays for AD-relevant phenotypes. We also examined additional fAD lines in (S1 Table). As the organoid continues to increase in size, the deeper regions of the tissue show evidence of necrosis, which is likely due to the absence of vasculature and lack of nutrients and oxygen penetration in the deeper layers. As the presence of necrosis could confound our measurements of AD-like phenotypes. We used immunohistochemistry for cleaved caspase 3 (CC3; S3A Fig) as an indicator of apoptotic cells and calculated the average distance between the surface and regions of increasing CC3 immunoreactivity. We found that the region of neuron-rich, CC3-sparse tissue extended an average of 250 μm from the surface into the interior of the organoid (S3A Fig). In addition, we incubated sections with secondary antibody alone (after blocking) to assess the presence of nonspecific binding. We observed that the inner tissue region also gave rise to nonspecific antibody binding compared to the more superficial region (S3B Fig). Based on these data, we established a limit of 250 μm from the surface of the organoid for our further characterization.

Alzheimer's disease phenotypes in organoids
Amyloid beta. Neurons derived from the APP Dp 1-1 line had previously exhibited increased levels of secreted Aβ 40 and Aβ 42 (Israel et al., 2012). To determine whether this phenotype was perpetuated in 3D culture, we subjected culture media from fAD and control organoids to ELISA. In agreement with the previous study, we detected significantly higher levels of Aβ in the media from fAD organoids culture compared to controls (Fig 1A). We performed immunohistochemistry on fixed cryosections from control and fAD organoids to examine AD-like pathology using two different antibodies that recognize Aβ. The first antibody, 4G8 (immunoreactive against amino acid residues 17-24 of Aβ) has been widely used to label both soluble and aggregated Aβ [68]. In addition, we co-labeled sections with the anti-Aβ antibody   and Aβ 1-42 from supernatant of control (Ctrl; CS-0020-01) and familial AD (fAD; APP Dp 1-1) organoid cultures, measured by ELISA, as well as the ratio of Aβ 1-42 to Aβ 1-40 concentrations. Unpaired two-tailed t-test with equal variance: *p = 0.047 (Aβ 1-40 ), unpaired twotailed t-test with Welch's correction for unequal variance: **p = 0.004 (Aβ 1-42 ), p = 0.48 (Aβ 1-42 /Aβ 1-40 ). (B) Tissue sections from fAD (APP Dp 1-1) and control (Ctrl; CS-0020-01) organoids were processed for immunoreactivity against amyloid β (Aβ) using two antibodies (D54D2: white, 4G8: green), as well as antibodies against the neuronal marker MAP2 (red) and stained with the nuclear dye Hoechst (blue). Insets demonstrate Aβ immunoreactivity that appears both extracellular (i, arrow) and intracellular (ii, arrowhead) based upon MAP2 colocalization. (C) Z-projection of immunolabeled tissue sections from 90 day old Ctrl and fAD organoids showing immunoreactivity for Aβ D54D2, which recognizes several isoforms of amyloid (Aβ 37 , Aβ 38 , Aβ 39 , Aβ 40 , and Aβ 42 ). In 90d organoids, we detected aggregates that were immunopositive for both Aβ antibodies ( Fig  1B), while 4G8 immunoreactivity appeared against intracellular amyloid that co-localizes with MAP2 as well as putative extracellular aggregates (Fig 1B, S4A Fig and S1 and S2 Movies). We then used the D54D2 antibody to quantify the size and number of Aβ aggregates. The 90d fAD organoids contained numerous Aβ aggregates compared to the control organoid tissue ( Fig  1C). Analysis at 60d and 90d of culture identified a progressive increase in number and size of Aβ aggregates in the fAD organoids compared to controls (Fig 1D and S4B Fig). Additionally, we performed Western blot on the 90d whole organoids lysates and observed increased levels of Aβ oligmers in fAD organoids compared to controls (data not shown). These data demonstrate the presence of robust and spontaneous Aβ aggregation in AD patient-derived organoid culture models that appears to develop in a time-dependent manner.
Hyperphosphorylated Tau (pTau). Another hallmark of AD is the abnormal phosphorylation, mislocalization, and aggregation of the tau protein. Beta-plated sheets of hyperphosphorylated tau (pTau) leads to the disruption of neuronal microtubule assemblies and neurodegeneration [19]. To examine control and fAD organoids for the presence of tau pathology (tauopathy), we conducted immunohistochemistry in organoid sections using antibodies against pTau (Ser396 or Thr181). At 90d, the fAD organoids exhibit significantly greater pTau immunoreactivity than did control sections ( Fig 1E and 1F and S4C Fig). Interestingly, there was no significant difference in pTau levels between fAD and control organoids at 60 days ( Fig  1F). This is in contrast to the amyloid phenotype, in which more Aβ aggregates were observed at both 60d and 90d in the fAD organoids ( Fig 1D). The Thioflavin-S dye binds aggregates of β-pleated sheets and is used as an indicator of tau pathology in human brain and mouse models [69]. We observed a greater total area covered by Thioflavin-S dye labeling in fAD organoids compared to controls, as well as a higher number of Thioflavin-S positive particles (S4D and S4E Fig). Together, these various measures demonstrate that the AD patient-derived organoids have the power to recapitulate both amyloid and pTau phenotypes.

Organoids from multiple AD lines recapitulate AD pathology
To ascertain whether the AD phenotypes of Aβ aggregation and Tau hyperphosphorylation are generalizable to fAD organoids from different sources, rather than being a phenomenon of the APP Dp 1-1 line, we examined additional organoids created from patient-derived and healthy control iPSCs (S1 Table). At 90d of culture, we processed and analyzed organoids created from an additional APP duplication line (APP Dp Table). Organoids derived from fAD patients with APP duplication (APP Dp 2-3) or the PSEN1 A264E mutation exhibited increased numbers of Aβ aggregates verses the control lines ( Fig 2B). While the fAD organoids from PSEN1 M146I patient cells exhibited a trend towards higher amyloid levels, this was not significant. We also examined organoids from these additional fAD lines for the presence of pTau immunoreactivity. Similar to the pattern observed with amyloid aggregation, the APP Dp 2-3 and the PSEN1 A264E fAD organoids exhibited increased pTau (Ser396) immunoreactivity at 90 days of culture, while the PSEN1 M146I organoids did not differ from control ( Fig 2C).

Attenuation of AD pathology in neural organoids by β-and γ-secretase inhibitor treatment
These data indicate that organoids created from multiple fAD patient iPSC lines demonstrate robust and relevant AD-like phenotypes. To determine whether these phenotypes are indeed a result of altered Aβ production, we treated the fAD organoids with two compounds well known to reduce amyloid aggregation: the γ-secretase inhibitor Compound E (Comp-E; γ2) and a BACE-1 β-secretase inhibitor (β-Secretase Inhibitor IV, EMD Millipore) [40,74]. These experiments also allow us to determine the feasibility of pharmacologic manipulation in the organoid cultures. Comp-E and β-Secretase Inhibitor IV, or DMSO vehicle, were added to the culture media of 30-day-old fAD organoids and replenished with every media change (Fig 3A). At day 60 (30 days of treatment) we assessed Aβ and pTau pathology ( Fig 3A) and showed that compound treatment significantly reduced the number of amyloid aggregates in the fAD organoids in dose dependent manner, compared to vehicle treated fAD tissue ( Fig 3B). This reduction was particularly evident in the in 90-day-old fAD organoids treated for 60-days ( Fig 3C). In contrast to the dramatic reduction of Aβ aggregation with compound treatment, immunoreactivity for pTau was unaffected following 30 days of treatment compared to vehicle-treated controls ( Fig 3D). However, following 60 days of treatment, the 90 day-old organoids showed significantly less pTau immunoreactivity compared to vehicle-treated controls in dose dependent manner (Fig 3D). These results indicate that specific AD-like phenotypes observed in patient iPSC-derived organoids can be ameliorated with drug treatment, suggesting that this system is amenable to compound testing.

Discussion
In diseases such as AD that are characterized by protein aggregation, the presence of a true interstitial compartment is important for modeling pathology. Previous three-dimensional (3D) tissue engineering approaches have embedded neural progenitors or cell types of interest in a matrix or a scaffold [40,58]. While these ingenious approaches can model AD phenotypes, they do not recapitulate spontaneous pathology resulting from endogenous cellular characteristics, but rather necessitate the overexpression of fAD genes. In the current work, we took advantage of scaffold-free 3D tissue culture protocols to create neural organoids using iPSCs from fAD patients and healthy controls. The dense nature of these cultures likely facilitates protein aggregation, while remaining amenable to experimental manipulation such as compound treatment.
The ease with which large numbers of these organoids can be created, and their ability to respond to compound treatment, open the possibility for phenotypic and mechanistic compound screening in complex human tissue models of multiple neurodegenerative diseases. With iPSC technology, we can access patient-derived cells carrying AD-associated variants in a number of genes, which are difficult to model as many are non-coding and/or associated with more than one significant single-nucleotide polymorphism (SNP) [75]. Moreover, rapid improvements in gene editing technology allows researchers to create isogenic iPSC and ESC lines that carry disease-associated coding variants, both well-known (such as in the APOE gene, [76,77] or emerging (i.e. SORL1 [50,78], TREM2, [79]). Together, these advancements allow us to conceive of systems in which we can test candidate therapies on complex neural tissue systems targeted to defined subpopulations of late-onset or sporadic AD patients.
One powerful aspect of the current model is the spontaneous appearance of both [1] amyloid and tau pathology, and the distinct timeline on which these phenotypes appear. Modeling both these facets of AD has been challenging in mouse models. Most mouse AD models must carry multiple transgenes to achieve robust amyloid phenotypes, and rarely have significant tau pathology or neuronal loss (for review see [31]). Mouse models of tauopathy overexpress mutated human MAPT (which is causal for frontotemporal dementia, not AD) to induce tau pathology [80,81], and do not exhibit amyloid aggregation. Using the organoid model, we observed that amyloid pathology emerges prior to significant tau hyperphosphorylation in neural tissue derived from fAD patients carrying a duplication of the APP gene. While the sequential emergence of amyloid and tau pathology in human AD remains somewhat controversial [82][83][84][85], this timing is in close agreement with that observed in the scaffolded 3D cultures of Choi et al. 2014; [40]. Additionally, the inhibition of Aβ production using β and γsecretase inhibitors reduced tau hyperphosphorylation only at the later time point of treatment, after Aβ reduction was observed. Thus, AD-relevant phenotypes of Aβ accumulation emerge prior to tauopathy in this model. Moreover, the reductions in Aβ accumulation that occur from the inhibition of APP processing lead to a dose-and time-dependent amelioration of tauopathy in the fAD organoids, suggesting a causal relationship between these relevant pathologies in the neural organoid model.
In the current work, we focused our efforts on determining whether or not these organoid systems could model age-related AD-like pathology. The pioneering works that we drew our techniques from used the organoid system as a means to study neurodevelopment [53,64,66,67]. Since work by several groups has suggested that iPSC reprogramming "re-sets" the epigenome, and that other phenotypes associated with cellular aging, such as mitochondrial function and telomere length, are returned to a "juvenile-like" state [86,87], the obvious question is: to what extent can we model phenotypes associated with aging in human neural cells? While we observe robust AD-like phenotypes that increase with "age" in the organoids, the extent to which the organoid tissue represents the aged human brain has not been examined. We believe that the scaffold-free three dimensional model has good potential for studying neurological diseases. This will be important for future works to use this model system to examine other AD related phenotypes, such as neuroinflammation, gliosis, DNA damage, U1 tangles [88] and synaptic dysfunction. Also, without a means of tissue perfusion, the organoid suffers from the same issues as primary slice culture, in that the distance from the culture medium interface is correlated with tissue necrosis. There is currently great interest in the combination of three-dimensional neural culture systems with artificial blood-brain barrier technology [89][90][91], to address this issue.

Maintenance of PSC and 3D culture differentiation
Induced pluripotent stem cells (iPSCs) were created from human fibroblasts (S1 Table). Two of the iPSC lines carrying duplications in the gene for Amyloid Precursor Protein (APP; APP Dp 1-1 and APP Dp 2-3) were provided by Dr. Lawrence Goldstein at the University of California, San Diego, and have been described previously [44]. One control iPSC line (AG09173) was kindly provided by Dr. Bruce A. Yankner at McLean Hospital and Harvard Medical School. The other lines were generated from fibroblasts at the Picower institute of Learning and Memory, iPSC core facility at the Massachusetts Institute of Technology (MIT) using Sendai virus to overexpress OCT4, SOX2, KLF4, and c-MYC. S1 Table. details the sources and attributes of the cells used in this study. Pluripotency was confirmed by immunocytochemistry for TRA-1-81 and TRA-1-60 (S1 Fig). All reagents were purchased from Life Technologies Corporation, Grand Island, NY, unless mentioned otherwise. iPSCs were cultured on irradiated mouse embryonic fibroblasts (MEFs, MTI-GlobalStem, Gaithersburg, MD) in DMEM/F12 media supplemented with knockout serum replacement (KSR, 20% v/v), non-essential amino acids (NEAA-1X), GlutaMAX (1X), beta-Fibroblast Growth Factor (FGF2, PeproTech, Inc, Rocky Hill, NJ) and 2-mercaptoethanol (0.1 mM). The quality of cells was monitored daily and differentiated cells were mechanically removed under a light microscope in a biosafety hood. iPSCs were culture up to 80% confluence and dissociated into single cell suspension after treated with Accutase (diluted in PBS (1.5:1) containing Rock inhibitor (Y-27632 dihydrochloride, Tocris Biosciences, Minneapolis, MN) to improve cell survival. The MEFs and iPSCs were separated by plating the single cell suspension onto 0.1% gelatin (0.1%, EMD Millipore, Billerica, MA) coated dishes for 45 min, after which time the MEFs attach to the substrate and the non-adherent iPSCs were collected.
To create 3D cultures, or neural organoids, we followed a published protocol [64] with some modifications. Embryoid bodies (EBs) were formed by loading 12,000 iPSCs per well into 96-well plates with cone-shaped wells (Nunc 1 96-well Conical Bottom plates, VWR International, Bridgeport, NJ) pre-coated with Pluronic acid (F-127, 1%, Sigma-Aldrich, Natick, MA). The 96 well plates were transferred to an incubator at 37 C°with 95% relative humidity and 5% CO 2 . Media used in EB culture consisted of Glasgow-MEM supplement with KSR (20% v/v), Sodium Pyruvate (1X), NEAA (1x), 2-mercaptoethanol (0.1mM), Rock inhibitor (20 μM), TGFβ-inhibitor (SB431532 compound, Tocris Biosciences, Minneapolis, MN; 5 μM), Wntinhibitor (IWRe1 compound, Tocris Biosciences, Minneapolis, MN; 3 μM). The EBs maintained in this medium for 18-20 days, with Dorsomorphin (BMP signal inhibitor, Tocris Biosciences, Minneapolis, MN; 2 μM) added to the culture for the first three days to promote a neuronal lineage. Media was replaced every other day. The edges of the EBs began to appear translucent around day 10, and the tissue grew to be large than 0.6 mm in diameter by day 18. The cell aggregates, or organoids, were transferred to non-adherent petri-dishes (EZsphere dish) to prevent fusion of separate organoids, and cultured in a medium designed to promote neuroepithelial formation, which consisted of DMEM/F12 supplemented with Chemically Defined Lipid Concentrate (1X) and N2-supplement (1X) and maintained in an incubator with 5% CO 2 and 40% Oxygen. The aggregates were kept in this media for 15 to 20 days, at which time heparin (5 μM, Sigma-Aldrich, Natick, MA), FBS (10% v/v, Gemini Bio-Products, West Sacramento CA), and Matrigel (final 1% v/v, Corning Incorporated-Life Sciences, Oneonta, NY) were added to the medium. On day 70 the amount of Matrigel was increased to 2% and B27 supplement (1X) was added to the medium. The aggregates were maintained in this final medium for the remained of the culture period. Medium in the culture dishes was replaced every 4 to 5 days.

Thioflavin-S staining
All reagents for Thioflavin-S staining were purchased from Sigma-Aldrich, Natick, MA, unless otherwise indicated. Sections were rinsed with PBS for 5 minutes and then incubated in 0.05% potassium permanganate solution for 20 minutes followed by two washes in PBS and de-staining in 0.2% potassium metabisulfite and 0.2% oxalic acid (until the brownish color from the potassium permanganate is removed, less than 1 minute). The potassium metabisulfite/oxalic acid was washed out using PBS and sections were incubated in freshly-prepared 0.02% Thioflavin-S solution (in 40% ethanol) for four minutes in the dark. The remaining steps took place in the dark. Following staining, sections were developed with 50% ethanol for 15 minutes and followed by three time PBS wash and one time deionized water wash. Slides were coverslipped in Fluoromount-G and edges were sealed with nail polish. Fluorescent Thioflavin-S signals were imaged using a laser scan confocal microscope (LSM710, Carl Zeiss) by researchers blind to experimental condition. For each tissue section, we imaged five different regions.

Image processing
Images were processed using ImageJ software (NIH) by researchers blind to experimental condition. Measurements consisted of particle counts and size for β-amyloid (Aβ), EEA1, and transferrin immunoreactivity, as well as signal intensity for pTau. Particle count and size were measured using a macro that converted the desired image channel into grayscale with automatic thresholding, inverting, and calling up the Analyze Particle measurement tool. Particle counts were binned by size range. Images were decoded and plotted as fold increase compared to control. For pTau immunoreactivity, the mean intensity of the entire image was measured and plotted as fold increase compared to control.

Enzyme-linked immunoabsorbent assay (ELISA)
Aβ concentration was measured from organoid supernatants using commercially available ELISA kit for Aβ (1-40) and Aβ (1-42) (Life Technologies Corporation, Grand Island, NY) following the manufacturer's protocol. Briefly, media samples were incubated (4 hours) in primary antibodies against the COOH-terminus of the 1-40 or 1-42 Aβ sequence in precoated 96 well plates (pre-coated with monoclonal antibody specific to human Aβ 1-40 or 1-42) followed by aspiration and four washes (in washing buffer) prior to incubation with HRP-conjugated secondary antibodies followed by aspiration, four washes, and addition of HRP substrate (3,3',5,5'-tetramethylbenzidine). The reaction was stopped using 1 N sulfuric acid and absorption was measured at 450 nm in an EnSpire plate readers (Perkin Elmer). Absolute values were calculated from a standard curve and plotted as either picogram/ml (pg/ml) or Aβ42/40 ratio per organoid.

Transferrin assay
Live organoids were cut in small pieces using surgical blade and cultured in a solution of transferrin conjugated to Alexa Fluor-488 (Alexa Fluor-488, Life Technologies Corporation, Grand Island, NY). Following a 10 minute incubation in 200 μg/ml transferrin solution in media, organoids were washed three time in PBS and fixed using 4% paraformaldehyde. The fixed organoids were then washed with PBS, stained with Hoechst, and whole-mounted onto glass coverslip. The outer (flat) surface of tissue sitting on the glass coverslip were imaged using confocal microscope. The images were processed in ImageJ, by removing the background (thresholding) and counting the size of Alexa Fluor-488 positive particles.

Drug treatment
Organoids were treated for 30 or 60 days with a combination of beta secretase (BACE-1) inhibitor (β-Secretase Inhibitor IV, EMD Millipore) and gamma secretases inhibitor (Compound E; EMD Millipore) or equivalent DMSO vehicle. Each compound was diluted into culture medium from a 5 mM DMSO stock that had been stored at -20°C. The concentrations of BACE-1 and Comp-E that were used were 5 μM and 6 nM or 1 μM and 3 nM, respectively. Vehicle-treated cultures received the same concentration of DMSO. Drug treatment was begun at day 30 of the culture and maintained until the endpoint at 60 or 90 days of culture. Media containing treatment or vehicle was replaced every 4 to 5 days. Organoids were processed for immunohistochemistry as described above.

Statistical Analysis
The data were plotted using GraphPad Prism software. Groups were compared via Student's two tailed t-test (two groups) or one-way analysis of variance (ANOVA; multiple groups). Please see figure legends for details.