WO2020214798A1

WO2020214798A1 - Epigenetic signatures of alzheimer's disease

Info

Publication number: WO2020214798A1
Application number: PCT/US2020/028491
Authority: WO
Inventors: Yujiang G. SHI; Irfete FETAHU
Original assignee: The Brigham And Women's Hospital, Inc.
Priority date: 2019-04-17
Filing date: 2020-04-16
Publication date: 2020-10-22

Abstract

The invention, in general, relates to a method of determining whether a human subject will develop Alzheimer's disease, the method including measuring the quantity of 5-methyl-cytosine (5mC) nucleobases in one or more of genes ADA2, PRKACA, NFIX, NFATC1, GSE1, KIF26A, NACC2, FTCH1, MIR153-2, PKHD1, PCDHA2, ACKR3, NR4A2, ECE1, CASZ1, and ARHGEF16 in a neuron obtained from the subject, wherein a finding that the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer's disease. Kits and cells relating to such methods are also provided.

Description

EPIGENETIC SIGNATURES OF ALZHEIMER'S DISEASE

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims benefit of U.S. Provisional Application No. 62/835,159, filed April 17,

2019, the contents of which are incorporated herein by reference in their entirety.

SEQUENCE LISTING

The instant application contains a Sequence Listing which has been submitted electronically in ASCII format and is hereby incorporated by reference in its entirety. The ASCII copy, created on April 14,

2020, is named 51397-002W02_Sequence_Listing_04.14.20_ST25 and is 2,250 bytes in size.

BACKGROUND OF THE INVENTION

The invention relates to methods and compositions for diagnosing and prognosing Alzheimer’s disease.

Alzheimer’s disease (AD) is clinically characterized by the accumulation of amyloid-b (Ab) plaques and neurofibrillary tangles, synaptic and neuronal loss, and cognitive decline (Lardenoije et al., Prog Neurobiol, (2015). Mutations in the Amyloid Precursor Protein (APP), Presenilin 1 or 2 (PSEN1 &2) genes lead to autosomal dominant AD, with disease onset occurring before the age of 65 (Lanoiselee et al., PLoS Med 14, e1002270 (2017)), while the presence of two copies of Apolipoprotein e4 (APOE4) is associated with late onset AD (onset >60 years) (Corder et al., Science 261 , 921 -923 (1993), Strittmatter et al., Proc Natl Acad Sci U S A 90, 1977-1981 (1993)). Recently, large scale genome-wide association studies (GWAS) have identified several additional genes involved with the onset and progression of AD (Bagyinszky et al., Clin Interv Aging 9, 535-551 (2014)), however, the identified genes have not had a meaningful prognostic value for the onset of AD when compared to the above-mentioned genes.

Altogether, these genetic studies provide a framework of AD-associated gene networks and a classic panel for genetic screening of familial AD, which accounts for less than 20% of the AD patient population (Povova et al., Biomed Pap Med Fac Univ Palacky Olomouc Czech Repub 156, 108-1 14 (2012)). On the other hand, the identification of effective biomarkers for the early detection and diagnosis of sporadic AD, which accounts for the vast majority of AD cases, remains a major challenge.

DNA methylation at the 5th-position of cytosine (5mC) plays an important role in neuronal gene expression and neural development. Aberrant DNA methylation is associated with many neuronal disorders, including AD (Rudenko et al., Neuron 79, 1 109-1 122 (2013), Klein et al., Nat Genet 43, 595- 600 (201 1), Kaas et al., Neuron 79, 1086-1093 (2013), Zhang et al., Cell Stem Cell 13, 237-245 (2013)). 5mC can be further oxidized to 5hmC, 5fC, and 5caC by the ten-eleven-translocation (TET) family of dioxygenases. The oxidized products of 5mC accumulate during brain development and early adulthood, suggesting an independent and critical role in neuronal development and function (Song et al., Nat Biotechnol 29, 68-72 (201 1), Lister et al., Science 341 , 1237905 (2013)). Several studies have suggested that dysregulated DNA methylation/demethylation is linked to the AD onset and progression (Chouliaras et al., Neurobiol Aging 34, 2091 -2099 (2013), Bradley-Whitman et al., Mech Ageing Dev 134, 486-495 (2013), Coppieters et al., Neurobiol Aging 35, 1334-1344 (2014), Lashley et al., Neuropathol Appl Neurobiol 41 , 497-506 (2015), De Jager et al„ Nat Neurosci 17, 1 156-1 163 (2014), Zhao et al.,

Alzheimers Dement 13, 674-688 (2017)), however, the relationship between altered levels of 5mC, 5hmC, 5fC, and 5caC and AD is not known. This is primarily due to the lack of comprehensive (base resolution) integrated reference maps of these epigenetic marks during neural cell differentiation, maturation, and brain development. It is also in part due to lack of experimental AD models and research strategies to map, analyze, and define AD-specific changes of the methylome and their oxidized derivatives.

To diagnose AD, the medical community evaluates a subject’s signs and symptoms and conducts several tests. An accurate diagnosis of AD is an important first step to ensure an individual has appropriate treatment, care, family education, and plans for the future.

Accordingly, there is a need in the art for tools that quickly and accurately diagnose AD. The present invention addresses this need.

SUMMARY OF THE INVENTION

The present invention provides, inter alia, a method of diagnosing Alzheimer's disease (AD) by detecting levels of 5mC, 5hmC, and 5fC/caC in iPSC, neural cells, or brain tissues. According to our invention, gene regions, described herein as "epigenetic signatures" allow for distinguishing between healthy iPSCs/neurons/brain tissue and AD-patient derived iPSCs/neurons/brains. These epigenetic signatures coupled with the ability to generate neurons from blood or skin cells provide a powerful platform and demonstrate how this information is used to develop epigenome-based diagnostic tools for AD using minimally invasive approaches.

The invention generally features various methods of diagnosing Alzheimer’s disease by measuring neural cytosine modifications in particular genes of interest.

In one aspect, the invention, in general, features method of determining whether a human subject will develop Alzheimer’s disease, the method including measuring the quantity of 5-methyl-cytosine (5mC) nucleobases in one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , and ARHGEF16 in a neuron obtained from the subject, wherein a finding that the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or more genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

In another aspect, the invention features method of determining whether a human subject will develop Alzheimer’s disease, the method including measuring the quantity of 5mC nucleobases in one or both of genes LINC02055 and KHDRBS3 in a neuron obtained from the subject, wherein a finding that the quantity of 5mC nucleobases in the one or both genes is significantly greater than a reference level of 5mC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or both genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

In yet another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including measuring the quantity of 5-hydroxymethyl- cytosine (5hmC) nucleobases in one or more of genes H1 F00, UNCX, HTRA1 , and C15orf62 in a neuron obtained from the subject, wherein a finding that the quantity of 5hmC nucleobases in the one or more genes is significantly less than a reference level of 5hmC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the one or more genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

In yet another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including measuring the quantity of 5hmC nucleobases in gene 2K4 in a neuron obtained from the subject, wherein a finding that the quantity of 5hmC nucleobases in the gene is significantly greater than a reference level of 5hmC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the gene as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

In yet still another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including measuring the quantity of 5-formyl- cytosine (5fC) or 5-carboxy-cytosine (5caC) nucleobases in one or both of genes INHBB and HLA-A in a neuron obtained from the subject, wherein a finding that the quantity of 5fC or 5caC nucleobases in the one or both genes is significantly less than a reference level of 5fC or 5caC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the one or both genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

In another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including measuring the quantity of 5fC or 5caC nucleobases in gene MIR4532 in a neuron obtained from the subject, wherein a finding that the quantity of 5fC or 5caC nucleobases in the gene is significantly greater than a reference level of 5fC or 5caC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the gene as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

In another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including:

a. reprogramming a population of somatic cells obtained from the subject into induced pluripotent stem cells (iPSCs); and

b. differentiating the iPSCs into neurons and measuring the quantity of 5mC nucleobases in one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , and ARHGEF16 in the neurons;

wherein a finding that the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or more genes as measured in neurons that are obtained by

differentiating neural progenitor cells (NPCs) derived from a human subject that does not have

Alzheimer’s disease. In still another embodiment, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5mC nucleobases in one or both of genes LINC02055 and KHDRBS3 in the neurons;

wherein a finding that the quantity of 5mC nucleobases in the one or both genes is significantly greater than a reference level of 5mC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or both genes as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease. In still another embodiment, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5hmC nucleobases in one or more of genes H1 F00, UNCX, HTRA1 , and C15orf62 in the neurons;

wherein a finding that the quantity of 5hmC nucleobases in the one or more genes is significantly less than a reference level of 5hmC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the one or more genes as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease. In still another embodiment, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5hmC nucleobases in gene 2K4 in the neurons; wherein a finding that the quantity of 5hmC nucleobases in the gene is significantly greater than a reference level of 5hmC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the gene as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease. In still another embodiment, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5fC or 5caC nucleobases in one or both of genes INHBB and HLA-A in the neurons; wherein a finding that the quantity of 5fC or 5caC nucleobases in the one or both genes is significantly less than a reference level of 5fC or 5caC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the one or both genes as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease. In still another embodiment, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5fC or 5caC nucleobases in gene MIR4532 in the neurons; wherein a finding that the quantity of 5fC or 5caC nucleobases in the gene is significantly greater than a reference level of 5fC or 5caC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease. In one embodiment, the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the gene as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease. In still another embodiment, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

In any of the aforementioned embodiments, the invention further includes somatic cells which are reprogrammed into iPSCs by transfecting the somatic cells with one or more of genes OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC. For example, such somatic cells may be transfected by

electroporation in the presence of one or more vectors that together encode OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC.

Still further, in any of the aforementioned embodiments, the iPSCs are differentiated into neurons by (i) differentiating the iPSCs into NPCs by contacting the iPSCs with mouse embryonic fibroblasts (MEFs), a rho kinase inhibitor, and fibroblast growth factor 2 (FGF2), and subsequently (ii) differentiating the NPCs into neurons. For example, in some embodiments, the rho kinase inhibitor is Y-27632. In other examples, the NPCs are differentiated into neurons by contacting the NPCs with an inhibitor of bone morphogenetic protein (BMP) signaling and a transforming growth factor b (TGF-b) receptor inhibitor. Such inhibitors of BMP signaling include noggin or dorsomorphin. Inhibitors of the TGF-b receptor inhibitor include SB431542. In some embodiments, 5mC or 5hmC nucleobase modifications are measured by way of oxidative bisulfite sequencing. In still other embodiments, 5fC or 5caC nucleobase modifications are measured by way of methylase-assisted bisulfite sequencing.

Still further, in any of the aforementioned embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in two or more of the genes recited above. For example, in some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in three or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in four or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in five or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in six or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in seven or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in eight or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in nine or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in 10 or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in 15 or more of the genes recited above. In some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in 20 or more of the genes recited above. For example, in some embodiments, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured in all 26 of the genes recited above.

Still further, in any of the aforementioned embodiment, the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured at a chromosomal site set forth in any one of Tables 3A - 3C, 4A - 4C, and 5A - 5C.

The invention further includes methods of diagnosing Alzheimer’s disease by monitoring cytosine modification patterns across different stages of neural development.

Accordingly, in yet another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including:

a. reprogramming a population of somatic cells obtained from the subject into iPSCs and measuring the quantity of 5fC or 5caC nucleobases in the iPSCs;

b. differentiating the iPSCs into NPCs and measuring the quantity of 5fC or 5caC nucleobases in the NPCs; and

c. differentiating the NPCs into neurons and measuring the quantity of 5fC or 5caC nucleobases in the neurons; wherein a finding that the quantity of 5fC or 5caC nucleobases in the NPCs is not significantly greater than (i) the quantity of 5fC or 5caC nucleobases in the iPSCs and (ii) the quantity of 5fC or 5caC nucleobases in the neurons identifies the subject as one that will develop Alzheimer’s disease.

In still another aspect, the invention features a method of determining whether a human subject will develop Alzheimer’s disease, the method including:

a. reprogramming a population of somatic cells obtained from the subject into iPSCs and measuring the quantity of 5hmC nucleobases in the iPSCs; b. differentiating the iPSCs into NPCs and measuring the quantity of 5hmC nucleobases in the NPCs; and

c. differentiating the NPCs into neurons and measuring the quantity of 5hmC nucleobases in the neurons; wherein a finding that the quantities of 5hmC nucleobases in the iPSCs, the NPCs, and the neurons do not significantly differ from one another identifies the subject as one that will develop Alzheimer’s disease.

a. reprogramming a population of somatic cells obtained from the subject into iPSCs and measuring the quantity of 5mC or 5hmC nucleobases in the iPSCs;

b. differentiating the iPSCs into NPCs and measuring the quantity of 5mC or 5hmC nucleobases in the NPCs; and

c. differentiating the NPCs into neurons and measuring the quantity of 5mC or 5hmC nucleobases in the neurons; wherein: (i) a finding that the quantity of 5mC or 5hmC nucleobases in the iPSCs is significantly greater than an iPSC reference level of 5mC or 5hmC nucleobases identifies the subject as one that will develop Alzheimer’s disease; and/or (ii) a finding that the quantity of 5mC or 5hmC nucleobases in the NPCs is significantly lower than a NPC reference level of 5mC or 5hmC nucleobases identifies the subject as one that will develop Alzheimer’s disease; and/or

(iii) a finding that the quantity of 5mC or 5hmC nucleobases in the neurons is significantly greater than a neuronal reference level of 5mC or 5hmC nucleobases identifies the subject as one that will develop Alzheimer’s disease.

In some embodiments of the aforementioned aspect, the iPSC reference level of 5mC or 5hmC nucleobases is a quantity of 5mC or 5hmC nucleobases measured in iPSCs that are obtained by reprogramming somatic cells from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPC reference level of 5mC or 5hmC nucleobases is a quantity of 5mC or 5hmC nucleobases measured in NPCs derived from a human subject that does not have Alzheimer’s disease.

In some embodiments, wherein the neuronal reference level of 5mC or 5hmC nucleobases is a quantity of 5mC or 5hmC nucleobases measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease. In some embodiments, the somatic cells are reprogrammed into iPSCs by transfecting the somatic cells with one or more of genes OCT4, SOX2, NANOG, LIN28, KLF4, and L- MYC. In some embodiments, the somatic cells are transfected by electroporation in the presence of one or more vectors that together encode OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC. In some embodiments, the iPSCs are differentiated into NPCs by contacting the iPSCs with MEFs, a rho kinase inhibitor, and FGF2. For example, in some embodiments, the rho kinase inhibitor is Y-27632.

In other embodiments, the NPCs are differentiated into neurons by contacting the NPCs with an inhibitor of BMP signaling and a TGF-b receptor inhibitor. For example, in some embodiments, the inhibitor of BMP signaling is noggin or dorsomorphin. In other embodiments, the TGF-b receptor inhibitor is SB431542.

In still other embodiments of the aforementioned aspects, Alzheimer’s disease is early-onset Alzheimer’s disease, late-onset Alzheimer’s disease, familial Alzheimer’s disease, or sporadic

Alzheimer’s disease.

In still other embodiments of the aforementioned aspects, the 5fC or 5caC nucleobase modifications are measured by way of methylase-assisted bisulfite sequencing.

In still other embodiments of the aforementioned aspects, the 5mC or 5hmC nucleobase modifications are measured by way of oxidative bisulfite sequencing.

Still further the invention includes kits for diagnosing Alzheimer’s disease.

Accordingly, in one aspect, the invention features a kit including a bisulfite salt and a plurality of nucleic acid primers suitable for amplification of one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , ARHGEF16, LINC02055 KHDRBS3, H1 F00, UNCX, HTRA1 , C15orf62, 2K4, INHBB, HLA-A, and MIR4532. In some embodiments, the kit further includes a DNA polymerase. In some other embodiments, the kit includes a package insert instructing a user to perform the method of any of the aforementioned aspects or embodiments of the described methods.

In still another aspect, the invention features a kit including a CpG methyltransferase and a plurality of nucleic acid primers suitable for amplification of one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 ,

CASZ1 , ARHGEF16, LINC02055 KHDRBS3, H1 F00, UNCX, HTRA1 , C15orf62, 2K4, INHBB, HLA-A, and MIR4532. In some embodiments, the CpG methyltransferase is M.Sssl methyltransferase. In some other embodiments, the kit includes a package insert instructing a user to perform the method of any of the aforementioned aspects or embodiments of the described methods.

In some other embodiments of any of the aforementioned kits, the kit further includes a panel of one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , ARHGEF16, LINC02055 KHDRBS3, H1 F00, UNCX, HTRA1 , C15orf62, 2K4, INHBB, HLA-A, and MIR4532 obtained from a human subject that does not have Alzheimer’s disease.

The invention further includes non-naturally occurring cells exhibiting particular cytosine modifications which are prepared according to methods disclosed herein.

Accordingly, in one aspect, the invention features a neuron including one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2,

ACKR3, NR4A2, ECE1 , CASZ1 , and ARHGEF16, wherein the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes. In some embodiments, the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or more genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

In another aspect, the invention features a neuron including one or both of genes LINC02055 and KHDRBS3, wherein the quantity of 5mC nucleobases in the one or both genes is significantly greater than a reference level of 5mC nucleobases in the one or both genes. In some embodiments, the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or both genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

In still another aspect, the invention features a neuron including one or more of genes H1 F00, UNCX, HTRA1 , and C15orf62, wherein the quantity of 5hmC nucleobases in the one or more genes is significantly less than a reference level of 5hmC nucleobases in the one or more genes. In some embodiments, the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the one or more genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

In still yet another aspect, the invention features a neuron including gene 2K4, wherein the quantity of 5hmC nucleobases in the gene is significantly greater than a reference level of 5hmC nucleobases in the gene. In some embodiments, the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the gene as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

In yet another aspect, the invention features a neuron including one or both of genes INHBB and HLA-A, wherein the quantity of 5fC or 5caC nucleobases in the one or both genes is significantly less than a reference level of 5fC or 5caC nucleobases. In some embodiments, the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the one or both genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease. In some other embodiments, the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

In another aspect, the invention features a neuron including gene MIR4532, wherein the quantity of 5fC or 5caC nucleobases in the gene is significantly greater than a reference level of 5fC or 5caC nucleobases in the gene. In some embodiments, the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the gene as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease. In some embodiments, the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

In still further embodiments of any of the aforementioned neurons, the neuron is a human neuron.

The invention provides numerous advantages. We have identified a set of AD-specific CpG signatures across all states of DNA cytosine methylation in early onset, late onset, familial, and sporadic AD models. The 27 signature regions and the unique 39 CpG sites (described herein) flagged in the roadmap, highlight key distinguishable and measurable features of AD epigenome from the norm, making them valuable and suitable for early molecular diagnosis of AD. These signatures are AD-specific and age-independent, which were further validated in a large clinical cohort. Significantly, a major group of the signatures occur on the genes implicated in AD or AD-related pathways, or on genomic regions that are critical for proper neurodevelopment. Finally, these 27 signatures and more specifically the 39 CpG sites identified can easily be converted into a minimally invasive diagnostic panel that allows for testing of these predictable molecular signatures for detection of AD.

Our findings demonstrate that the existing 5mC, 5hmC, and 5fC/caC changes associated with our signatures may precede the disease process, and they are, therefore, not secondary or consequence to the later stages of the disease progression. Furthermore, although epigenetic landscapes have been suggested to change with increasing age, our identified signatures are age-independent; rather, our 5mC, 5hmC, and 5fC/caC signatures appeared to be limited to AD condition, regardless of familial or sporadic origin. Our findings not only provide mechanistic insight into disease etiology, but also potentially identify strategies for early detection and therapeutic intervention at preclinical stages.

The present invention further advantageously provides comprehensive and comparative reference maps of genome-wide distribution of all three DNA methylation states at base resolution during neural differentiation from iPSCs to mature neurons, and in post-mortem brain tissues of both normal donor and AD patients. These maps not only uncovered the dynamic changes of the landscapes of 5mC, 5hmC, and 5fC/caC during neural differentiation of normal and AD-derived iPSCs, but also allowed us to pin-down the specific changes at any given cytosine and that of three methylation states. These findings identified that the dynamic and coordinated changes of three DNA methylation states, are important epigenetic features, specifically associated with neural differentiation process. Importantly, we found that the observed epigenetic features in normal cells were dysregulated in AD cell models, indicating that precise regulation, proper establishment, and maintenance of these epigenetic features likely plays important roles in regulation of neural lineage commitment, maturation, and function.

In addition, the cost of an AD epigenetic signatures test is substantially lower than magnetic resonance imaging and molecular neuroimaging with positive emission tomography.

Other features and advantages of the invention will be apparent from the following description of the preferred embodiments thereof, and from the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 shows a schematic of our study model. Abbreviations: iPSC: Induced Pluripotent Stem Cells, NPC: Neural Precursor Cells, N: Neurons, AD: Alzheimer’s disease, APOE4: Apolipoprotein E isoform 4, PSEN1 : Presenilin 1 , PSEN2: Presenilin 2, WT: Wild Type. MAB-Seq: Methylase-Assisted Bisulfite Sequencing, OXBS-Seq: Oxidative Bisulfite Sequencing. DMRs: differentially methylated regions, DHMRs: differentially hydroxymethylated regions, and DFCRs: differentially formyl/carboxylated regions.

FIG. 2 shows global methylation trends in iPS cells and iPS cell-derived NPCs and neurons in WT cell lines. FIG. 2A-2B show pluripotency, neural stem cell, and neural cell markers determined by qRT-PCR and immunofluorescence. Scale bar=200pm. FIG. 2C shows distribution of

5mC/5hmC/5fC/caC levels in WT iPS cells, NPCs, and neurons. The y axis indicates the levels of 5mC, 5hmC, or 5fC/caC at each reference cytosine (at least 10 reads required). FIG. 2D shows single base profiles of 5mC, 5hmC, and 5fC/caC for the pluripotency gene OCT4 and neural-specific gene MAP2 in iPSCs and the consecutive stages of iPSC-derived NPCs and neurons. Genomic coordinates for OCT4: chr6:31 ,132,042-31 ,139,528 and MAP2 : chr2:210,002,875-210,783,41 1. In all samples the scale for 5mC, 5hmC, and 5fC/caC was 0-100%.

FIG. 3 shows directed differentiation of human iPS cells to neural progenitors and cortical neurons, and expression of stage-specific markers. Related to Figure 2. FIG 3A. shows induced pluripotent stem cells expressing pluripotency markers, such as ALPL, KLF4, NANOG, PODXL, and OCT4 ( POU5F1 ), determined by RNA-seq. FIG3B. shows that over the ~15 day neural induction period, OCT4-expressing hiPS cells differentiate at high efficiency to FOXGf-expressing neural stem cells. Transcriptome data indicate the absence of detectable FOXGf-expressing cells in iPSC stage and of OCT4 at NPC stage. Expression of additional neural progenitor cell markers, such as NES and VIM, confirm complete induction of iPS cells to NPC. FIG3C shows that neurogenesis was stimulated by withdrawal of mitogens, following which increasing numbers of neurons were generated as determined by expression of TUBB3 and MAP2, neuron-specific tubulins, and CTIP2 ( BCL11B ) expression marking corticothalamic projection neurons of layer.

FIG. 4 shows disease and developmental differences in WT and AD neurons. FIG. 4A-4C show distribution of 5mC/5hmC/5fC/caC levels in PSEN2, PSEN1 , and APOE4 patient-derived cell lines. The y axis indicates the levels of 5mC, 5hmC, or 5fC/caC at each reference cytosine (at least 10 reads required). FIG. 4D shows representative regions showing 5mC, 5hmC, and 5fC/5caC levels at single base resolution in WT and AD-patient derived cell lines at HLA gene cluster (chr6:32,440,096- 32,643,715).

FIG. 5 shows iPSCs and iPS cell-derived NPCs and neurons in PSEN2 lines are characterized by distinct methylomes compared to the WT. FIG. 5A shows total 5mC, 5hmC, and 5fC/caC levels in iPSCs and iPS cell-derived NPCs and neurons in WT and PSEN2. The x-axis indicates the levels of 5mC,

5hmC, or 5fC/caC at each reference cytosine (at least 10 reads required). Asterisks mark p<2.2e-16, determined by Wilcoxon rank sum test. FIG. 5B shows levels of 5mC, 5hmC, and 5fC/caC in WT and PSEN2 cell lines at TSS and gene body in iPSCs, NPCs, and neurons. For each cytosine modification gene body was normalized to 0-100%. Normalized density is plotted from 5kbp upstream of TSSs to 5kbp downstream of TESs. FIG. 5C shows stacked bars show the differentially 5mC, 5hmC, and 5fC/caC regions presented as gains and losses during directed differentiation of iPS cells to neurons in WT and PSEN2 lines. FIG. 5D shows representative regions showing 5mC, 5hmC, and 5fC/5caC levels at single base resolution in WT and PSEN2 cell lines. In all samples the scale for 5mC and 5fC/caC was 0-100%, while for 5hmC the scale was 0-70%. Genome coordinates for the images shown were: PCDHB8:

chr5:140,556,693-140,560,759, ANK1: chr8:41 ,731 ,1 12-41 ,755,347, HLA-DPA1 : chr6:33,030,742- 33,049,833.

FIG. 6 shows a biological interpretation of genomic regions identified in the DMRs/DHMRs/ DFCRs in PSEN2 and WT cell lines. Related to Figure5 A-C show a gene ontology functional analysis enrichment was performed using cluterProfiler in Bioconductor. Results were filtered using p<0.05 as inclusion cutoff, and only top 10 terms were kept in each group. Abbreviations: GO: Gene ontology, DMR: differentially methylated regions, DHMR: differentially hydroxy methylated regions, DFCR: differentially formyl/caroboxylated regions.

FIG. 7 shows epigenetic signatures associated with Alzheimer’s disease. FIG. 7A-7C show the outermost circle (1) presents chromosome ideograms (in Mb). Changes in 5mC (FIG. 7A), 5hmC (FIG. 7B), or 5fC/caC (FIG. 7C) levels are shown as differences of 5mC, 5hmC, or 5fC/caC levels between normal and AD. Red bars indicate gain of 5mC, 5hmC, or 5fC/caC in AD vs. controls, while blue bars mark loss of 5mC, 5hmC, 5fC/caC in AD vs. normal. The second layer of the circle (2) shows genes associated with AD epigenetic signatures. Genes in red letters fall in the ultraconserved regions. Layers 3-7 show 5mC, 5hmC, and 5fC/caC ratios, respectively, in AD vs. control: (3) PSEN2 neurons/WT neurons, (4) PSEN1 neurons/WT neurons, (5) APOE4 neurons/WT neurons, (6) AD-frontal lobe/Normal brain, and (7) AD-left frontal lobe/Normal brain. FIG. 7D shows 5mC, 5hmC, and 5fC/caC signature regions narrowed down to individual CpG sites. Using methylation variation difference of <±1 fold-change for 5mC and <±1.5 fold-change for 5hmC and 5fC/caC, we identified 19 regions for 5mC signatures, 5 regions for 5hmC signatures, and 3 regions for 5fC/caC signatures. We then further filtered our 5mC, 5hmC, and 5fC/caC signatures by using methylation difference of <-0.1 or >0.1 fold-change as a final cutoff. This yielded 27 CpG sites in the 5mC signatures, out of which 21 and 6 CpG sites were marked by loss or gain of methylation, respectively in AD samples vs. controls. In the 5hmC group, we identified 3 CpG sites and 1 CpG site, which were marked by loss or gain of 5hmC, respectively, in AD samples compared to controls. In the 3 signature regions of 5fC/caC, we identified an equal number of CpG sites (4 in each group) that were marked by gain or loss of 5fC/caC levels in AD vs. controls.

FIG. 8 shows receiver operative characteristics revealed that SVM and Decision Tree models have the best predictive ability with an AUC=0.9128 (specificity=98% and sensitivity=48%) and

AUC=0.8235 (specificity=96% and sensitivity=61 %), respectively. This was followed by predictive ability generated by Logistic Regression, AUC=0.7941 (specificity=86% and sensitivity=55%) and QDA, AUC=0.7905 (specificity=87% and sensitivity=54%).

DETAILED DESCRIPTION OF THE INVENTION

As is described above, Alzheimer’s Disease (AD), a progressive neurodegenerative disorder, is the most common untreatable form of dementia. Below we describe, in a series of Examples, an experimental and analytical model characterizing epigenetic alterations during AD onset and progression.

We generated an integrated base-resolution genome-wide maps of the distribution of 5-methyl (5mC)-, 5-hydroxymethyl (5hmC)-, and 5-formyl/carboxy (5fC/caC)-cytosine in normal and AD neurons. These maps uncovered the dynamic changes of the three key states of DNA methylation, their distinctive patterns during normal neural differentiation, and their aberrance in AD. We identified 27 AD- region- specific and 39 CpG site-specific epigenetic signatures that were independently validated across our familial and sporadic AD models, and in an independent clinical cohort. These results provide a useful model and strategy to study the epigenetic alterations underlying AD onset and progression and provides a set of highly reliable AD-specific epigenetic signatures useful for early diagnosis and prognosis of the disease.

EXAMPLES

Results

Description of the study model and subjects

The overview of our experimental paradigm is outlined in Fig. 1 . We employed new base- resolution mapping and analytical technologies, such as oxidative bisulfite deep sequencing (OXBS-seq) and methylase-assisted bisulfite deep sequencing (MAB-seq) to characterize dynamic genome-wide distributions of 5mC, 5hmC, and 5fC/5caC modifications. We established cell culture models of neurogenesis by directing the differentiation of iPS cells derived from healthy control (WT) and AD-patient derived cells to neural progenitor cells, and then to cortical neurons, as previously described (Shi et al., Nat Protoc 7, 1836-1846 (2012)). The cell lines include a normal cell line, WT; two early-onset familial AD (EOAD) cell lines, PSEN1 and PSEN2, and a late-onset familial AD cell line (LOAD), APOE4, which are the strongest genetic risk factors for developing AD (Bagyinszky et al., Clin Interv Aging 9, 535-551 (2014)). AD cell lines either carrying mutations in PSEN1 , PSEN2, or homozygous APOE4 served as ‘AD-in-dish” models to mimic EOAD and LOAD, respectively. To ensure generalizability of the epigenetic features and signatures of healthy controls and AD-associated epigenetic alterations found in cell culture models, we extended our epigenomic studies to post-mortem brain tissues of a normal donor and sporadic AD cases. Finally, we validated the AD-specific 5mC signatures defined in our study using publicly available normal and AD cohorts.

Identification of 5mC, 5hmC, and 5fC/caC landscapes during neuroqenesis

Pluripotency genes, such as NANOG, OCT4, SOX2 are expressed in iPSC-WT. Expression of NPC markers, including FOXG1 , NES, and TBR2 confirmed the complete induction of iPS cells to NPCs. Subsequently, successful differentiation of NPCs to neurons (N) was confirmed with the expression of neuronal specific marks, such as CUX1 , MAP2, TBR1 , TUJ1 (Figs. 2A&B). Similarly, all iPS cells derived from AD patients can also be differentiated to NPCs and neurons (Figs. 3A-C).

Until recently, uncovering the exact patterns of 5mC, 5hmC, and 5fC/caC at base resolution has proven to be a challenging task due to the lack of technologies that allow to distinguish 5mC from its oxidized modifications (5hmC and 5fC/caC) at a given single cytosine base. To circumvent this problem, we developed an in-house protocol and analytical algorithm to identify genome-wide distribution of sites and levels of 5mC, 5hmC, 5fC/caC based on recently reported OXBS-seq (identifies 5mC and 5hmC, where the combined levels range from 0-100%) and MAB-seq (identifies 5fC/caC, where levels range from 0-100%) (Booth et al., Nat Protoc 8, 1841 -1851 (2013), Neri et al., Cell Rep, (2015)). Globally, the levels of 5mC were higher in both NPCs and neurons (89.6% and 89.3%, respectively, p<2.2e-16) compared to iPS cells (82.5%), indicating that these 5mC sites tend to be more frequently methylated in the differentiated neural stages than in iPS cells (Fig. 2C, left). These findings are consistent with previously published reports (22, 23). Levels of 5hmC were lower (p<2.2e-16) in NPCs (8.3%) and neurons (7.7%) compared to iPS cells (9.8%) (Fig. 2C, middle). The levels of 5fC/caC were higher in the iPS cells (36%) than in neurons (34%, p<2.2e-16) (Fig. 2C, right). Interestingly, we observed a surge of 5fC/caC level (48%) in NPCs, which is in agreement with previous studies in mice showing that levels of 5fC/5caC transiently increase during NPC stage and rapidly decline as cells commit to neural lineages (Wheldon et al., Cell Rep 7, 1353-1361 (2014)), highlighting the critical roles of this modification in lineage commitment. We should note, due to its technical nature, MAB-seq is unable to distinguish between 5fC and 5caC. Therefore, the identified cytosine modification by MAB-seq reflects the total modification of 5fC/caC. Detailed description and quantification of genome-wide sites of 5mC, 5hmC, and 5fC/caC at base resolution is described below under“Quantification of genome-wide sites of 5mC, 5hmC, and 5fC/caC at base resolution”) and in Table 1. Regional- and gene-specific examples of 5mC, 5hmC, and 5fC/caC at base-resolution are shown in Fig. 2D.

Quantification of genome-wide sites of 5mC, 5hmC, and 5fC/caC at base resolution

To gain insight into the global and dynamic changes of genome-wide distribution of 5mC, 5hmC, and 5fC/caC during directed differentiation of iPSCs to neuronal precursor cells (NPC) and neurons (N), we first identified the total number of the modified cytosines in each respective sample. In WT cells, we identified 13637104, 15049856, and 21918240 sites of methylated cytosine (5mC) at iPSC, NPC, and N stages, respectively. By comparative analysis, we observed a 16% and 34% increase of 5mC sites in NPCs and Ns, respectively, compared to iPSCs (Table 1). For 5hmC, we identified 7041346, 7364564, and 10485793 sites in WT iPSCs, NPCs, and Ns, respectively. Changes in 5hmC sites were marked by a 28% decrease in NPCs compared to iPSCs. In contrast, we observed a gain of 5hmC sites as cells differentiate from NPCs to Ns, reaching to similar numbers to that of the iPSC cells (Table 1). For 5fC/caC, we identified 2586021 , 1797954, and 2307023 sites in WT iPSC, NPC, and N, respectively. The number of 5fC/caC sites was reduced by 10% and 21 % in NPCs and neurons, respectively, compared to iPSCs (Table 1).

Table 1. Total 5mC, 5hmC, and 5fC/caC sites in WT and AD cell lines

Disease and developmental differences in 5mC, 5hmC, and 5fC/caC profiles

Using similar analytical approaches as in WT cells, we have also successfully mapped and analyzed at base-resolution 5mC, 5hmC, and 5fC/caC distribution in AD-patient derived cell lines (for technical details and description of each SEQ-datasets, see Table 1). Overall, global levels of 5mC and 5hmC in EOAD models, PSEN1 and PSEN2 cell lines followed similar patterns as the WT cells, which were characterized by gain of 5mC and loss of 5hmC in neurons compared to the iPSCs, whereas levels of 5fC/caC decreased in neurons compared to the iPSCs (Figs. 4A&B). Intriguingly, in both PSEN1 and PSEN2 cell lines, levels of 5fC/caC failed to peak at the NPC stage as opposed to the trend in the WT NPCs. In the LOAD model cell line, APOE4, we noted that 5mC levels increased during differentiation, however, the differences between iPSCs and neurons were not as prominent when compared to the WT or the EOAD cell lines (Fig. 4C). Changes in 5hmC levels were marginal between the different stages as well, while similar to the EOAD models, 5fC/caC failed to peak in the NPC stage, which appears to be a common feature in all AD cell lines (Figs. 4A-C). Furthermore, both PSEN1 and PSEN2 cell lines show similar global patterns of 5mC, 5hmC, and 5fC/caC during differentiation of iPSCs to neurons that appear to be independent of PSEN1 or PSEN2 gene mutations, but common to the EOAD model. Changes in methylation of cytosine at base level for 5mC, 5hmC, and 5fC/caC across our different cell line models as well as during the differentiation process in normal and disease settings are depicted in Fig. 4D.

Differentially methylated DNA regions demarcate key Alzheimer’s disease genes

To determine the loci-specific similarities and differences of epigenetic profiles between WT and AD cells, we next performed more detailed comparative analysis of the 5mC, 5hmC, and 5fC/caC profiles between WT and PSEN2-AD cell models. We focused our initial comparative analysis between WT and PSEN2 lines as the N141 I mutation in PSEN2 has been well characterized and gives rise to autosomal dominant early onset AD in patients younger than 65 years old (Lanoiselee et al., PLoS Med 14, e1002270 (2017)). In both WT and PSEN2, levels of 5mC increased in NPCs and neurons compared to iPSCs (Fig. 5A, top). Gain of 5mC levels was marked by loss of 5hmC levels in both WT and PSEN2 lines during neural differentiation (Fig. 5A, middle). Of note, at any given stage, we observed that both 5mC and 5hmC of PSEN2 were slightly, but significantly higher (p<2.2e-16) in iPSCs and Ns, but lower (p<2.2e-16) in NPCs compared to the WT lines. Interestingly, the dynamic changes of 5fC/caC levels in WT and PSEN2 were markedly different from that of 5mC and 5hmC during neural differentiation. The peak in 5fC/caC levels in the NPC stage as observed in our WT line appears to be an epigenetic feature that is critical in neurodevelopment (Wheldon et al., Cell Rep 7, 1353-1361 (2014)). Yet this pattern fails to be established in PSEN2 cells (Fig. 5A, bottom), suggesting an aberrant 5fC/caC methylome in the PSEN2 AD cell lines.

We next analyzed the distribution of 5mC, 5hmC, and 5fC/caC across averaged RefSeq genes in WT and PSEN2 cell lines. Profiles of 5mC in both WT and PSEN2 cells were similar at all differentiation stages, characterized by a large drop around the TSS and an increase in gene body toward TES, dropping again around TES (Fig. 4B, top row). 5hmC profiles also showed a dip at TSS, followed by a gradual accumulation after TSS, finally surging at TES at all stages in both cell lines (Fig. 5B, middle row). On the other hand, 5fC/caC profiles are more complex than those of 5mC and 5hmC during differentiation in the two cell lines (Fig. 5B, bottom row). In iPSCs and NPCs, the profiles were characterized by a peak at TSS. Of note, in NPCs the peak was greatly diminished in PSEN2 cells. In neurons the peaks observed at TSS in iPSC and NPC disappear and become a dent.

To identify the regions that contain differential levels of 5mC (DMR), 5hmC (DHMR), and 5fC/caC (DFCR) during differentiation (iPSCs to neurons) in both WT and PSEN2 cells, we analyzed stage- specific differences in DMRs, DHMRs, and DFCRs between WT and PSEN2 cell lines (Fig. 5C). The detailed characterization of each stage-specific group of DMRs, DHMRs, or DFCRs between WT and PSEN2 lines is depicted under“Disease and developmental differences in 5mC, 5hmC, and 5fC/caC landscapes in WT and PSEN2 cell lines” below . Importantly, gene ontology enrichment analyses for the genes identified in DMRs (8713), DHMRs (1052), and DFCRs (2147) revealed a significant enrichment on neurodevelopmental processes, including forebrain development (p<10-6), synapse organization (p<10- 9), differentiation of neurons (p<10-8), etc., highlighting the link between 5mC, 5hmC, and 5fC/caC methylome changes and disease-critical neural gene programs. The top ten most highly significant gene sets ranked by p-value are presented in Figs. 6A-C. Focusing solely on the pathways identified in the gene ontology enrichment analysis that contained known AD risk genes revealed that changes in the epigenetic component overlap with diverse and critical biological pathways, including immune response, metabolism, and oxidative stress response, all of which have been shown to be disrupted in AD (Tables 2A-C).

Other relevant genes identified in DMRs, DHMRs, or DFCRs, include ANK1 , BACE2, BIN1 , CLU, HLA-DPA1 , PCDHB8, which are known AD risk genes and have well-established roles in AD pathology. These findings demonstrate that changes in 5mC, 5hmC, and 5fC/caC epigenetic profiles are directly associated with a network of key AD susceptibility genes in PSEN2 AD cell models. For instance, we observed loss of 5mC at TSS and gene body of PCDHB8 at all stages in PSEN2 cells compared to the WT cells. This gene codes for neural-cadherin like protein and has been reported as an AD risk gene. ANK1 is a well-established gene with a critical role in AD pathology, and few studies have reported its epigenetic deregulation in AD (De Jager et al., Nat Neurosci 17, 1156-1163 (2014), Lunnon et al., Nat Neurosci 17, 1 164-1 170 (2014)). ANK1 is generally characterized by accumulation of 5hmC in the PSEN2 model in comparison to the WT. An example from the DFCRs is HLA-DPA1 gene. The role of HLA-DPA1 has also been studied in the context of AD, and genetic polymorphisms of this gene increase risk for AD (26). We observed stark differences in 5fC/caC levels along the entire HLA-DPA1 gene in the PSEN2 cell model compared to the WT (Fig. 5D). Collectively, our findings highlight the pivotal role of proper timing and landscape of the three DNA methylation states in regulation of AD-critical genes. Since the patterns of methylation of all states are disrupted in AD-patient derived lines at all differentiation stages, it is postulated that the dysregulation of the 5mC, 5hmC, and 5fC/caC on these key AD genes may intrinsically and directly be linked to the AD pathology.

Disease and developmental differences in 5mC, 5hmC, and 5fC/caC landscapes in WT and

PSEN2 cell lines

To characterize changes in the methylome during differentiation along the neural trajectory, we scanned methylomes of the WT and PSEN2 cell lines to comprehensively identify the regions that contain different levels of 5mC (DMR), 5hmC (DHMR), and 5fC/caC (DFCR). In the DMR group we identified a total of 12417 regions, out of which 27.7% belong to the iPSC-specific DMRs with 11 .2% being hypermethylated in PSEN2 compared to the WT, while 16.5% was marked by loss of methylation in PSEN2-iPSCs compared to WT-iPSCs. The NPC-specific group was mostly associated with gain of 5mC in PSEN2 (gain DMRs=26.6%) and only a small fraction of 2.9% had loss of 5mC in PSEN2 compared to the WT NPCs. Finally, the neuron-specific DMRs, which consist the largest DMR group (42.9%), showed gain of methylation in 5.6% and loss of methylation in 37.3% of the total DMRs in PSEN2 neurons compared to the WT neurons. In the DHMR group we identified in total 1 10 regions, where the largest DHMR group was specific to the iPS cells with 12.7% of the regions marked by gain and 25.5% of the regions marked by loss of 5hmC in PSEN2 iPSCs compared to the WT iPSC. The number of NPC- specific DHMRs was marked by identical number of regions (16.4%) that had gains and losses of 5hmC PSEN2 compared to the WT. Lastly, 10% and 19.1 % of neuron-specific DHMRs were marked by gain and loss of 5hmC, respectively in PSEN2 cells compared to the WT lines. In the last group, DFCRs, we identified a total of 694 DFCRs, out of which 12.7% and 14% had gain or loss of 5fC/caC, respectively in PSEN2 iPSC compared to the WT. In the NPC-stage, 10.2% of DFCRs gained 5fC/caC, while 16.3% were marked by loss of 5fC/caC in PSEN2 compared to WT. Finally, in neurons, 27.2% of DFCR gained 5fC/caC, whereas 19.6% of the regions had lost 5fC/caC in PSEN2 compared to the WT (Fig. 5C).

Discovery of Alzheimer’s disease-specific 5mC, 5hmC, and 5fC/caC signatures

To verify the aforementioned global changes in DNA methylation at all three states in additional AD cell lines, we applied the same in-depth analysis approaches to characterize DMRs, DHMRs, and DFCRs in another EOAD cell model, PSEN1 and the LOAD model, APOE4. The comparison analysis identified a total of 2197 DMRs, 22 DHMRs, and 128 DFCRs, which were shared among all three AD cell models. To broaden our findings from cell culture models and to ensure that the changes of DNA methylation states are disease-specific and not due to the aging process itself or a cell culture artifact, we also mapped and analyzed 5mC, 5hmC, and 5fC/caC of DNA samples derived from post-mortem brain tissues of healthy donor and AD patients. Finally, we overlapped our two sets of DMRs, DHMRs, and DFCRs (originating from cell lines or brain tissues), which resulted in 1447 DMRs, 7 DHMRs, and 23 DFCRs.

We further extracted these shared DMRs/DHMRs/DFCRs to a final of 71 AD-specific epigenetic signature regions located in autosomes, which were defined according to the following criteria: 1) all signature loci must be consistently presenting the same trends of gain or loss of 5mC, 5hmC, or 5fC/caC across all disease models compared to the control, 2) because 5mC is the most abundant modification we applied more stringent criteria using a methylation difference of <-1 or >1 as a cutoff to reduce confounding background signal, while cutoff for 5hmC and 5fC/caC was <-0.1 or >0.1 . These 71 regions were associated with 56 different genes (Table 3A-B). Because biological reproducibility is key, we applied another filter to our regional signatures based on methylation variation (using a cutoff of <±1 fold- change for 5mC and <±1 .5 fold-change for 5hmC and 5fC/caC) between the samples we employed in our study.

Abbreviations: APOE4: Apolipoprotein E isoform 4, PSEN1 : Presenilin 1 , PSEN2: Presenilin 2, WT: wild type, AFL: AD frontal lobe, ALFL: AD left frontal lobe, NLB: normal left brain

This led to 19 signature regions for the 5mC modification, out of which, 17 were marked by loss of 5mC and 2 were marked by gain of 5mC in AD samples compared to the controls (Fig. 6A). Using the same approach, we narrowed down our 5hmC signature regions to 5, out of which 4 were marked by loss of 5hmC and 1 was characterized by gain of 5hmC in AD samples vs. controls (Fig. 6B). Similarly, for the 5fC/caC group of signatures we identified final 3 signatures, where 2 had loss of 5fC/caC, while one had gain of 5fC/caC in AD samples compared to normal controls (Fig. 6C).

Most of the 5mC signatures were (8/19) located in intronic regions, followed by 5/19 in promoters,

3/19 in distal intergenic regions, 2/19 in exons, and 1/19 in 3’UTRs. The 5hmC gene signatures were physically located as follows: 1/5 in distal intergenic regions and the rest were equally distributed in promoters (2/5) and 3’UTRs (2/5). All 5fC/caC (3/3) signatures were located in distal intergenic regions

(Tables 4A-C).

The genes harboring epigenetic signatures can be classified into four major subgroups according to their diversified biological function: 1) neurodevelopment and neuronal transcription factors, 2) critical cellular processes, 3) RNA and associated proteins (including non-coding RNAs), and 4) cell signaling. Additional information for each signature region is provided in Table 4 and below under“Functional classification of the genes that were associated with AD-specific signatures”. Collectively, we have identified 27 regional signatures of the methylated DNA cytosine that are specifically associated with AD.

Finally, since our approaches allow us to investigate these cytosine modifications at single base resolution, we next aimed to narrow down our regional signatures to individual CpG sites based on the following criteria: 1) identify the CpG sites that overlapped in the signature regions in all study samples, 2) sort CpG sites that have same trends of methylation (gain or loss) in AD vs. controls, and 3) apply cutoff of methylation fold-change difference of <-0.1 and >0.1 to avoid false positives. This led to a total of 39 AD-specific CpG sites that were consistently modified in early onset, late onset, familial, and sporadic AD models, which contain 27 5mC-, 4 5hmC-, and 8 5fC/caC-specific sites (ST, Fig. 7D, Table 5A-C). Finally, most of our 5mC, 5hmC and 5fC/caC signatures were directly associated with genes involved in neuronal functions and/or have been implicated in Alzheimer’s disease.

Functional classification of the genes that were associated with AD-specific signatures

We identified a set of 27 signature regions that were associated with 26 genes. These genes were then grouped into four categories based on their biological functions: 1) neurodevelopment and neuronal transcription factors, 2) critical cellular processes, 3) RNA and associated proteins (including non-coding RNAs), and 4) cell signaling. In the first group, which was also the largest group (10/26), we identified several genes that have been reported to be involved with AD onset or progression, including NR4A2, HTRA1 , and INHBB. For example, NR4A2, a transcription factor, which was characterized by loss of 5mC in all AD samples was shown to have a neuroprotective role against inflammation, and its loss of expression was associated with neurodegenerative diseases (36). HTRA1 was marked by loss of 5hmC in all AD samples. This gene specifically degrades APOE4, and inhibition of HTRA1 was associated with increased accumulation of Ab plaques (Chu et al., J Am Chem Soc 138, 9473-9478 (2016), Grau et al., Proc Natl Acad Sci U S A 102, 6021 -6026 (2005)). In addition, INHBB, a member of TGFp family, which regulates age-related inflammation processes, was marked by loss of 5fC/caC in AD vs. controls (Fig. 7A-C, Table 4). In the second category (8/26), genes include those involved in DNA repair, inflammation, and mitochondrial functions, all pathways critical in AD pathology. For instance,

ECE1 was marked by loss of 5mC in all AD samples. Lack of or lower expression of ECE1 in mice and humans, respectively, was associated with increased amyloid b production (Eckman et al., J Biol Chem 278, 2081 -2084 (2003), Funalot et al., Mol Psychiatry 9, 1 122-1 128, 1059 (2004)). In the third group (5/26), we identified various noncoding RNAs, including long noncoding RNAs and microRNAs. One of the noncoding RNAs, miR-153, which is downregulated in AD and it targets APP (Long et al., J Biol Chem 287, 31298-31310 (2012)), was marked by loss of 5mC in all AD samples (Fig. 5A, Table 4). In the fourth group (3/26), we identified PTCH1 , which encodes for a protein that is involved in the Hedgehog signaling pathway, was marked by loss of 5mC in all AD samples. Overexpression of PTCH1 in Down syndrome was associated with increased levels of APP/AICD (APP intracellular domain) system (Trazzi et al., J Biol Chem 288, 20817-20829 (2013)) (Fig. 7A, Table 4).

Validation of AD-specific epigenetic signatures in an independent cohort and their clinical application

To demonstrate that our identified epigenetic signatures are a common molecular feature of AD, we validated our signatures in a replication cohort (ROS/MAP cohort, control: n=163, AD: n=371), where DNA methylation profiles were determined using lllumina 450k array (De Jager et al., Nat Neurosci 17, 1156-1 163 (2014), Bennett et al., J Alzheimers Dis 64, S161-S189 (2018)). Samples with Braak&Braak score 0-II were defined as controls, while samples with Braak&Braak score IV-VI were considered as AD.

Notably, 14/18 of the 5mC signature regions were identified in the ROS/MAP cohort. We employed methylation levels of 14 signatures that were present in the ROS/MAP validation cohort as well as the available general clinical features: age, sex, and race for the diagnosis of AD. To determine the beta coefficients and the significance of each signature region we performed logistic regression, which allowed us to model data with binomial distribution. Following corrections for multiple hypothesis testing, 8 genes: ACKR3, ARHGEF16, CASZ1 , ECE1 , KIF26A, MIR153-2, NACC2, NFATC1 were found to be significant in ROS/MAP cohort as well. In addition, clinical features: age and sex, were also significant and highlighted as critical in predicting AD (Table 6).

Finally, to determine overall accuracy of our signatures in predicting AD, we used methylation levels in our signature regions along with the clinical features to calculate Receiver Operating

Characteristic (ROC) curve and Area Under the Curve (AUC) by applying and comparing 4 classification models: support vector machine (SVM), decision tree, logistic regression, and quadratic discriminant analysis (QDA). As shown in Fig. 8, the SVM model, showed the best predictive capacity with an AUC=0.9128 (p<2.2e-16, specificity=98%, sensitivity=48%). This was followed by the predictive ability generated by decision tree with an AUC=0.8235 (p<2.2e-16, specificity=96%, sensitivity=61 %), logistic regression with an AUC=0.7941 (p=4.746e-05, specificity=86%, sensitivity=55%), and QDA with an AUC=0.7905 (p=7.233e-06, specificity=87%, sensitivity=54%). Thus, as a proof-of-concept, the newly identified AD-specific epigenetic signatures in this study can be independently validated in another cohort, and have a tremendous clinical implication in early detection and diagnosis of AD.

To gain insight into the global and dynamic changes of genome-wide distribution of 5mC, 5hmC, and 5fC/caC during directed differentiation of iPSCs to neuronal precursor cells (NPC) and neurons (N), we first identified the total number of the modified cytosines in each respective sample. In WT cells, we identified 13637104, 15049856, and 21918240 sites of methylated cytosine (5mC) at iPSC, NPC, and N stages, respectively. By comparative analysis, we observed a 16% and 34% increase of 5mC sites in NPCs and Ns, respectively, compared to iPSCs (Table 1). For 5hmC, we identified 7041346, 7364564, and 10485793 sites in WT iPSCs, NPCs, and Ns, respectively. Changes in 5hmC sites were marked by a 28% decrease in NPCs compared to iPSCs. In contrast, we observed a gain of 5hmC sites as cells differentiate from NPCs to Ns, reaching to similar numbers to that of the iPSC cells (Table 1). For 5fC/caC, we identified 2586021 , 1797954, and 2307023 sites in WT iPSC, NPC, and N, respectively. The number of 5fC/caC sites was reduced by 10% and 21 % in NPCs and neurons, respectively, compared to iPSCs (Table 1). Identification of AD-specific 5mC, 5hmC, and 5fC/caC signature sites

To narrow down our regional signatures to individual CpG sites, we first identified the total number of CpG sites that were present in these final signature regions. For the 19 regional signatures in the 5mC groups, in total we identified 355 CpGs, which were then sorted into loss of 5mC CpGs (74) and gain of 5mC CpGs (8) in AD vs. control samples. To avoid background signal and false positives, we applied a final cutoff of methylation fold change difference of <-0.1 and >0.1 , which resulted in 21 loss of 5mC CpGs and 6 gain of methylation CpGs in AD compared to controls (Fig. 7D). In the final 5hmC regional signatures, we identified 220 total CpG sites, where 14 and 9 CpG sites were marked by loss and gain of 5hmC, respectively, in AD compared to controls. Similar to the 5mC final CpG sites, we applied the final cutoff of methylation fold change difference of <-0.1 and >0.1 , resulting in 3 loss of 5hmC CpGs and 1 gain of 5hmC CpG site in AD vs. normal (Fig. 7D). Using the same approach as for 5mC and 5hmC, we identified a total of 1 1 CpG sites in 5fC/caC signature region, with a final 4 gain and 4 loss of 5fC/caC CpG sites in AD compared to controls (Fig. 7D, Table 5).

Validation of AD-epiqenetic signatures in an independent sample set

To validate our 5mC AD signatures we employed previously published datasets (De Jager et al., Nat Neurosci 17, 1 156-1 163 (2014)). We must note that in the previous study, the authors used bisulfite sequencing that cannot distinguish between 5mC and 5hmC. In addition, the authors used the lllumina 450k array that only covers the coding part of the genome, therefore, it is possible that some of our signature regions might not be present in the ROS/MAP cohort. In our analyses, we employed whole- genome OXBS sequencing that allowed us to identify separately an average of 17,176,031 5mC and 8,479,266 5hmC individual sites. Thus, the lllumina 450k array covers only ~1 .6% of our combined 5mC+5hmC sites.

Materials and Methods

The following material and methods were employed in arriving at the above-described results.

Cell lines and tissue samples

Normal and AD patient-derived iPS cells, neural progenitor cells, and cortical neuronal cells were obtained from Axol Biosciences (Cambridge, UK). Control or disease iPS lines were generated using episomal vector reprogramming of somatic cells (newborn, male). In addition to control lines, in our study, we employed iPS cells carrying mutation L286V in PSEN1 and mutation N141 I in PSEN2, both of which are associated with EOAD. Age of patients when the skin cells were harvested for reprogramming was 38 years (female) and 81 years (female) for PSEN1 and PSEN2 lines, respectively. The iPS cells derived from a LOAD patient carrying homozygous APOE4 were generated by reprogramming fibroblasts harvested from a female patient at the age of 87 years. Directed differentiation of iPS cells to cortical neurons was performed as described previously (Shi et al., Nat Protoc 7, 1836-1846 (2012)).

Characterization of iPS cells, neural progenitor cells, and cortical neurons in normal and AD patient- derived lines was performed at Axol Biosciences (Cambridge, UK). DNA samples originating from brain tissues of healthy (60 years, female), AD-frontal lobe region (87 years, male), and AD-left frontal lobe region (75 years, male) were purchased from Biochain (CA, USA). All samples were collected and processed according to protocols approved by the Brigham and Women’s Hospital. Oxidative bisulfite and methylase-assisted sequencing at single-base pair resolution

To map and quantify the levels of 5mC and 5hmC genome-wide at base resolution we employed OXBS-seq in our samples using TrueMethyl Seq according to the recommended protocol by Cambridge Epigenetix (CEGX, Cambridge, UK) (Booth et al„ Nat Protoc 8, 1841 -1851 (2013)). Briefly, 1 ug of DNA extracted following phenol-chlorophorm protocol was spiked-in with synthetic DNA sequencing controls, and then DNA was sheared to ~800bp using Covaris E220 sonicator. Following that, DNA samples were split into two halves for the library module: ~500ng for bisulfite conversion and ~500ng for oxidative bisulfite conversion, generating in parallel two libraries for each sample, enabling us to distinguish at base resolution 5mC and 5hmC. MAB-seq was performed according to the protocol described previously by Neri et al. (Neri et al., Cell Rep, (2015)). Briefly, 1 pg of DNA was methylated using M.Sssl (NEB, MA, USA) and then sheared to 350bp using Covaris M220 sonicator. DNA libraries were prepared according to the lllumina protocol (TruSeq DNA PCR-Free Library Preparation Kit, CA, USA). Adaptor ligated libraries were treated with M.Sssl (NEB, MA, USA) to methylate bases introduced through end repair, bisulfite converted using EpiTect bisulfite kit (Qiagen, CA, USA), and then amplified using KAPA HiFi Hotstart Uracil+ Readymix (KAPA Biosystems, MA, USA). Samples were sequenced using lllumina HiSeq X Ten platform, generating at least 100GB/sample.

RNA isolation, reverse transcription , and real time quantitative RT-PCR

RNA was isolated using TRIzol reagent following the recommended protocol (Invitrogen, CA, USA). Reverse transcription was performed using iScript reverse transcription reagents from Bio-Rad (CA, USA). Real time qRT-PCR was performed with iQ SYBR green super mix (Bio-Rad, CA, USA).

Primer sequences for NES, MAP2, PAX6, SOX2 genes were described previously (Shi et al., Nat Protoc 7, 1836-1846 (2012), Giorgetti et al., Proc Natl Acad Sci U S A 109, 12556-12561 (2012), Garner et al., PLoS One 10, e0145052 (2015)). The rest of the primer sequences are listed in Table 7.

Immunofluorescence staining

Cells grown on sterile glass cover slips were fixed with 3.7% paraformaldehyde in PBS for 20 min and washed extensively with PBS. Samples were permeabilized with 0.2% Triton-X in PBS for 20 min and blocked with 5% goat serum (or 3% BSA) in PBS for 45 min. Cells were incubated with primary antibodies for 1 h at room temperature. IgG was used as negative control. After extensive washing with PBS, samples were incubated with respective secondary antibodies. Nuclei were stained with DAPI for 10 min and samples were mounted using Fluoromount-G (Southern Biotech, AL, USA). Whole-genome sequencing data quality

To assess quality of BS/OXBS-Seq data, spike-in DNA controls were mixed with genomic DNA samples. We used CEGX_QC package with default parameters to detect the conversion rate of 5mC and 5hmC. The conversion efficiency was determined based on synthetic DNA controls from CEGX and was considered successful if 090% for BS and OXBS libraries, while hmC<5% in BS libraries and >80% in OXBS libraries (Table 8A). Similarly, to determine the methylation efficiency of M.Sssl in MAB-seq, we spiked-in unmethylated CG lambda DNA (Promega, Wl, USA) and observed almost complete methylation of unmethylated CGs (~97%) and high bisulfite conversion efficiency (~97%) (Table 8B). We randomly selected 30 million reads from each trimmed sequence files, mapped them onto lambda genome using bsmap, and extracted methylation signals using command“methratio.py -t 2 -r -m 5 -z.” The M.Sssl enables methylation of Cs with high efficiency in CG context, instead of CH context, while the bisulfite treatment converts unmethylated Cs into Ts in both CG or CH contexts. Therefore, the M.Sssl methylase efficiency was calculated by the methylated-CG % based on cytosine methylation level in CG context, while bisulfite conversion efficiency was identified by unmethylated-CH % based on lack of cytosine methylation in the CH context. Owning to the fact that MAB-seq allows to confidentially determine

5fC/caC only in CG context (Wu et al.,Nat Biotechnol 32, 1231 -1240 (2014)) and that 5hmC is almost exclusively located in CG dinucleotides (Yu et al., Nat Protoc 7, 2159-2170 (2012)), we focused our studies of 5mC and 5hmC in CG sites in order to generate comparable and comprehensive genome-wide maps for all three cytosine modifications.

Table 8A. Summary of whole-genome OXBS-sequencing experiments

Table 8B. Summary of whole-genome MAB-sequencing experiments

Identification of 5mC, 5hmC, and 5fC/5caC sites and regions

Raw sequencing data were trimmed using Trim Galore (v0.4.0) to remove low quality bases and adaptor sequences. Trimmed reads were mapped onto the reference genome (hg19 for samples of human origin and mm9 for samples of mouse origin) using bsmap (v2.74) (Xi et al., BMC Bioinformatics 10, 232 (2009)), followed by removal of PCR duplicates. Methylation signals were extracted using methratio.py, a script in the bsmap package (v3.4.2) (Song et al., PLoS One 8, e81 148 (2013)). An R- script was generated to maintain chromosome coordinates consistent for methylation sites between each pair of BS and OXBS samples, allowing to identify 5mC and 5hmC signal in these common sites with the mlml script in methpipe using the default settings. To calculate 5fC/5caC sites, we used the following formula: 5fC/5caC=NT/(NC+NT), where NT and NC are the number of Ts and Cs in CG context, respectively. M.Sssl methylase error rate and bisulfite conversion inefficiency was 1 .64%, which was determined in lambda genome in our pilot experiment in mESC WT and mESC Tdg KO cell lines (Table 8B). Error rate was used in binomial distribution and q-value package to adjust for the 5fC/5caC signal. The number of 5fC/5caC sites was determined by a binomial test as described previously (Yu et al., Cell 149, 1368-1380 (2012)). Only 5fC/5caC sites with cutoff of coverage>10, p-value<0.01 , and FDR<0.01 were kept for the downstream analyses. It is possible that our samples may carry mutations different from reference genome, which could result in higher false positives. To address this issue, we used biscuit package to identify these mutations from trimmed mapped reads and removed these potential mutations from 5mC, 5hmC, and 5fC/5caC sites. The remaining sites were used for the downstream analysis.

To call for DMRs, DHMRs, and DFCRs between the two groups of methylomes, the methpipe package was used (Song et al., PLoS One 8, e81 148 (2013)). We started by assembling a proportion table containing read proportions for all target methylomes with‘merge-methcounts’ program included in methpipe. After creating the proportion table and specifying the design matrix, we performed differential methylation analyses, which consists of 1) regression (‘radmeth regression’ default parameters), 2) combining significance, and 3) multiple testing adjustment steps (‘radmeth adjust -bins 1 :200:1’). All DMRs/DHMRs/DFCRs were filtered by p-value<0.01 and contain at least 5 differentially methylated CG sites in each region. To call for DMRs, DHMRs, and DFCRs between a pair of methylomes, we generated an R script that enables to use 200bp-step-size across entire genome with 2kb-bin and calculate the methylation difference along with p-value (student’s t-test) in both samples. If adjacent bins had continual methylation differences between samples, which were identified with cutoff of fold-change>2 and p- value<0.05, these bins were iteratively merged together, and methylation difference was calculated for the merged region.

Data normalization

To allow comparison of 5mC, 5hmC, and 5fC/caC signals among samples, we normalized 5mC, 5hmC, and 5fC/caC signals across samples using non-overlapping 1 kb bins spanning the whole-genome. For 5mC and 5hmC, the signal in each bin was represented by TNC/(TNC+TNT), where TNT and TNC are the total number of Ts and Cs within a bin, respectively. For 5fC/5caC, the signal in each bin was calculated with average signal within the bin, which was due to the lower density of 5fC/5caC sites across whole-genome compared to 5mC or 5hmC sites. Finally, the signals of 5mC, 5hmC, and 5fC/5caC were smoothed over whole-genome, which allows to compare the signal differences of same type modification between samples.

Pathway analysis

We used the clusterProfiler package in Bioconductor to perform enrichment analysis for identified genes (Yu et al., OMICS 16, 284-287 (2012)). The script enrichGO was used to perform Gene Ontology (GO) pathway enrichment analysis. The enrichGO calculates p-values using the hypergeometric distribution.

ROC, specificity, and sensitivity calculation

To determine the beta coefficients and the significance of each signature in predicting AD, we performed logistic regression using R function glm(family = binomial(link = "logit")), in which the parameters allowed us to model data with binomial distribution. Half of the data were used as training dataset and the other half as test dataset. To evaluate the overall accuracy of our signature regions we calculated Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC) by applying and comparing 4 classification models: Support Vector Machine (SVM), decision tree, logistic regression, and quadratic discriminant analysis (QDA). P-values were calculated with exact binomial test.

All publications and patents mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication or patent was specifically and individually indicated to be incorporated by reference.

From the foregoing description, one skilled in the art can easily ascertain the essential characteristics of this invention, can make various changes and modifications of the invention to adapt it to various usages and conditions. Thus, other embodiments are also within the claims.

Claims

CLAIMS What is claimed is:

1. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising measuring the quantity of 5-methyl-cytosine (5mC) nucleobases in one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , and ARHGEF16 in a neuron obtained from the subject, wherein a finding that the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease.

2. The method of claim 1 , wherein the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or more genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

3. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising measuring the quantity of 5mC nucleobases in one or both of genes LINC02055 and KHDRBS3 in a neuron obtained from the subject, wherein a finding that the quantity of 5mC nucleobases in the one or both genes is significantly greater than a reference level of 5mC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease.

4. The method of claim 3, wherein the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or both genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

5. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising measuring the quantity of 5-hydroxymethyl-cytosine (5hmC) nucleobases in one or more of genes H1 F00, UNCX, HTRA1 , and C15orf62 in a neuron obtained from the subject, wherein a finding that the quantity of 5hmC nucleobases in the one or more genes is significantly less than a reference level of 5hmC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease.

6. The method of claim 5, wherein the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the one or more genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

7. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising measuring the quantity of 5hmC nucleobases in gene 2K4 in a neuron obtained from the subject, wherein a finding that the quantity of 5hmC nucleobases in the gene is significantly greater than a reference level of 5hmC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease.

8. The method of claim 7, wherein the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the gene as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

9. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising measuring the quantity of 5-formyl-cytosine (5fC) or 5-carboxy-cytosine (5caC) nucleobases in one or both of genes INHBB and HLA-A in a neuron obtained from the subject, wherein a finding that the quantity of 5fC or 5caC nucleobases in the one or both genes is significantly less than a reference level of 5fC or 5caC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease.

10. The method of claim 9, wherein the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the one or both genes as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

11 . A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising measuring the quantity of 5fC or 5caC nucleobases in gene MIR4532 in a neuron obtained from the subject, wherein a finding that the quantity of 5fC or 5caC nucleobases in the gene is significantly greater than a reference level of 5fC or 5caC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease.

12. The method of claim 11 , wherein the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the gene as measured in a neuron obtained from a human subject that does not have Alzheimer’s disease.

13. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

wherein a finding that the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease.

14. The method of claim 13, wherein the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or more genes as measured in neurons that are obtained by differentiating neural progenitor cells (NPCs) derived from a human subject that does not have Alzheimer’s disease.

15. The method of claim 14, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

16. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

wherein a finding that the quantity of 5mC nucleobases in the one or both genes is significantly greater than a reference level of 5mC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease.

17. The method of claim 16, wherein the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or both genes as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease.

18. The method of claim 17, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

19. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

wherein a finding that the quantity of 5hmC nucleobases in the one or more genes is significantly less than a reference level of 5hmC nucleobases in the one or more genes identifies the subject as one that will develop Alzheimer’s disease.

20. The method of claim 19, wherein the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the one or more genes as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease.

21 . The method of claim 20, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

22. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5hmC nucleobases in gene 2K4 in the neurons;

wherein a finding that the quantity of 5hmC nucleobases in the gene is significantly greater than a reference level of 5hmC nucleobases in the gene identifies the subject as one that will develop

Alzheimer’s disease.

23. The method of claim 22, wherein the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the gene as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease.

24. The method of claim 23, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

25. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

a. reprogramming a population of somatic cells obtained from the subject into iPSCs; and b. differentiating the iPSCs into neurons and measuring the quantity of 5fC or 5caC

nucleobases in one or both of genes INHBB and HLA-A in the neurons; wherein a finding that the quantity of 5fC or 5caC nucleobases in the one or both genes is significantly less than a reference level of 5fC or 5caC nucleobases in the one or both genes identifies the subject as one that will develop Alzheimer’s disease.

26. The method of claim 25, wherein the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the one or both genes as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease.

27. The method of claim 26, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

28. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

nucleobases in gene MIR4532 in the neurons;

wherein a finding that the quantity of 5fC or 5caC nucleobases in the gene is significantly greater than a reference level of 5fC or 5caC nucleobases in the gene identifies the subject as one that will develop Alzheimer’s disease.

29. The method of claim 28, wherein the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the gene as measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease.

30. The method of claim 29, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

31 . The method of any one of claims 13-30, wherein the somatic cells are reprogrammed into iPSCs by transfecting the somatic cells with one or more of genes OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC.

32. The method of claim 31 , wherein the somatic cells are transfected by electroporation in the presence of one or more vectors that together encode OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC.

33. The method of any one of claims 13-32, wherein the iPSCs are differentiated into neurons by

(i) differentiating the iPSCs into NPCs by contacting the iPSCs with mouse embryonic fibroblasts (MEFs), a rho kinase inhibitor, and fibroblast growth factor 2 (FGF2), and subsequently (ii) differentiating the NPCs into neurons.

34. The method of claim 33, wherein the rho kinase inhibitor is Y-27632.

35. The method of claim 33 or 34, wherein the NPCs are differentiated into neurons by contacting the NPCs with an inhibitor of bone morphogenetic protein (BMP) signaling and a transforming growth factor b (TGF-b) receptor inhibitor.

36. The method of claim 35, wherein the inhibitor of BMP signaling is noggin or dorsomorphin.

37. The method of claim 35 or 36, wherein the TGF-b receptor inhibitor is SB431542.

38. The method of any one of claims 1 -8, 13-24, and 31 -37, wherein 5mC or 5hmC nucleobase modifications are measured by way of oxidative bisulfite sequencing.

39. The method of any one of claims 9-12 and 25-37, wherein 5fC or 5caC nucleobase modifications are measured by way of methylase-assisted bisulfite sequencing.

40. The method of any one of claims 1 -39, wherein the 5mC, 5hmC, 5fC, or 5caC nucleobase is measured at a chromosomal site set forth in any one of Tables 3A - 3C, 4A - 4C, and 5A - 5C.

41 . A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

c. differentiating the NPCs into neurons and measuring the quantity of 5fC or 5caC nucleobases in the neurons;

wherein a finding that the quantity of 5fC or 5caC nucleobases in the NPCs is not significantly greater than (i) the quantity of 5fC or 5caC nucleobases in the iPSCs and (ii) the quantity of 5fC or 5caC nucleobases in the neurons identifies the subject as one that will develop Alzheimer’s disease.

42. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

a. reprogramming a population of somatic cells obtained from the subject into iPSCs and measuring the quantity of 5hmC nucleobases in the iPSCs;

b. differentiating the iPSCs into NPCs and measuring the quantity of 5hmC nucleobases in the NPCs; and

c. differentiating the NPCs into neurons and measuring the quantity of 5hmC nucleobases in the neurons;

wherein a finding that the quantities of 5hmC nucleobases in the iPSCs, the NPCs, and the neurons do not significantly differ from one another identifies the subject as one that will develop Alzheimer’s disease.

43. A method of determining whether a human subject will develop Alzheimer’s disease, the method comprising:

c. differentiating the NPCs into neurons and measuring the quantity of 5mC or 5hmC

nucleobases in the neurons;

wherein:

(i) a finding that the quantity of 5mC or 5hmC nucleobases in the iPSCs is significantly greater than an iPSC reference level of 5mC or 5hmC nucleobases identifies the subject as one that will develop Alzheimer’s disease; and/or

(ii) a finding that the quantity of 5mC or 5hmC nucleobases in the NPCs is significantly lower than a NPC reference level of 5mC or 5hmC nucleobases identifies the subject as one that will develop Alzheimer’s disease; and/or (iii) a finding that the quantity of 5mC or 5hmC nucleobases in the neurons is significantly greater than a neuronal reference level of 5mC or 5hmC nucleobases identifies the subject as one that will develop Alzheimer’s disease.

44. The method of claim 43, wherein the iPSC reference level of 5mC or 5hmC nucleobases is a quantity of 5mC or 5hmC nucleobases measured in iPSCs that are obtained by reprogramming somatic cells from a human subject that does not have Alzheimer’s disease.

45. The method of claim 43 or 44, wherein the NPC reference level of 5mC or 5hmC nucleobases is a quantity of 5mC or 5hmC nucleobases measured in NPCs derived from a human subject that does not have Alzheimer’s disease.

46. The method of any one of claims 43-45, wherein the neuronal reference level of 5mC or 5hmC nucleobases is a quantity of 5mC or 5hmC nucleobases measured in neurons that are obtained by differentiating NPCs derived from a human subject that does not have Alzheimer’s disease.

47. The method of claim 45 or 46, wherein the NPCs derived from the human subject that does not have Alzheimer’s disease are obtained by differentiating iPSCs that are obtained by reprogramming somatic cells from the human subject that does not have Alzheimer’s disease.

48. The method of any one of claims 41 -47, wherein the somatic cells are reprogrammed into iPSCs by transfecting the somatic cells with one or more of genes OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC.

49. The method of claim 48, wherein the somatic cells are transfected by electroporation in the presence of one or more vectors that together encode OCT4, SOX2, NANOG, LIN28, KLF4, and L-MYC.

50. The method of any one of claims 41 -49, wherein the iPSCs are differentiated into NPCs by contacting the iPSCs with MEFs, a rho kinase inhibitor, and FGF2.

51 . The method of claim 50, wherein the rho kinase inhibitor is Y-27632.

52. The method of any one of claims 41 -51 , wherein the NPCs are differentiated into neurons by contacting the NPCs with an inhibitor of BMP signaling and a TGF-b receptor inhibitor.

53. The method of claim 52, wherein the inhibitor of BMP signaling is noggin or dorsomorphin.

54. The method of claim 52 or 53, wherein the TGF-b receptor inhibitor is SB431542.

55. The method of any one of claims 41 -54, wherein the Alzheimer’s disease is early-onset Alzheimer’s disease, late-onset Alzheimer’s disease, familial Alzheimer’s disease, or sporadic

Alzheimer’s disease.

56. The method of any one of claims 41 and 48-55, wherein 5fC or 5caC nucleobase modifications are measured by way of methylase-assisted bisulfite sequencing.

57. The method of any one of claims 42-56, wherein 5mC or 5hmC nucleobase modifications are measured by way of oxidative bisulfite sequencing.

58. A kit comprising a bisulfite salt and a plurality of nucleic acid primers suitable for amplification of one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , ARHGEF16, LINC02055 KHDRBS3, H1 F00, UNCX, HTRA1 , C15orf62, 2K4, INHBB, HLA-A, and MIR4532.

59. The kit of claim 58, wherein the kit further comprises a DNA polymerase.

60. The kit of claim 58 or 59, wherein the kit comprises a package insert instructing a user to perform the method of any one of claims 1 -57.

61 . A kit comprising a CpG methyltransferase and a plurality of nucleic acid primers suitable for amplification of one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , ARHGEF16, LINC02055 KHDRBS3, H1 F00, UNCX, HTRA1 , C15orf62, 2K4, INHBB, HLA-A, and MIR4532.

62. The kit of claim 61 , wherein the CpG methyltransferase is M.Sssl methyltransferase.

63. The kit of claim 61 or 62, wherein the kit comprises a package insert instructing a user to perform the method of any one of claims 1 -57.

64. The kit of any one of claims 58-63, wherein the kit further comprises a panel of one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , ARHGEF16, LINC02055 KHDRBS3, H1 F00, UNCX, HTRA1 , C15orf62, 2K4, INHBB, HLA-A, and MIR4532 obtained from a human subject that does not have Alzheimer’s disease.

65. A neuron comprising one or more of genes ADA2, PRKACA, NFIX, NFATC1 , GSE1 , KIF26A, NACC2, FTCH1 , MIR153-2, PKHD1 , PCDHA2, ACKR3, NR4A2, ECE1 , CASZ1 , and ARHGEF16, wherein the quantity of 5mC nucleobases in the one or more genes is significantly less than a reference level of 5mC nucleobases in the one or more genes.

66. The neuron of claim 65, wherein the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or more genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease.

67. The neuron of claim 66, wherein the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

68. A neuron comprising one or both of genes LINC02055 and KHDRBS3, wherein the quantity of 5mC nucleobases in the one or both genes is significantly greater than a reference level of 5mC nucleobases in the one or both genes.

69. The neuron of claim 68, wherein the reference level of 5mC nucleobases is a quantity of 5mC nucleobases in the one or both genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease.

70. The neuron of claim 69, wherein the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

71 . A neuron comprising one or more of genes H1 F00, UNCX, HTRA1 , and C15orf62, wherein the quantity of 5hmC nucleobases in the one or more genes is significantly less than a reference level of 5hmC nucleobases in the one or more genes.

72. The neuron of claim 71 , wherein the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the one or more genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease.

73. The neuron of claim 72, wherein the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

74. A neuron comprising gene 2K4, wherein the quantity of 5hmC nucleobases in the gene is significantly greater than a reference level of 5hmC nucleobases in the gene.

75. The neuron of claim 74, wherein the reference level of 5hmC nucleobases is a quantity of 5hmC nucleobases in the gene as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease.

76. The neuron of claim 75, wherein the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

77. A neuron comprising one or both of genes INHBB and HLA-A, wherein the quantity of 5fC or 5caC nucleobases in the one or both genes is significantly less than a reference level of 5fC or 5caC nucleobases.

78. The neuron of claim 77, wherein the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the one or both genes as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease.

79. The neuron of claim 78, wherein the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

80. A neuron comprising gene MIR4532, wherein the quantity of 5fC or 5caC nucleobases in the gene is significantly greater than a reference level of 5fC or 5caC nucleobases in the gene.

81 . The neuron of claim 80, wherein the reference level of 5fC or 5caC nucleobases is a quantity of 5fC or 5caC nucleobases in the gene as measured in a neuron obtained by differentiating an NPC derived from a human subject that does not have Alzheimer’s disease.

82. The neuron of claim 81 , wherein the NPC derived from the human subject that does not have Alzheimer’s disease is obtained by differentiating an iPSC obtained by reprogramming a somatic cell from the human subject that does not have Alzheimer’s disease.

83. The neuron of any one of claims 65-82, wherein the neuron is a human neuron.