Abstract

   Human middle temporal gyrus (MTG) is a vulnerable brain region in early
   Alzheimer’s disease (AD), but little is known about the molecular
   mechanisms underlying this regional vulnerability. Here we utilize the
   10 × Visium platform to define the spatial transcriptomic profile in
   both AD and control (CT) MTG. We identify unique marker genes for
   cortical layers and the white matter, and layer-specific differentially
   expressed genes (DEGs) in human AD compared to CT. Deconvolution of the
   Visium spots showcases the significant difference in particular cell
   types among cortical layers and the white matter. Gene co-expression
   analyses reveal eight gene modules, four of which have significantly
   altered co-expression patterns in the presence of AD pathology. The
   co-expression patterns of hub genes and enriched pathways in the
   presence of AD pathology indicate an important role of
   cell–cell-communications among microglia, oligodendrocytes, astrocytes,
   and neurons, which may contribute to the cellular and regional
   vulnerability in early AD. Using single-molecule fluorescent in situ
   hybridization, we validated the cell-type-specific expression of three
   novel DEGs (e.g., KIF5A, PAQR6, and SLC1A3) and eleven previously
   reported DEGs associated with AD pathology (i.e., amyloid beta plaques
   and intraneuronal neurofibrillary tangles or neuropil threads) at the
   single cell level. Our results may contribute to the understanding of
   the complex architecture and neuronal and glial response to AD
   pathology of this vulnerable brain region.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s40478-022-01494-6.

   Keywords: Spatially resolved transcriptomics, Alzheimer’s disease,
   Vulnerability, Human middle temporal gyrus, Microglia,
   Oligodendrocytes, Astrocytes, Neurons, Weighted gene co-expression
   network analyses (WGCNA), Single-molecule fluorescent in situ
   hybridization

Introduction

   Alzheimer’s disease (AD) is the most common form of dementia in the
   elderly, affecting 40–50 million people worldwide [[63]1]. AD is
   pathologically characterized by extracellular amyloid beta (Aβ)
   plaques, neurofibrillary tangles (neuronal tau aggregates), gliosis
   induced by activated microglia and reactive astrocytes, and white
   matter (WM) degeneration possibly induced by dysfunctions in
   oligodendrocytes [[64]2–[65]6]. The distribution pattern of AD
   pathology from regions like the medial temporal lobe to the cortex has
   been extensively studied [[66]7–[67]10]. However, the molecular
   mechanisms underlying the cell- and region-specific distribution of AD
   pathology at early stages of the disease are still under-investigated.
   The transcriptome of the AD brain can pinpoint key differences in
   disease that may be crucial for elucidating the pathogenesis of AD and
   for developing disease-modifying therapeutics for the prevention and
   treatment of AD [[68]11].

   Compared with single-cell or single-nucleus (sn) RNA-Sequencing
   (RNA-Seq) techniques, the recent advent of spatially resolved
   transcriptomics (SRT) (e.g., 10 × Genomics Visium) allows us to compare
   transcriptomic profiles between AD and CT subjects without the loss of
   spatial information [[69]12–[70]17]. This novel technique enables a
   better understanding of molecular mechanisms underlying the
   neuropathology of AD within the spatial context. Recent applications of
   SRT on AD-like mouse models have revealed a plaque-induced gene (PIG)
   network, an oligodendrocyte gene (OLIG) response, and regionally
   differentially expressed genes (DEGs) in AD-like mice compared to
   controls [[71]16, [72]17]. However, comprehensive SRT profiling of
   human AD brain tissue and characterization of DEGs associated with
   variations of AD pathology in vulnerable cortical layers, have not been
   reported as far as we know.

   To fill this knowledge gap, we used the 10 × Genomics Visium platform
   combined with co-immunofluorescence staining of AD-associated
   pathological markers to define the spatial topography of gene
   expression in the human middle temporal gyrus (MTG), a vulnerable brain
   region in early AD [[73]18]. In this study, we investigated the MTG
   from human postmortem CT (Braak stages I-II, minimal tau pathology) and
   AD cases (Braak stages III-IV, moderate tau pathology) [[74]8]. To
   highlight gene changes associated with early stages of the disease and
   exclude most downstream gene expression changes that occur in late AD,
   we did not include AD cases at Braak stages V-VI in this study. We
   annotated the cortical layers and the WM of AD and CT MTG, identified
   specific marker genes for five cortical layers and the WM, and
   identified layer-specific DEGs in human AD compared to CT. Weighted
   gene co-expression network analyses (WGCNA) [[75]19] of 10,000 highly
   variable genes in both AD and CT cases revealed eight co-expression
   gene modules. The co-expression pattern of four gene modules
   dramatically changed in the presence of Aβ and/or tau pathology,
   implicating an important role of varying cell types and their
   co-expression in health and disease. Furthermore, we quantitated the
   expression of novel DEGs associated with AD pathology in microglia,
   astrocytes, neurons, and oligodendrocytes using RNAscope HiPlex
   single-molecule fluorescent in situ hybridization (smFISH) assay. We
   further validated our study and techniques by quantitating the
   expression of previously published DEGs [[76]12–[77]15, [78]20–[79]25].
   As far as we know, our study provides first-of-its-kind SRT profiling
   of human MTG and a unique spatial view of transcriptional alterations
   associated with two major AD hallmarks, Aβ plaques and pathological
   tau. Furthermore, this analysis reveals gene perturbations specific to
   each layer and AD pathology, as well as shared gene-expression
   perturbations, thus providing additional molecular insights into the
   regional vulnerability and pathogenesis of AD.

Methods

Human postmortem brain tissues

   Human fresh frozen brain blocks were provided by the Arizona Study of
   Aging and Neurodegenerative Disorders/Brain and Body Donation Program
   at Banner Sun Health Research Institute [[80]26], the New York Brain
   Bank at Columbia University Irving Medical Center [[81]27], and the
   Brain Bank & Biorepository at Ohio State University Wexner Medical
   Center. The demographics of human cases used in this study are listed
   in Additional file [82]2: Table S1. These specimens were obtained by
   consent at autopsy and have been de-identified and are IRB exempt to
   protect the identity of each patient. Frozen sections (10 μm) were cut
   from frozen blocks under RNase-free conditions.

Reagents

   Human/murine phospho-tau pSer202/Thr205 (AT8, Cat# MN1020), Tau46 (Cat#
   13-6400), Alexa Fluor dye-labeled cross-absorbed donkey secondary
   antibodies, and TRIzol RNA isolation reagent (Cat# 15596026) were
   purchased from ThermoFisher Scientific. Goat anti-Olig2 (Cat# AF2418)
   and GAD1 (Cat# AF2086) polyclonal antibody were purchased from R&D
   Systems. Rabbit anti-GFAP (Cat# G9269), WFS1 (Cat# 1158-1-AP), and Aβ
   (Cat# Ab2539) polyclonal antibodies were purchased from Sigma-Aldrich,
   Proteintech, Abcam, and DAKO, respectively. Rat anti-P2RY12 (Cat#
   848002) was purchased from BioLegend. RNAscope HiPlex Ancillary Kit
   (Cat# 324120) and human-specific RNA probes were purchased from
   Advanced Cell Diagnostics. TrueBlack lipofuscin autofluorescence
   quencher (Cat# 23007) was purchased from Biotium. Fluoromount-G
   Mounting Medium (Cat# 0100-01) was purchased from SouthernBiotech.

Sample selection and preparation for Visium

   Thirty milligrams of brain tissue were homogenized in 500 µl TRIzol RNA
   isolation reagent. RNA was extracted from each homogenate by following
   the TRIzol RNA extraction procedure. The RNA quality of each sample was
   assessed by RNA integrity number (RIN) via Agilent 2200 TapeStation
   system. The 10 µm sections from the human MTG were mounted within
   6.5 mm × 6.5 mm capture areas, which were subjected to the Visium gene
   expression assay. All samples passed the quality control for cDNA, post
   library construction, and sequencing set by 10 × Visium Spatial Gene
   Expression Reagent Kits User Guide (Additional file [83]3: Table S2).

   The time course of the tissue permeabilization for each sample was
   determined by 10 × Genomics Visium Spatial Tissue Optimization (STO)
   Kit (Slide Kit, Part# 1000191. Reagent Kit, Part# 1000192). Briefly,
   sections from selected square areas of the brain tissue were mounted on
   the capture areas of STO slides. Sections were fixed in the −20 °C
   prechilled methanol for 30 min and subjected to H&E staining. The
   20 × tile image for each sample was obtained by Zeiss Axio Observer
   microscope. After imaging, sections were permeabilized with
   permeabilization enzyme for varying amounts of time, and the reverse
   transcription (RT) was performed directly on the slide using
   Cy3-nucleotides. After RT, sections were imaged and aligned with H&E
   images. The best time course was chosen for Vsisum experiment if its
   image shows the strongest Cy3 signal and the minimum signal diffusion.

Visium SRT processing

   Each optimized human postmortem brain tissue was cryosectioned at 10 µm
   continuously for five sections and was labeled sequentially as No. 1 to
   No. 5. The middle section (No. 3, Gene Expression (GE) section) was
   mounted on the Visium GE slide for Visium profiling, sections No. 1, 2,
   4, 5 were used for IF staining of cell-type markers and AD pathology.
   SRT was performed using the 10 × Genomics Visium Spatial Gene
   Expression Kit (Slide Kit, Part# 1000188. Reagent Kit, Par# 1000189).
   The procedures prior to RT are the same as described for STO. For RT,
   cDNA was synthesized using nucleotides without the label of Cy3. By
   using Template Switch Oligo, the second strand cDNA (ss cDNA) was
   synthesized according to cDNA templates captured on the poly T probes.
   The ss cDNA was then denatured by 0.08 M KOH, washed off, and then
   amplified using PCR. The cDNA quality control was performed using an
   Agilent Bioanalyzer high sensitivity chip. After the cDNA concentration
   was determined, the sequencing library of each sample was constructed
   using the 10 × Library Construction kit (Part# 1000196). Briefly,
   optimized cDNA was obtained by enzymatic fragmentation and size
   selection. P5, P7, i7, and i5 sample indexes and TruSeq Read 2 were
   added via End Repair, A-tailing, Adaptor Ligation, and PCR. The cDNA
   library with correct sizes was then selected using the SPRIselect
   reagent (Beckman Coulter, Part# [84]B23318). To meet the required
   sequencing depth, cDNA libraries from four samples were pulled in a
   NovaSeq6000 SP v1.0 flowcell and paired-end sequencing was performed on
   an Illumina NovaSeq6000 sequencer at the Genomics Services Laboratory
   at Nationwide Children’s Hospital and the AGCT core at Cedars-Sinai
   Cancer.

IF staining

   As described above, sections No. 2 and 4, and 1 and 5 were 0 µm, and
   10 µm, apart from the GE section, respectively. Section No. 2 was
   sequentially stained with P2RY12/GFAP/AT8, section No. 4 with
   Olig2/Tau46/Aβ, section No. 1 with WFS1/AT8, and section No. 5 with
   GAD1. IF staining was performed as previously described [[85]14]. All
   sections were air dried at 60 °C for 10 min and then fixed and
   permeabilized by prechilled acetone at −20 °C for 15 min. For sections
   No. 2 and 4, slides were immersed in 1 × PBS for 3 h at 37 °C for
   antigen retrieval. For sections No. 1 and 5, antigen retrieval was
   performed by incubating slides in 10 mM sodium citrate (pH6.0, 95 °C)
   for 12 min. Antibodies were incubated sequentially to avoid
   false-positive results coming from co-immunostaining. On day 1, after
   1-h blocking by 10% donkey serum (in 1 × PBS), P2RY12 (1:1000) and
   Olig2 (1:500) primary antibodies were applied on section No. 2 and 4,
   respectively, overnight at 4 °C, while GAD1 (1:100 in PBS with 10%
   donkey serum) antibody was applied to sections No. 1 and 5 overnight at
   4 °C. On day 2, after three washes with 1 × PBS, secondary antibodies
   were incubated with corresponding sections for 2 h at 37 °C. The
   sequential staining was followed by 1-h re-blocking with 10% donkey
   serum in 0.3% PBST, primary antibody combinations for each section were
   incubated at 4 °C in the same way as described above for regular IF
   staining. The nuclei were stained with Hoechst33342. Autofluorescence
   was quenched with 0.5 × TrueBlack solution in 70% ethanol for 10 min.
   The coverslips were mounted with Fluoromount-G Mounting Medium, and the
   slides were then imaged by Zeiss Axio Observer microscope.

H&E and IF staining image alignment

   The 20 × tile images taken by Zeiss Axio Observer microscope were
   exported into merged and individual channel images by Zeiss Zen
   software (2.6 blue edition). Since adjacent continuous sections were
   collected from the same brain tissue block, the shape outline and the
   structure of each section are identical (Additional file [86]1: Fig.
   S1A). The merged channel images were landmarked according to DAPI and
   hematoxylin staining in corresponding H&E images using ImageJ
   “Multi-points” tool (Additional file [87]1: Fig. S1A). Multi-points
   landmarks in each merged channel image were then saved and applied to
   individual channel images. Image alignment was performed using the
   “Transform/Landmark correspondences” plugin in ImageJ as described by
   other groups [[88]16]. After the transformation, IF staining of AT8 and
   Aβ, as well as other staining like Tau 46 were used to check whether
   the alignment was accurate. H&E image and transformed channel images
   were stacked into.tiff image and imported into Loupe Browser (v5.0.1)
   and Space Ranger (v1.2.2) for further alignment with Visium spots.

Annotation of cortical layers and the WM of the human MTG

   To assign Visium spots to their corresponding layers, we first combined
   all 25,293 spots from our six samples and generated a cross-sample
   Uniform Manifold Approximation and Projection (UMAP) via Seurat
   (v.4.0.5) [[89]28] pipeline to visualize dimension-reduced data in 2D
   space with unsupervised manner. All spots were clustered into eight
   groups and visualized in the original sample using eight colors
   assigned to each cluster. Although UMAP can roughly reflect the
   cortical layers’ information and separate the WM from the gray matter
   well, it is not powerful to segregate six layers in the gray matter of
   cortex. Since certain layer(s) may be missing during the brain
   dissection or mounting brain sections on the GE slide, arbitrarily
   assigning clusters from Seurat clusters into a sample may not be
   accurate without manual annotations. Moreover, the pathology
   development in AD brains may also change the gene expression profile
   and introduce more confounders in clustering. To solve this problem, we
   used both UMAP and IF staining as references to manually label each