profiling microglia from ad donors and non-demented ... · 3/18/2020 · to examine the effect of...

1

Profiling microglia from AD donors and non-demented elderly in 1

acute human post-mortem cortical tissue 2

3

Astrid M. Alsema1, #, Qiong Jiang1, #, Laura Kracht1, #, Emma Gerrits1, Marissa L. Dubbelaar1, 4

Anneke Miedema1, Nieske Brouwer1, Maya Woodbury3, Astrid Wachter4, Hualin S. Xi3, 5

Thomas Möller3, Knut P. Biber4, Susanne M. Kooistra1, Erik W.G.M Boddeke1,a and Bart J.L. 6

Eggen1,a * 7

8

1 Department of Biomedical Sciences of Cells & Systems, Section Molecular Neurobiology, 9

University of Groningen, University Medical Center Groningen, Antonius Deusinglaan 1, 10

9713AV Groningen, the Netherlands. 11

3 Foundational Neuroscience Center, AbbVie Inc., Cambridge, Massachusetts, United States of 12

America 13

4 Neuroscience Discovery, AbbVie Deutschland GmbH & Co. KG, Ludwigshafen, Germany 14

#These authors contributed equally to this work. 15

aThese authors share senior authorship. 16

*Correspondence: 17

Prof. dr. Bart J.L. Eggen 18

[email protected] 19

(which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted March 19, 2020. ; https://doi.org/10.1101/2020.03.18.995332doi: bioRxiv preprint

mailto:[email protected]

https://doi.org/10.1101/2020.03.18.995332

2

Abstract 20

Microglia are the tissue-resident macrophages of the central nervous system (CNS). Recent 21

studies based on bulk and single-cell RNA sequencing in mice indicate high relevance of 22

microglia with respect to risk genes and neuro-inflammation in Alzheimer’s disease. Here, we 23

investigated microglia transcriptomes at bulk and single cell level in non-demented elderly and 24

AD donors using acute human post-mortem cortical brain samples. We identified 9 human 25

microglial subpopulations with heterogeneity in gene expression. Notably, gene expression 26

profiles and subcluster composition of microglia did not differ between AD donors and non-27

demented elderly in bulk RNA sequencing nor in single-cell sequencing. 28

Keywords: Alzheimer's disease, microglia, single-cell RNA sequencing, barcoded Smart-seq2, 29

human 30

31

Introduction 32

Alzheimer’s disease (AD), one of the most prevalent age-related neurodegenerative disorders, 33

is characterized by extracellular deposition of β-amyloid protein (Aβ) and intra-neuronal 34

neurofibrillary tangles in the neocortex [1]. 35

Functional changes occurring in microglia cells have been proposed as an important factor in 36

AD pathology [2,3]. AD single nucleotide polymorphism (SNP) heritability was recently found 37

to be highly enriched in microglia enhancers [4]. Multiple genes associated with increased 38

susceptibility for sporadic AD are preferentially expressed in microglia, including APOE, CR1, 39

CD33, INPP5D, PLCG2, MS4A6A and TREM2 [5,6]. In AD mouse models, microglia have 40

been implicated in Aβ seeding, Aβ plaques are surrounded by activated microglia, microglia 41

protrusions physically interact with insoluble Aβ aggregates and microglia around plaques 42

undergo transcriptional changes [7–12]. Sustained depletion of microglia in 5xFAD mice 43

prevents Aβ plaque formation in parenchymal tissue, rather showing Aβ accumulation in the 44

brain vasculature [13]. The functional changes occurring in microglia during AD pathology 45

seem to be diverse [14] and the exact role that microglia play in AD pathology is still unknown. 46

Many efforts have been made in AD mouse models to identify subpopulations of microglia that 47

are associated with AD pathology. A microglial neurodegenerative subpopulation was 48

discovered by Krasemann and colleagues that was associated with Aβ plaques and triggered by 49

phagocytosis of apoptotic neurons [9]. This transcriptional phenotype was characterized by 50

increased Spp1, Itgax, Axl, Lilrb4, Clec7a, Ccl2, Csf1, and Apoe and decreased P2ry12, 51


https://doi.org/10.1101/2020.03.18.995332

3

Tmem119, Olfml3, Csf1r, Rhob, Cx3cr1, Tgfb1, Mef2a, Mafb, and Sall1 expression levels [9]. 52

At the same time, a highly similar gene expression change, associated with microglia 53

surrounding Aβ plaques was reported by Keren-Shaul and colleagues [8], termed disease-54

associated microglia (DAMs). Using single-cell RNA-sequencing these DAMs were 55

subdivided into two sequential stages, a Trem2-independent stage, marked by increased 56

expression of Tyrobp, Apoe, and B2m and decreased expression of homeostatic genes, followed 57

by a Trem2-dependent activation stage marked by induction of genes involved in lipid 58

metabolism and phagocytosis (Trem2, Spp1, Itgax, Axl, Lilrb4, Clec7a, Cts7, Ctsl, Lpl, Cd9, 59

Csf1, Ccl6, Cd68 and more). Sala et al. [15] described a microglia subpopulation that appears 60

in response to Aβ deposition in AppNL-G-F mice and shares gene expression changes with DAMs. 61

They identified mutually exclusive subtypes of activated response microglia (ARMs) 62

overlapping with DAMs and, in addition, an independent subtype of interferon response 63

microglia (IRMs). 64

Studies investigating human microglia subtypes are limited, probably due to the technical and 65

logistical difficulties of isolating pure, viable microglia from acute human brain tissue. Olah 66

and colleagues [16] investigated single human microglia from donors with a large variety of 67

neuropathological backgrounds. They observed 23 clusters of microglia, where 5 out of 23 68

clusters were enriched for DAM signature genes. However, the neuropathological background 69

of donors was too diverse to associate the observed changes to AD. Mathys and colleagues [17] 70

used single-nucleus sequencing and subclustered ~2400 microglia of 48 donors. This study was 71

focused on cell-type specific responses to AD development and around 50 microglia per donor 72

was insufficient to fully define microglia diversity in AD. 73

In this study, we aimed to identify transcriptomic changes in human microglia at the end-stage 74

of AD, by applying both bulk and single-cell sequencing of microglia acutely isolated from 75

acute post-mortem CNS tissue. We isolated and sequenced a pure population of microglia after 76

CD11B+CD45+-based FACS sorting and investigated effects of gender, brain region, tissue 77

dissociation methods and diagnosis. 78

79

Materials and methods 80

Human brain specimens 81

Autopsy brain specimens from the LPS, GFS and lateral temporal lobe were obtained from 25 82

donors of the Netherlands Brain Bank (NBB, https://www.brainbank.nl/) and 2 donors of the 83


https://www.brainbank.nl/

https://doi.org/10.1101/2020.03.18.995332

4

NeuroBiobank of the Institute Born-Bunge (NBB-IBB, Wilrijk, Antwerp, Belgium, ID: 84

BB190113). Mechanical isolation of microglia and bulk microglia sequencing was performed 85

with 22 specimens (11 GFS, 11 LPS) of 15 donors (6 AD, 6 CTR+, 3 CTR). From nine donors 86

both regions were collected and from the other donors either GFS (2016-038, 2016-095, 2017-87

102) or LPS (2017-098, 2017-124). From three donors (2016-038, 2016-046, 2016-080), LPS 88

or GFS specimens were divided for mechanical as well as enzymatic dissociation. Single-cell 89

sequencing was performed with 16 specimens (15 LPS, 1 temporal) from 15 donors (6 AD, 5 90

CTR+, 4 CTR), from which 2 were overlapping with donors used in bulk analysis (2018-18, 91

2018-021). All donors have given informed consent for autopsy and use of their brain tissue for 92

research purposes. The performed procedures, information - and consent forms of the NBB 93

have been approved by the Medical Ethics Committee of the VU Medical Centre. On average, 94

the autopsies were performed within 6h after death. Detailed information about brain specimens 95

used for bulk and single-cell sequencing can be found in Table S1 and S2, respectively. 96

Microglia isolation and sorting 97

The mechanical isolation of microglia was performed as described previously [18,19] with 98

minor modifications. All procedures were performed on ice and all centrifugation steps were 99

performed at 4°C. The tissue was homogenized by mechanical dissociation in Medium A 100

(HBSS (Gibco, 14170-088) containing 15 mM HEPES (Lonza, BE17-737E) and 0.6% glucose 101

(Sigma-Aldrich, G8769) and was filtered through a 300 and 106 µm sieve. Cells were 102

centrifuged at 220g for 10 min and myelin and other lipids were removed through two percoll 103

gradient centrifugation steps. A 100% Percoll solution was prepared consisting of 90% Percoll 104

(GE Healthcare, UK) and 10% 10x HBSS (Gibco, 14180-046), from which the dilutions were 105

prepared. First, cells were resuspended in 24.5% (vol/vol) Percoll in Medium A. A layer of PBS 106

was added and the gradient was centrifuged at 950g for 20 min with reduced acceleration speed 107

and brakes off. After the supernatant was removed, cells were resuspended in 60% (vol/vol) 108

Percoll in 1x HBSS (Gibco, 14170-088) and a layer of 30% (vol/vol) Percoll in 1x HBSS 109

(Gibco,) and PBS, respectively, were added and centrifuged at 800g for 25 min (acc: 4, brake: 110

0). The cells in between the 30%/60% percoll layer were collected and washed in 1x HBSS 111

(Gibco, 14175-053) and centrifuged at 600g for 10 min. 112

To examine the effect of the isolation protocol on the microglia expression profile, for three 113

tissue samples (2016-038, 2016-046, 2016-080) mechanical and enzymatic dissociation on two 114

halves of the same tissue was performed. 1.5 gram of tissue was incubated for 80 minutes at 115

37°C continuously shaking in enzyme solution (1x Trypsin/EDTA (Gibco, A1217701), 0.1 mM 116

HEPES (Lonza, BE17-737E), 0.5 mg/ml DNase I (Roche, 10104159001) in 1x HBSS (Gibco, 117


https://doi.org/10.1101/2020.03.18.995332

5

14170-088)). Next, it was continued with the Percoll gradient centrifugation steps as described 118

above. 119

Cells were incubated with anti-human Fc receptor (0,005 µg/ml eBioscience, 14-9161-73) for 120

10 min in Medium A without phenol red (HBSS (Gibco, 14170-053) containing 15 mM HEPES 121

(Lonza, BE17-737E), 0.6% glucose (Sigma-Aldrich, G8769), 1mM EDTA (Invitrogen, 15575-122

038)), followed by the incubation with 5 µg/ml DAPI (Biolegend, 422801), eBioscience 123

DRAQ5 (Thermofisher Scientific, 63351), FITC anti-human CD45 (BioLegend, 304006) and 124

PE anti-human CD11B (BioLegend, 301306). Single, viable microglia defined as DAPI-, 125

DRAQ5+, CD45+ and CD11B+, were FACS sorted on a Beckman Coulter MoFlo XDP or 126

Astrios. 127

For bulk microglia RNA sequencing, microglia were sorted into low retention Eppendorf tubes 128

(Sigma, Z666548-250EA) containing 200 µl RNA later (Qiagen, 76104). Following, 129

centrifugation at 4°C and 5000g for 10 minutes, supernatant was carefully removed and 130

microglia were resuspended in 350 µl RLTplus lysis buffer (Qiagen, 1053393) and stored at -131

80°C. For barcoded 3’ single-cell sequencing, 15792 single microglia were collected in 384-132

well PCR plates containing cell lysis buffer (0,2% Triton (Sigma-Aldrich, T9284), 4 U RNAse 133

inhibitor (Takara, 2313A), 10mM dNTPs (Thermo Scientific, #R0193) and 10µM barcoded 134

oligo-dT primer) and were stored for maximally one month at -80°C until further processing. 135

For 10x Genomics Chromium single cell RNA sequencing, approximately 25000 single 136

microglia were sorted from two samples (2018-135, 2ß19-010) into low retention Eppendorf 137

tubes (Sigma, Z666548-250EA) containing 5 µl Medium A and were immediately processed 138

with the Single Cell 3’ Reagent Kit v2 (10x Genomics). FACS data was analyzed with FlowJo 139

(Becton, Dickinson & Company). 140

Bulk microglia RNA sequencing library preparation 141

Total RNA was extracted from the bulk sorted microglia samples using the RNeasy Plus Micro 142

kit (Qiagen, 74034) according to the manufacturer's protocol. RNA quality and quantity were 143

determined with the Experion RNA HighSens Analysis Kit (Bio-Rad, #7007105). All 25 RNA 144

samples, with RIN values varying between 5.7 and 9.9, were enriched for poly(A)+ messenger 145

RNA using NEXTflex Poly(A) Beads (BIOO Scientific, #NOVA-512980) according to the 146

protocol in the manual. Fourteen µL of this mRNA-enriched poly(A)-tailed RNA was used as 147

the input for the NEXTflex Rapid Directional qRNA-Seq kit (Bioo Scientific, #NOVA-5130-148

04). Library preparation was performed according to the manufacturers protocol. Quality and 149

concentration of libraries from individual samples were assessed using the High Sensitivity 150


https://doi.org/10.1101/2020.03.18.995332

6

dsDNA kit (Agilent, 067-4626) on a 2100 Bioanalyzer (Agilent) and a Qubit 2.0 Fluorometer 151

(Life Technologies). Subsequently, individual libraries were combined into 2 sequencing pools 152

of 13 samples each with equal molar input. 75bp paired-end sequencing was performed on an 153

Illumina NextSeq 500 system. PhiX was added at 5% to both pools as an internal control before 154

sequencing. 155

Single-cell RNA sequencing library preparation 156

The single-cell RNA library preparation method that was used here is based on the Smart-157

seq2 protocol by Picelli [20], with the modification of obtaining 3’ instead of full-length 158

RNA/cDNA libraries as in Uniken Venema [21]. After cell lysis and barcoded poly-dT primer 159

annealing (73°C, 3 min), RNA was reversed transcribed (RT) based on the template 160

switching oligo mechanism using 0.1 µM biotinylated barcoded template switching oligo 161

(BC-TSO, 5’-AAGCAGTGGTATCAACGCAGAGTACATrGrG+G-biotin-3’), 25 U 162

SmartScribe reverse transcriptase, first-strand buffer and 2mM DTT (Takara, 639538), 1 U 163

RNAse inhibitor (Takara, 2313A) and 1M betaine (Sigma-Aldrich, B0300-5VL) with the 164

following PCR program: 1) 42 °C 90 min, 2) 11 cycli of 50 °C 2 min, 42 °C 2 min, 3) 70 °C 165

15 min. To account for amplification bias and to allow multiplexing of cells and samples, the 166

barcoded poly-dT primer contains a cell-specific barcode and a unique molecular identifier 167

(UMI) and a known sequence that is used as a primer binding site during the first 168

amplification step. This same primer binding site is linked to the BC-TSO enabling the use of 169

one primer pair (custom primer) during the first amplification. After the RT reaction, primer-170

dimers and small fragments were removed by 0.5 U Exonuclease (GE Healthcare, E70073Z) 171

treatment for 1 h at 42 °C. cDNA libraries were amplified with KAPA Hifi HotStart 172

ReadyMix (KAPA Biosystems, KK2602) and custom PCR primer (5’-173

AAGCAGTGGTATCAACGCAGAGT-3’) with the following PCR program: 1) 98 °C 3 min, 174

25 cycles of 98 °C 20 s, 67 °C 15 s 72 C 6 min, 3) 72 C 5 min. cDNA libraries of 84 cells 175

were multiplexed and short fragments were eliminated by Agencourt Ampure XP beads 176

(Beckman Coulter, A63880, ratio of 0.8:1 beads to library volume). The quality of 177

multiplexed cDNA libraries was examined with a 2100 Bioanalyzer (Agilent) according to the 178

manufacturer’s protocol. cDNA libraries with an average size of 1.5- 2 kb were tagmented 179

and indexed during a second PCR amplification step with the Illumina Nextera XT DNA 180

preparation kit (Illumina, FC-131-1024). Tagmentation was performed according to the 181

manufacturer’s protocol with an input of 500 pg cDNA and amplicon tagment mix for 5 min 182

at 55 °C. The tagmentation reaction was stopped using NT buffer. Next, tagmented cDNA was 183

amplified with Nextera PCR master mix, the Nextera indices (12 pool-specific indices, 184


https://doi.org/10.1101/2020.03.18.995332

7

Illumina, FC-131-2001) and 10 µM P5-TSO hybrid primer (5’-185

AATGATACGGCGACCACCGAGATCTACACGCCTGTCCGCGGAAGCAGTGGTATCA186

ACGCAGAGT*A*C-3’) with the following PCR program: 1) 72 °C 3 min, 2) 95 °C 30 s, 3) 187

10 cycles of 95 °C 10 s, 55 °C 30 s, 72 °C 30 s, 4) 72 °C 5 min). Tagmented cDNA libraries 188

were purified by 0.6:1 ratio of Agencourt Ampure XP beads (Beckman Coulter, A63880) to 189

library volume. The quality and concentration of tagmented cDNA libraries were determined 190

with a 2100 Bioanalyzer (Agilent). cDNA pools with an average size of 300-600 bp were 191

multiplexed using a balanced design with pools from 10 different donors (in total 840 cells) 192

per sequencing run. In other words, cells from each donor were distributed over several 193

sequencing runs to avoid potential batch effects. To eliminate short fragments, the final 194

superpool was gel-purified from 2% agarose gel (Invitrogen, 10135444) with the Zymoclean 195

Gel DNA Recovery kit (Zymo Research, D4007). The concentration was determined using a 196

2100 Bioanalyzer (Agilent) and Qubit 3.0 (ThermoFisher Scientific) according to the 197

manufacturers protocol. Pools were loaded on an Ilumina NextSeq at a final concentration of 198

2 pM with7% spike in of PhiX DNA. 0.3 µM BC read 1 primer (5’-199

GCCTGTCCGCGGAAGCAGTGGTATCAACGCAGAGTAC-3’) was used for the 200

sequencing run. The libraries were sequenced on an Illumina NextSeq 500 system with an 201

average sequencing depth of 17,391 UMIs per cell. 202

10x Genomics Chromium single cell 3’ library construction 203

The single-cell barcoded libraries were constructed according to the instructions of the Single 204

Cell 3’ Reagent Kits v2 (10x Genomics). Briefly, cells were loaded into a slot of a Chromium 205

chip and GEMs were incubated in a thermal cycler to generate barcoded cDNA. After 206

amplification, the cDNA was fragmented and processed for sequencing by ligating adapters and 207

sample indices. The libraries were sequenced on an Illumina NextSeq 500 system with a 208

sequencing depth up to 20,000 reads per cell. 209

Immunohistochemistry 210

Immunohistochemistry was performed as described previously [11]. Briefly, 16 μm sections of 211

PFA-fixed human brain tissue were vacuum-dried, post-fixated for 10 minutes with 4% PFA, 212

and washed with PBS. Heat-induced antigen retrieval was performed in sodium citrate solution 213

(pH 6.0) for 10 min in a microwave. Endogenous peroxidase was blocked by incubating the 214

slides in 0.3% H2O2 for 30 min. After three washing steps with PBS, primary antibodies against 215

IBA1 (WAKO, 019-19741, 1:1000), Phospho-TAU (ThermoFisher, MN1020, clone AT8, 216

1:750) and Beta-Amyloid (Dako, M0872, 1:100) were diluted in Bright Diluent (ImmunoLogic, 217

BD09-500) to prevent specific background staining and incubated overnight at 4°C. After three 218


https://doi.org/10.1101/2020.03.18.995332

8

washing steps in PBS, secondary biotinylated horse anti-mouse IgG antibody (0.000125 mg/ml 219

Vector BA-2001) was incubated for 1 hour at room temperature. The tissue sections were 220

washed three times in PBS. The signal was amplified by induction with avidin-biotin complexes 221

(Vectastain Elite ABC‐HRP (Vector, PK‐6100)) for 30 min at RT and visualized with 3,3′‐222

diaminobenzidine (Sigma, D-5637). Additionally, after the phospho-TAU staining we 223

performed a crystal violet counterstaining. The slides were dehydrated with an ethanol series 224

(50%, 70%, 80%, 90%, 2x 96% and 3x 100% ethanol) and air-dried for 30 min prior to 225

mounting a coverslip with DePeX (Serva, 18243). Imaging was performed with a Hamamatsu 226

Nanozoomer at a 40x magnification. 227

Preprocessing of RNA-sequencing data 228

For bulk samples, the first preprocessing step was stripping the NEXTflex barcode (9 base pairs) 229

from the sequence. The barcode was saved in a separate file for further use. Next, microglia 230

bulk sequencing reads were aligned with HISAT2 (version 2.1.0) to the GRCh38.92 reference 231

genome with Ensembl annotation [22]. Further processing was done with samtools (version 1.9) 232

and Picard Tools (version 1.140). This included sorting, assigning reads to a read group, 233

verification of mate pair information and marking duplicates. After these processing steps, reads 234

were quantified using featureCounts (Subread version 1.6.2) [23]. Reads from bulk samples 235

were deduplicated using a bash script by BIOO Scientific (v2, date 11/1/14) , using the 236

NEXTflex barcode that was saved in the additional file. This allows proper elimination of PCR 237

duplicates taking advantage of stochastic labeling of UMIs. 238

Reads from bc-Smart-seq2 single cells were demultiplexed based on cell barcodes using UMI-239

tools (version 0.5.3) [24]. We created a whitelist of cell barcodes by applying the whitelist 240

function of UMI-tools. Next, the whitelist is compared to the true cell barcodes, and these 241

barcodes are saved as a cell barcode list that is used for downstream processing. The UMI-tools 242

extract function was used to extract the cell barcode and the UMI from each read and add this 243

information to the read name of the sequence. Reads were single-end aligned with HISAT2 244

(version 2.1.0) to the GRCh38.91 reference genome with Ensembl annotation using the default 245

parameters, followed by sorting and indexing of the BAM files. 246

Primary counts were quantified with the function featureCounts (Subread version 1.6.0) using 247

the flag –primary and -R BAM to save the BAM file. PCR duplicates were removed and unique 248

molecules were counted per gene and per cell using the function UMI-tools count [24]. 249

Seventeen pools with less than 10% of reads assigned to genes were excluded. 250

251

Reads from 10x Genomics Chromium single cells were demultiplexed and aligned to the 252

GRCh38 genome with Ensembl transcriptome annotation with Cell Ranger using default 253


https://doi.org/10.1101/2020.03.18.995332

9

parameters. The demultiplexed fastq files were used as input for the 10x Genomics pipeline 254

Cell ranger. Barcode filtering was performed with the R package DropletUtils with a threshold 255

of > 100 UMIs per barcode [25]. 256

Downstream analysis 257

For bulk samples, DAFS filtering was used to remove lowly expressed genes [26] and principal 258

component analysis was computed on rlog transformed counts using the DESeq2 R package 259

(v1.24.0). We applied upperquartile normalization to adjust for library size and used edgeR 260

(version 3.26.8) for differential gene expression analysis. Average library size after filtering 261

was 3.9M counts (± standard deviation 1.9M). To model gene expression levels we used a 262

negative binomial generalized log-linear model as implemented in edgeR, adjusting for gender, 263

analyzing different brain regions separate. Differences between groups were tested with 264

likelihood ratio tests. Thresholds were set at FDR > 0.05 and abs(logFC) > 1 to define DEGs. 265

For bc-Smart-seq2 single cells, approximately 25% of the cells were filtered out during 266

preprocessing. Downstream analysis started with 16 samples and 12,861 single cells with 267

detectable cell barcodes from 15 different donors. One donor (iB6399-BA7) contributed tissue 268

from both the temporal cortex and the superior parietal cortex. Cell library sizes before filtering 269

were quite different across donors; this is most likely due to differences in tissue quality. To 270

remove empty cells while respecting variation across donors, we set a threshold for each donor 271

individually, removing cells with library sizes exceeding median of log(total counts) ± 3 median 272

absolute deviation (mad) [27]. In addition, cells with more than 3000 unique genes were 273

considered doublets (genes per cell median 520; mad ±276) and cells with >10% mitochondrial 274

reads were excluded. We excluded ~10% of the cells (differs per donor) based on these criteria. 275

In total, this left 11,419 cells for analysis. Genes not detected in at least 3 cells were removed. 276

After filtering, we observed a median of 13,394 UMIs and median 520 genes per cell for bc-277

Smart-seq2 data. 278

For 10X data, similarly, low quality cells with >10% mitochondrial gene (MT) content were 279

removed in donor 2018-135. Donor 2019-010 had very high cell quality, so a >5% MT 280

threshold was applied. Duplicate cells were filtered by setting an upper UMI threshold that was 281

based on the multiplet rate as mentioned in the 10x Genomics user guide. We analyzed 3,077 282

single cells for MCI donor 2019-010 and 2,881 single cells for AD donor 2018-135. 283

Raw counts were normalized by total expression per cell, scaled by 10,000, and log-transformed 284

with the CRAN package Seurat (v3.0.0). We used the mean variability method to select highly 285

variable genes (HVGs). Briefly, this method identifies variable genes while controlling for the 286

strong relation between gene variability and gene average expression. We allowed lowly 287

expressed genes in the highly variable gene list, since some disease associated genes (e.g. 288

TREM2, TYROBP, CTSD) are biologically relevant but also lowly expressed. These extra 289

~632 lowly expressed HVGs did not change clustering results and were included in the final 290


https://doi.org/10.1101/2020.03.18.995332

10

clustering analysis. Unwanted sources of variation, such as number of detected genes, 291

ribosomal content, and mitochondrial content were regressed out. We used the first 20 principal 292

components for clustering analysis applying PCA-Louvain maximum modularity clustering as 293

implemented by Seurat. Cluster resolution was set at 0.5, since it formed the most stable plateau 294

in cluster number when considering resolutions from 0.1 to 2.0. 295

10X single cells were analyzed identically. Each donor was analyzed individually and cluster 296

resolution was adjusted due to higher cell numbers. For donor 2019-135 cluster resolution 0.6 297

and for donor 2019-010 cluster resolution 0.8 was set. Marker genes between clusters were 298

identified using MAST implemented in Seurat’s FindAllMarkers function with default 299

thresholds (logfc.threshold = 0.25, min.pct = 0.1). Gene ontology (GO) analysis on upregulated 300

marker genes was calculated using the ClusterProfiler (v3.12.0) [27] with p-value cut-off of 301

0.01 and q-value cut-off of 0.05. For FDR control we used the R package q-value (version 302

3.1.1). 303

304

Gene set analysis 305

Briefly, raw expression values were normalized to counts per million and log(cpm + 1) 306

transformed. The gene set score was an average over all genes in the set per cell. We compared 307

if, on average, cells in any one cluster showed a higher mean expression of the gene set than all 308

other cells by using multiple linear regression with gene set score as dependent variable and 309

cell library size, donor, and cluster as independent variables. Afterwards, p-values were 310

adjusted with a Bonferroni correction. Visualizations were made with the R package ‘ggplot2’. 311

312

Results 313

Isolation of pure microglia from acute post-mortem brain tissue 314

To investigate transcriptomic changes in microglia during AD, bulk and single-cell sequencing 315

were performed. Post-mortem tissue samples of the superior parietal lobe (LPS), superior 316

frontal gyrus (GFS) and lateral temporal lobe were obtained from 27 donors. The samples were 317

classified into 3 experimental groups based on a clinical diagnostic report provided by the 318

Netherlands Brain Bank/ NeuroBiobank Born-Bunge and immunohistochemical analysis of Aβ 319

and hyperphosphorylated tau (PHF-tau): CTR (no dementia, absence of Aβ plaques and PHF-320

tau), CTR with plaques (CTR+) (no dementia, presence of Aβ plaques and/or PHF-tau) and AD 321

(dementia, AD diagnosis, presence of Aβ plaques and/or PHF-tau). One donor diagnosed with 322

mild cognitive impairment (MCI) and presence of Aβ and PHF-tau plaques was included as 323

well. Representative images of Aβ and PHF-tau immunohistochemical stainings are available 324

in Figure S1. The goal of the stratification between CTR and CTR+ was to ensure that the CTR 325

group is free of undiagnosed AD donors. Microglia were isolated from enzymatically or 326


https://doi.org/10.1101/2020.03.18.995332

11

mechanically dissociated tissue and purified by fluorescence-activated cell sorting (FACS), 327

based on single, viable CD11B and CD45 positive cells. Twenty-five samples (12 LPS and 13 328

GFS; 22 mechanically and 3 enzymatically dissociated) from 14 donors were sorted and 329

sequenced as bulk samples and 16 samples (15 LPS, 1 temporal lobe) from 15 donors were 330

single-cell sorted and sequenced (bc-Smart-seq2 and 10x Genomics) (Fig 1A). Two donors 331

were included in both the single-cell and bulk cohort. The employed strategy resulted in a pure 332

microglia population, as confirmed by expression of known microglia genes, that were based 333

on previously published adult human microglia [19] and human CNS nuclei [28] gene 334

expression profiles (Fig 1B). 335

336

Figure 1 Microglia gene expression profiling of four groups of donors from acute post-337

mortem human brain tissue. 338

(A) Tissue samples from GFS and LPS were classified into four experimental groups (CTR, 339

CTR+, AD, MCI) based on Aß and Tau immunohistological stainings and the clinical report of 340

the donor. Microglia were mechanically or enzymatically isolated and collected by 341

CD11B+CD45+-based FACS sorting. Sorted microglia were used for bulk or single-cell 342

sequencing (barcoded Smartseq2 and 10x Genomics). Scale bar= 50µm. (B) Expression of top 343

25 previously published celltype-specific marker genes in bulk-sorted microglia [17, 18]. 344

Abbreviations Aß: amyloid beta, HPF-tau: hyperphosphorylated tau, AD: clinically diagnosed 345

Alzheimer’s Disease with Aβ plaques and/or hyperphosphorylated tau, CTR: Control, CTR+: 346

Control with Aβ plaques and/or hyperphosphorylated tau, GFS: superior frontal gyrus, LPS: 347


https://doi.org/10.1101/2020.03.18.995332

12

superior parietal lobe, MCI: mild cognitive impairment, NPC: Neural Progenitor Cells, OPC: 348

Oligodendrocyte Precursor Cells, SC: single cell. 349

350

No transcriptomic differences between CTR, CTR with plaques (CTR+) and AD donors 351

in acute post-mortem tissue by bulk RNA-sequencing 352

To investigate general transcriptional characteristics of microglia in AD, we analyzed bulk 353

sorted and sequenced microglia from CTR, CTR+ and AD samples. Principal component 354

analysis (PCA) showed no segregation between donor groups (Fig 2A). The second principle 355

component showed segregation of microglia samples on gender, but not on brain region or age 356

(Fig 2A). 357

To further examine the effect of gender, male and female GFS microglia samples were 358

compared. Here, 33 genes were differentially expressed with an absolute log fold change > 1, 359

FDR-value < 0.05. One third of these differentially expressed genes (DEGs) were X or Y-linked 360

genes (Table S3). 361

Differences between microglia from frontal and parietal brain regions were assessed. For 9 362

donors, microglia were isolated from both the LPS and GFS region, and a within-subject 363

comparison was applied. Only the expression of JAML and TREM1 was significantly higher in 364

LPS compared to GFS (Table S4). This indicates that the gene expression profiles of microglia 365

isolated from frontal and parietal brain regions were very similar. 366

The effect of the dissociation method on gene expression was determined by comparing the 367

bulk RNA-sequencing profiles of GFS microglia from enzymatically and mechanically 368

dissociated tissue of donors 2016-038 and 2016-080 (4 samples). 10 genes were significantly 369

higher expressed in microglia isolated using enzymatic dissociation: CXCL8, SOCS1, NR4A1, 370

ID1, FOSB, ATF3, NR4A2, EGR2, OSM and HSPA7 (Table S5). As the enzymatic dissociation 371

procedure induced expression of these 10 genes in the bulk samples of these 2 donors, only 372

mechanical dissociation was used for single-cell sequencing. 373

Bulk samples from AD and CTR donors were compared while correcting for the effect of 374

gender on gene expression. LPS- (n=11) and GFS-derived (n=11) microglia were analyzed 375

separately. In the analysis of microglia from both LPS and GFS regions no differentially 376

expressed genes from AD donors compared to CTR donors were found. 377

Lastly, expression levels of the top 50 genes upregulated in DAMs [8] were investigated in bulk 378

microglia. These mouse top DAM genes were not differentially expressed between AD-derived 379

compared to control microglia in LPS and GFS regions (Fig 2B). For several known DAM 380


https://doi.org/10.1101/2020.03.18.995332

13

genes, such as APOE, TREM2, ITGAX a small increase in expression in AD donors compared 381

to CTR can be recognized, albeit it is not statistically significantly (Fig 2C). 382

To summarize, in bulk transcriptomes of microglia from AD and CTR donors, no significant 383

gene expression differences were detected. 384

385

386

Figure 2 Transcriptomic analysis of microglia populations isolated from CTR, CTR+ and 387

AD donors. 388

(A) PCA of RNA-sequencing data from post-mortem isolated microglia illustrating the effect 389

of tissue dissociation method, donor (each sample is indicated with the donor label), donor 390

groups, gender, brain region, and age. 391

(B) Expression of disease-associated microglia genes in bulk microglia samples from three 392

donor groups (z scores). Donors are ordered based on age within their group. 393


https://doi.org/10.1101/2020.03.18.995332

14

(C) Selected examples of disease-associated microglia gene expression (log2(CPM)) in bulk 394

microglia samples. Abbreviations: AD: Alzheimer’s Disease, CPM: counts per million, CTR: 395

Control, CTR+: Control with Aβ plaques and/or hyperphosphorylated tau, F: female, M: male, 396

GFS: superior frontal gyrus, LPS: superior parietal lobe. 397

398

Single-cell gene expression profiling identifies 9 subsets of microglia but no AD-associated 399

subpopulation 400

As AD-associated changes possibly only occur in a small subpopulation of plaque-associated 401

microglia and thus might be masked by bulk RNA-sequencing, we next employed single-cell 402

sequencing to investigate microglia heterogeneity. A median of 753 microglia per donor from 403

4 CTR, 5 CTR+, a donor with mild cognitive impairment (MCI) and 5 AD donors were 404

analyzed. The read depth per cell for each donor was comparable, however the median number 405

of detected genes for each donor varied (Fig S2A,B). Clustering analysis identified 9 microglial 406

subsets, indicating heterogeneity in microglia transcriptomes (Fig 3A). Sequencing depth, 407

uniquely detected genes and detected mitochondrial genes across clusters are visualized in Fig 408

S2. The expression of cell type-specific markers across all clusters showed that all analyzed 409

cells were microglia, and that this population was free from neuronal, erythrocyte, and other 410

glial cell contaminations (Fig 3B). Figure 3C depicts the percentage of cells from a certain 411

donor that contribute to each cluster. Except for cluster 3, all clusters exhibited roughly equal 412

donor contributions. Cluster 3 has a high contribution from AD donor 2018-077 and cells from 413

this donor had a relatively low fraction of mitochondrial genes (Fig S2C) compared to other 414

cells. Gene expression in cluster 3 will strongly depend on this particular donor. Therefore, 415

cluster 3 (9.2% of the cells) was excluded from further analysis. 416

417

Differential gene expression analysis identified marker genes that were significantly enriched 418

in each cluster compared to all other clusters. However, marker genes significantly enriched in 419

a cluster (Table S6) and the biological annotation of those genes (Fig S3) did not point towards 420

clear biological functions of identified clusters. Expression of homeostatic genes such as 421

CSF1R, ITGAM, TMEM119 and CX3CR1 was quite variable between cells in a cluster (Fig 422

3D). These genes were not identified as marker genes in any of the clusters (Table S6). Many 423

microglial genes previously associated with antigen presentation (HLA genes), the complement 424

system (C1QA), neurodegenerative diseases (TREM2, TYROBP, B2M), lysosomal proteases 425

(CTSB) and phagocytosis (CD14, AXL) were generally expressed at higher levels than 426

homeostatic genes and were found to be marker genes of cluster 2 (Fig 3D). 427


https://doi.org/10.1101/2020.03.18.995332

15

Unexpectedly, cluster 2 was not enriched by AD-derived microglia and overall no AD-428

associated microglia cluster was detected. APOE expression is associated with AD and was 429

elevated but not significantly enriched in cluster 2 according to differential expression analysis 430

(Fig3D, Table S6). The majority of AD-derived microglia contributed to cluster 0. Cluster 0 431

was comprised of 58% AD and 42% CTR/CTR+ microglia (Fig 3C). 432

The small clusters 4-8 contain relatively few cells (0.8-1.8%), they showed increased 433

expression of genes involved in protein degradation such as PSMD11, PSMB2, PSMB3, 434

EDEM3, FOXN2, UBQLN1 and DERL2. Possibly, the small microglia clusters 4-8 reflect cell 435

states with increased stress responses. 436

In 5XFAD mice, disease-associated microglia (DAMs) were characterized by expression 437

changes in set of genes [8]. To investigate if the increased gene expression in DAMs was also 438

present in one of the identified clusters, the average DAM gene set expression per microglia 439

was calculated. The average DAM gene expression level per cluster was compared with all 440

other clusters. Clusters 0 and 7 showed significantly lower average DAM gene expression levels 441

and microglia in cluster 2 showed significantly higher average DAM gene expression levels (S 442

Table 7), p < 0.001. Although statistically significant, these changes are likely too small to be 443

biologically relevant. These observations were confirmed in a different gene set from a meta-444

analysis of multiple amyloid mouse models [14]. Again, microglia in cluster 0 showed a 445

significant decrease in neurodegenerative-related gene expression levels. To summarize, only 446

the decrease in gene expression in cluster 0 was consistent and no cluster was associated with 447

increased neurodegeneration-related gene expression (Fig 3E). 448

Taken together, single-cell analysis of approximately 11,419 microglia which were 449

CD11B+CD45+-based FACS sorted and profiled with bc-Smart-seq2, identified microglial 450

heterogeneity, but no AD-specific microglia subpopulation. 451


https://doi.org/10.1101/2020.03.18.995332

16

452

Figure 3 Single-cell expression profiling of microglia identified 9 subsets of microglia but 453

no AD-associated cluster using bc-Smart-seq2. 454


https://doi.org/10.1101/2020.03.18.995332

17

(A) UMAP visualization and unsupervised clustering of microglia derived from AD (n=5), MCI 455

(n=1), CTR+ (n=5) and CTR (n=4) donors. Each dot represents a cell, colors represent cluster 456

identity. (B) Expression of cell type-specific markers for neurons (RBFOX3), oligodendrocytes 457

(MOG), astrocytes (GFAP), infiltrating immune cells as monocytes (CCR2), erythrocytes 458

(HBA1) and microglia (ITGAX). (C) Cluster contribution normalized to donor. Bar height 459

indicates the percentage of cells from a donor that contributed to the clusters. (D) Violin plots 460

show the probability density distribution of normalized gene expression across clusters for 461

selected genes. (E) The average mouse DAM gene set expression in each cell across clusters 462

(left). The average human myeloid neurodegenerative-related gene set expression in each cell 463

(right). Abbreviations: AD: Alzheimer’s Disease, CPM: counts per million, CTR: Control, 464

CTR+: Control with Aβ plaques and/or hyperphosphorylated tau, DAM: disease-associated 465

microglia, UMAP: uniform manifold approximation and projection 466

467

Microglia diversity but no DAM-like cluster in an individual MCI and AD donor with 468

high microglial cell numbers 469

The proportion of DAM-like microglia in human AD brain might be relatively low and could 470

potentially be missed by bc-Smart-seq2 expression profiling. Therefore, two donors with ~3000 471

microglia each were investigated using the 10X Genomics platform. First, 2881 cells were 472

analyzed from post-mortem LPS tissue of a female AD donor, 81 years old, with high Aβ 473

burden and modest but visible PHF-tau protein (2018-135) (Fig S1). Second, 3077 single cells 474

from an MCI donor (2019-010), 77 years old, with moderate Aβ pathology and high levels of 475

PHF-tau protein in the LPS were analyzed (Fig S1). We applied an identical clustering 476

procedure as for bc-Smart-seq2 libraries, now analyzing each donor individually, only adjusting 477

the clustering resolution for higher cell numbers. Per donor, three clusters were identified (Fig 478

4A, B). As clustering is performed within one sample, the identified clusters are not confounded 479

by post-mortem delay, gender, tissue quality or other donor-associated factors. Despite testing 480

multiple cluster resolutions, a DAM-like cluster could not be detected. 481

As described for figure 3, the average gene expression levels were calculated for each cluster 482

and compared to all other clusters. The gene expression averages per cluster were calculated 483

for the mouse DAM gene set [8] and for the neurodegenerative-related gene set [14] (Fig 4B, 484

C, E, F). A significant increase of the DAM gene set and the neurodegenerative gene set was 485

observed in microglia in cluster 0 in the MCI donor (Fig 4 E,F), but not in the AD donor (Fig 4 486

B,C). In short, no cluster with DAM-like microglia was detected in either donor, although a 487

high number of microglia was analyzed (~3000 cells/donor). 488


https://doi.org/10.1101/2020.03.18.995332

18

489

Figure 4 Single-cell expression profiling of microglia from individual donors with high 490

cell numbers using 10X Genomics. 491

(A,B) Unsupervised clustering identified 3 subsets of microglia in AD donor 2018-135 (A) and 492

in MCI donor 2019-010 (D). The average DAM gene set expression per cell and the myeloid 493

neurodegenerative-related gene set per cell across clusters for AD donor 2018-135 (B, C) and 494

for MCI donor 2019-010 (C, F). Abbreviations: AD: Alzheimer’s Disease, CPM: counts per 495

million, CTR: Control, CTR+: Control with Aβ plaques and/or hyperphosphorylated tau, DAM: 496

disease-associated microglia, UMAP: uniform manifold approximation and projection 497

498

Discussion 499

In this study we aimed to identify transcriptomic changes in human microglia at the end-stage 500

of AD by applying both bulk and single-cell RNA sequencing of microglia isolated from acute 501

post-mortem CNS tissue. In acute parietal and frontal cortex, we analyzed microglia as bulk 502

samples allowing the most sensitive detection of small gene expression changes. Here, 503

transcriptomic differences between AD and CTR were not detected and disease-associated 504


https://doi.org/10.1101/2020.03.18.995332

19

genes identified in AD mouse models were not enriched in AD microglia. Single cell 505

sequencing analysis was applied to detect microglial subtype populations, possibly consisting 506

of low cell numbers, that might be unnoticed in the average transcriptome obtained by bulk 507

analysis. Also in single microglia transcriptomes, the relative contribution to microglia clusters 508

did not differ between AD and control donors, when using the bc-Smart-seq2 protocol on 509

CD11B and CD45 sorted cells. 510

Expression of the average neurodegenerative-related gene sets were not significantly increased 511

in any cluster. The neurodegenerative disease-associated microglial subtype originally 512

described by Krasemann et al. [9] and Keren-Shaul et al. [8] represented a relatively small 513

fraction of the total microglia population. To rule out that the lack of a DAM-like cluster in our 514

data was due to the analysis of too low cell numbers, a higher number of microglia from two 515

donors were single-cell sequenced using the 10X Genomics platform. Clustering was 516

performed per individual donor to avoid donor variation which might mask a potential DAM 517

cluster. Concordant to the bc-Smart-seq2 data, the top DAM genes were expressed to a similar 518

extent in all clusters. In conclusion, microglial transcriptomes between AD and CTR donors 519

did not differ and a DAM-like microglial subtype was absent in AD donors. 520

When comparing the clustering results of the 10X Genomics and bc-Smart-seq2 single-cell 521

RNA sequencing techniques, differences were observed. Clusters 0-2 in the bc-Smart-seq2 data 522

contain the vast majority of microglia and are most similar to the three clusters identified per 523

donor in the 10X Genomics data set. The smaller clusters in the bc-Smart-seq2 data (4-8) were 524

not identified in the microglia profiled by 10X Genomics and are possibly associated with the 525

plate-based protocol as we observed similar small clusters in a different bc-Smart-seq2 dataset 526

as well (unpublished results). 527

There are multiple possible explanations for the absence of AD-associated changes in acute 528

bulk and single cell microglia. The lack of detectable AD-associated changes could be due to 529

limited sample size and donor variation. Another explanation for the lack of AD-related 530

transcripts in bulk and single cell microglia could be that the relevant microglia associated with 531

AD plaques are more vulnerable to the isolation procedure. This would imply that, using 532

conventional isolation and sorting of microglia would enrich a population of cells that are not 533

related to AD pathology. Streit and colleagues first introduced the concept of dystrophic 534

microglia that occur around neuronal structures positive for hyperphosphorylated tau protein 535

[29,30] and that can occur around Aβ plaques as well [31]. Possibly, dystrophic microglia and 536

microglia embedded inside the Aβ plaque are more vulnerable and therefore preferentially lost 537

during FACS gating of live, single cells. 538


https://doi.org/10.1101/2020.03.18.995332

20

When comparing AD mouse models to human AD, the distinction between parenchymal and 539

plaque-associated microglia might be more pronounced for amyloid mouse models than for 540

human end-stage AD samples. In transgenic amyloid mouse models, especially 5xFAD mice, 541

Aβ is over-expressed in a non-physiological manner, resulting in very fast Aβ plaque formation. 542

Transgenic mouse models lack regional brain atrophy and show less widespread 543

neurodegeneration than human AD cases [32]. Possibly, the parenchymal microglia and the rest 544

of the CNS are less affected in amyloid mouse models. Furthermore, quantifying the cortical 545

dense-core plaque burden in 5XFAD and APPswePS1ΔE9 mouse models showed that in mice, 546

the plaque burden eventually reaches 10 times higher levels than in human AD brain tissue [33]. 547

Additionally, inter-individual variation will influence human microglia transcriptomes more 548

than mouse microglia transcriptomes. Together these observations might lead to a more 549

pronounced change between parenchymal and plaque-associated microglia in amyloid mouse 550

models than in human AD samples. For this reason, single human microglia studies will most 551

likely require much larger cell numbers to capture sufficient plaque-associated microglia than 552

studies using AD mouse models. 553

DAMs were not only associated with neurodegenerative diseases, but also with natural ageing 554

[8,9]. This was confirmed by Sala and colleagues investigating an AppNL-G-F-associated 555

microglia subpopulation, ARMs, which overlap with DAMs. ARMs already comprised a few 556

percent of microglia in the brains of wild-type mice at young age and evolve naturally with 557

ageing [15]. Furthermore, a consensus gene expression network module co-occurring both 558

during ageing and neurodegeneration was detected. The described module included DAM 559

signature genes such as Csf1, Spp1, Apoe, Axl, B2m, Ctsz, Cd9, Cstb, and Cst7. Biological 560

annotation of module-specific genes included phagosome and lysosomal pathways [34]; 561

functions previously associated with DAMs [8]. Taken together, this suggests the presence of 562

DAM-like microglia could be expected, albeit at low levels, in aged controls as well as AD 563

donor-derived microglia. 564

It is still an unresolved question whether a subpopulation resembling DAMs exist among human 565

microglia. Three other studies previously addressed this question. Olah et al. [16] observed 23 566

clusters of human microglia, where 5 out of 23 clusters were enriched for DAM signature genes. 567

Three of the 15 donors suffered from AD pathology, making it difficult to connect their 568

microglia subpopulations with AD-induced gene expression changes. Srinivasan et al. [35] 569

investigated frozen myeloid cells from AD brain tissue and observed that from the 100 DAM 570

genes, only expression of APOE did indeed change in myeloid cells from AD donors compared 571

to controls. Mathys and colleagues used single-nuclei sequencing and subclustered ~2400 572


https://doi.org/10.1101/2020.03.18.995332

21

microglia of 48 donors into 4 subpopulations. They highlighted the Mic1 cluster as AD-573

pathology-associated human microglia [17]. From the 257 DAM genes investigated, only 28 574

were overlapping with marker genes for the Mic1 cluster and 16 of these 28 overlapping marker 575

genes were ribosomal genes. A larger-scale investigation of microglia nuclei will be needed to 576

reveal AD-associated microglia subpopulations in humans. 577

Single nucleus gene expression faithfully recapitulates cellular gene expression profiles [36,37]. 578

Therefore, single-nucleus sequencing offers an alternative to single-cell sequencing that is 579

especially useful in tissues from which recovering intact cells is difficult [36–38]. An important 580

advantage of single-nucleus sequencing is the possibility to use frozen samples from brain 581

banks containing large, well-characterized donor cohorts. For example, an improved donor 582

cohort to differentiate the effects of natural ageing and AD pathology would be possible by 583

comparing aged-matched (young) control donors to early-onset familial AD cases. In future, 584

single-nucleus sequencing of microglia and including tissues of early, pre-symptomatic stages 585

of AD will be most promising to potentially identifying (early) microglia biomarkers for AD. 586

587

Conflict of Interest 588

MW, AW, SX, TM and KB were full time employees of Abbvie during the time of the studies. The remaining 589 authors declare that the research was conducted in the absence of any commercial or financial relationships that 590 could be construed as a potential conflict of interest. 591

592

Author Contributions 593

EB, BE and SK were responsible for the overall conception of the project and provided supervision. QJ, AA, LK, 594 conducted the experimental work, and/or analyzed the data, prepared the figures and wrote the manuscript. EG 595 assisted with sample processing, data discussion and conducted all immunohistochemistry. NB, AM, MD, MW, 596 AW, SX, KB, TM, YH, and TO assisted with maintenance of availability of reagents, sample processing, data 597 discussion, optimization issues and/or data analysis. All authors contributed to the editing of the paper. 598

599

Funding 600

MW, SX, AW, TM and KB are employed by AbbVie, Inc., which has subsidized the study. BE acquired funding 601 from the foundation Alzheimer Nederland. AA, EG are funded by Abbvie. AA was supported by the Jan Kornelis 602 de Cock- Hadders foundation. QJ was funded by the Li Ka-shing Foundation at Shantou University Medical 603 College, China, the Abel Tasman Talent Program, University Medical Center Groningen/ University of Groningen, 604 The Netherlands, and the Graduate School of Medical Sciences (GSMS), University of Groningen, the Netherlands. 605 LK holds a scholarship from the GSMS, University of Groningen, the Netherlands. 606

607


https://doi.org/10.1101/2020.03.18.995332

22

Acknowledgments 608

We thank the Netherlands Brain Bank and the NeuroBiobank of the Institute Born-Bunge, Belgium. We thank 609 Ortiz, T. Oshima, M. Wijering, S. Eskandar, R. van der Pijl, E. Wesseling, C. Grit and C.B. Haas for assistant 610 and/or maintenance of sample processing and bulk and/or single cell isolations. We are grateful to W. Abdulahad, 611 G. Mesander, T. Bijma and J. Teunis from The Central Flowcytometry Unit of the University of Groningen, 612 University Medical Center Groningen (UMCG), for technical assistance on FACS sorting. We thank P. van der 613 Vlies, D. Brandenburg, N. Festen and W. Uniken Venema for their assistance setting up the bc-SmartSeq2. We 614 thank M. Meijer for ICT-related support. We appreciate sequencing-related advices provided by D. Spierings, K. 615 Hoekstra-Wakker, J. Beenen and N. Halsema. 616

617

Data availability 618

The data supporting the conclusions of this manuscript will be made available on Gene Expression Omnibus upon 619 publication. 620

621 References 622

1. Hyman BT, Trojanowski JQ. Consensus recommendations for the postmortem diagnosis of Alzheimer 623 disease from the National Institute on Aging and the Reagan Institute Working Group on diagnostic 624 criteria for the neuropathological assessment of Alzheimer disease. Journal of neuropathology and 625 experimental neurology. England; 1997. pp. 1095–1097. doi:10.1097/00005072-199710000-00002 626

2. Zhang B, Gaiteri C, Bodea L-G, Wang Z, McElwee J, Podtelezhnikov AA, et al. Integrated systems 627 approach identifies genetic nodes and networks in late-onset Alzheimer’s disease. Cell. 2013;153: 707–628 720. doi:10.1016/j.cell.2013.03.030 629

3. Mhatre SD, Tsai CA, Rubin AJ, James ML, Andreasson KI. Microglial malfunction: the third rail in the 630 development of Alzheimer’s disease. Trends Neurosci. 2015;38: 621–636. 631 doi:10.1016/j.tins.2015.08.006 632

4. Nott A, Holtman IR, Coufal NG, Schlachetzki JCM, Yu M, Hu R, et al. Brain cell type-specific 633 enhancer-promoter interactome maps and disease-risk association. Available: 634 http://science.sciencemag.org/ 635

5. Ulrich JD, Finn MB, Wang Y, Shen A, Mahan TE, Jiang H, et al. Altered microglial response to Abeta 636 plaques in APPPS1-21 mice heterozygous for TREM2. Mol Neurodegener. 2014;9: 20. 637 doi:10.1186/1750-1326-9-20 638

6. Sarlus H, Heneka MT. Microglia in Alzheimer’s disease. J Clin Invest. 2017;127: 3240–3249. 639 doi:10.1172/JCI90606 640

7. Kamphuis W, Kooijman L, Schetters S, Orre M, Hol EM. Transcriptional profiling of CD11c-positive 641 microglia accumulating around amyloid plaques in a mouse model for Alzheimer’s disease. Biochim 642 Biophys Acta. 2016;1862: 1847–1860. doi:10.1016/j.bbadis.2016.07.007 643

8. Keren-Shaul H, Spinrad A, Weiner A, Matcovitch-Natan O, Dvir-Szternfeld R, Ulland TK, et al. A 644 Unique Microglia Type Associated with Restricting Development of Alzheimer’s Disease. Cell. 645 2017;169: 1276-1290.e17. doi:10.1016/j.cell.2017.05.018 646


https://doi.org/10.1101/2020.03.18.995332

23

9. Krasemann S, Madore C, Cialic R, Baufeld C, Calcagno N, El Fatimy R, et al. The TREM2-APOE 647 Pathway Drives the Transcriptional Phenotype of Dysfunctional Microglia in Neurodegenerative 648 Diseases. Immunity. 2017;47: 566-581.e9. doi:10.1016/j.immuni.2017.08.008 649

10. Venegas C, Kumar S, Franklin BS, Dierkes T, Brinkschulte R, Tejera D, et al. Microglia-derived ASC 650 specks cross-seed amyloid-beta in Alzheimer’s disease. Nature. 2017;552: 355–361. 651 doi:10.1038/nature25158 652

11. Yin Z, Raj D, Saiepour N, Van Dam D, Brouwer N, Holtman IR, et al. Immune hyperreactivity of Abeta 653 plaque-associated microglia in Alzheimer’s disease. Neurobiol Aging. 2017;55: 115–122. 654 doi:10.1016/j.neurobiolaging.2017.03.021 655

12. Rogers J, Lue LF. Microglial chemotaxis, activation, and phagocytosis of amyloid beta-peptide as linked 656 phenomena in Alzheimer’s disease. Neurochem Int. 2001;39: 333–340. doi:10.1016/s0197-657 0186(01)00040-7 658

13. Spangenberg E, Severson PL, Hohsfield LA, Crapser J, Zhang J, Burton EA, et al. Sustained microglial 659 depletion with CSF1R inhibitor impairs parenchymal plaque development in an Alzheimer’s disease 660 model. Nat Commun. 2019;10: 3758. doi:10.1038/s41467-019-11674-z 661

14. Friedman BA, Srinivasan K, Ayalon G, Meilandt WJ, Lin H, Huntley MA, et al. Diverse Brain Myeloid 662 Expression Profiles Reveal Distinct Microglial Activation States and Aspects of Alzheimer’s Disease 663 Not Evident in Mouse Models. Cell Rep. 2018;22: 832–847. doi:10.1016/J.CELREP.2017.12.066 664

15. Sala Frigerio C, Wolfs L, Fattorelli N, Thrupp N, Voytyuk I, Schmidt I, et al. The Major Risk Factors for 665 Alzheimer’s Disease: Age, Sex, and Genes Modulate the Microglia Response to Abeta Plaques. Cell 666 Rep. 2019;27: 1293-1306.e6. doi:10.1016/j.celrep.2019.03.099 667

16. Olah M, Menon V, Habib N, Taga M, Yung C, Cimpean M, et al. A single cell-based atlas of human 668 microglial states reveals associations with neurological disorders and histopathological features of the 669 aging brain. 2018; 1–42. doi:10.1101/343780 670

17. Mathys H, Davila-Velderrain J, Peng Z, Gao F, Mohammadi S, Young JZ, et al. Single-cell 671 transcriptomic analysis of Alzheimer’s disease. Nature. 2019;570: 332–337. doi:10.1038/s41586-019-672 1195-2 673

18. Galatro TF, Vainchtein ID, Brouwer N, Boddeke EWGM, Eggen BJL. Isolation of Microglia and 674 Immune Infiltrates from Mouse and Primate Central Nervous System. Methods Mol Biol. 2017;1559: 675 333–342. doi:10.1007/978-1-4939-6786-5_23 676

19. Galatro TF, Holtman IR, Lerario AM, Vainchtein ID, Brouwer N, Sola PR, et al. Transcriptomic analysis 677 of purified human cortical microglia reveals age-associated changes. Nat Neurosci. 2017;20: 1162–1171. 678 doi:10.1038/nn.4597 679

20. Picelli S, Faridani OR, Bjorklund AK, Winberg G, Sagasser S, Sandberg R. Full-length RNA-seq from 680 single cells using Smart-seq2. Nat Protoc. 2014;9: 171–181. doi:10.1038/nprot.2014.006 681

21. Uniken Venema WT, Voskuil MD, Vila AV, van der Vries G, Jansen BH, Jabri B, et al. Single-Cell 682 RNA Sequencing of Blood and Ileal T Cells From Patients With Crohn’s Disease Reveals Tissue-683 Specific Characteristics and Drug Targets. Gastroenterology. 2019;156: 812-815.e22. 684 doi:10.1053/j.gastro.2018.10.046 685

22. Kim D, Langmead B, Salzberg SL. HISAT: a fast spliced aligner with low memory requirements. Nat 686 Methods. 2015;12: 357–360. doi:10.1038/nmeth.3317.HISAT 687


https://doi.org/10.1101/2020.03.18.995332

24

23. Liao Y, Smyth GK, Shi W. FeatureCounts: An efficient general purpose program for assigning sequence 688 reads to genomic features. Bioinformatics. 2014;30: 923–930. doi:10.1093/bioinformatics/btt656 689

24. Smith T, Heger A, Sudbery I. UMI-tools: modelling sequencing error in Unique Molecular Identifiers to 690 improve quantification. Cold Spring Harb Lab Press. 2017; 491–499. doi:10.1101/gr.209601.116 691

25. Griffiths JA, Richard AC, Bach K, Lun ATL, Marioni JC. Detection and removal of barcode swapping 692 in single-cell RNA-seq data. Nat Commun. 2018;9. doi:10.1038/s41467-018-05083-x 693

26. George NI, Chang C-W. DAFS: a data-adaptive flag method for RNA-sequencing data to differentiate 694 genes with low and high expression. BMC Bioinformatics. 2014;15: 92. doi:10.1186/1471-2105-15-92 695

27. Yu G, Wang L-G, Han Y, He Q-Y. clusterProfiler: an R package for comparing biological themes 696 among gene clusters. OMICS. 2012;16: 284–287. doi:10.1089/omi.2011.0118 697

28. Lake B, Shen R, Ronaghi M, Fan J, Wang W, Zhang K. Neuronal subtypes and diverstiy revealed by 698 single-nucleus RNA sequencing of human brain. 2016;35: 1586–90. 699

29. Streit WJ, Sammons NW, Kuhns AJ, Sparks DL. Dystrophic microglia in the aging human brain. Glia. 700 2004;45: 208–212. doi:10.1002/glia.10319 701

30. Streit WJ, Braak H, Xue Q-S, Bechmann I. Dystrophic (senescent) rather than activated microglial cells 702 are associated with tau pathology and likely precede neurodegeneration in Alzheimer’s disease. Acta 703 Neuropathol. 2009;118: 475–485. doi:10.1007/s00401-009-0556-6 704

31. Streit WJ, Braak H, Del Tredici K, Leyh J, Lier J, Khoshbouei H, et al. Microglial activation occurs late 705 during preclinical Alzheimer’s disease. Glia. 2018;66: 2550–2562. doi:10.1002/glia.23510 706

32. Drummond E, Wisniewski T. Alzheimer’s disease: experimental models and reality. Acta Neuropathol. 707 2017;133: 155–175. doi:10.1007/s00401-016-1662-x 708

33. Liu P, Reichl JH, Rao ER, McNellis BM, Huang ES, Hemmy LS, et al. Quantitative Comparison of 709 Dense-Core Amyloid Plaque Accumulation in Amyloid-beta Protein Precursor Transgenic Mice. J 710 Alzheimers Dis. 2017;56: 743–761. doi:10.3233/JAD-161027 711

34. Holtman IR, Raj DD, Miller JA, Schaafsma W, Yin Z, Brouwer N, et al. Induction of a common 712 microglia gene expression signature by aging and neurodegenerative conditions: a co-expression meta-713 analysis. Acta Neuropathol Commun. 2015;3: 31. doi:10.1186/s40478-015-0203-5 714

35. Srinivasan K, Friedman BA, Etxeberria A, Huntley MA, van der Brug MP, Foreman O, et al. 715 Alzheimer’s patient brain myeloid cells exhibit enhanced aging and unique transcriptional activation. 716 bioRxiv. 2019; 610345. doi:10.1101/610345 717

36. Gerrits E, Heng Y, Boddeke HWGM, Eggen BJL. Transcriptional profiling of microglia; current state of 718 the art and future perspectives. Glia. 2019. doi:10.1002/glia.23767 719

37. Lake BB, Codeluppi S, Yung YC, Gao D, Chun J, Kharchenko P V, et al. A comparative strategy for 720 single-nucleus and single-cell transcriptomes confirms accuracy in predicted cell-type expression from 721 nuclear RNA. Sci Rep. 2017;7. doi:10.1038/s41598-017-04426-w 722

38. Grindberg R V, Yee-Greenbaum JL, McConnell MJ, Novotny M, O’Shaughnessy AL, Lambert GM, et 723 al. RNA-sequencing from single nuclei. Proc Natl Acad Sci U S A. 2013;110: 19802–19807. 724 doi:10.1073/pnas.1319700110 725

726

Supplementary materials 727


https://doi.org/10.1101/2020.03.18.995332

25

Figure S1 Donor classification based on immunohistochemistry 728

(A,B) Representative immunohistochemical images of Aß and PHF-tau staining for the LPS (A) 729

and GFS (B) of the four groups (CTR, CTR+, AD, MCI). Scale bar= 100µm. AD: clinically 730

diagnosed Alzheimer’s Disease with Aβ plaques and/or hyperphosphorylated tau, CTR: Control, 731

CTR+: Control with Aβ plaques and/or hyperphosphorylated tau, MCI: clinically diagnosed 732

mild cognitive impairment with Aβ plaques and hyperphosphorylated tau. 733

Figure S2 bc-Smart-seq2 single cell RNA sequencing quality of donors and clusters 734

Boxplots displaying the number of detected UMIs, unique genes and mitochondrial genes per 735

donor (A,B,D) and per cluster (D,E,F). 736

Figure S3 Biological annotation of marker genes for each cluster 737

Dotplots displaying the gene ratio per gene ontology term for each cluster identified in bc-738

Smart-seq2 analysis. 739

Supplementary tables 740

Table S1 Detailed donor information of bulk sequenced samples. 741

Table S2 Detailed donor information of single-cell sequenced samples. 742

Table S3 Differential gene expression analysis of male vs. female bulk microglia GFS samples. 743

Table S4 Differential gene expression analysis of LPS vs. GFS bulk microglia samples. 744

Table S5 Differential gene expression analysis of enzymatic vs. mechanical dissociated bulk 745

microglia samples. 746

Table S6 Cluster markers of the identified 9 clusters with bc-Smart-seq2 single cell RNA 747

sequencing. 748

Table S7 Statistics of the average gene set expression changes in the clusters identified. 749

Table S8 Cluster markers of the identified 3 clusters of donor 2018-135 with 10X Genomics 750

single cell RNA sequencing. 751

Table S9 Cluster markers of the identified 3 clusters of donor 2019-10 with 10X Genomics 752

single cell RNA sequencing. 753

754

755

756


https://doi.org/10.1101/2020.03.18.995332

26

757


https://doi.org/10.1101/2020.03.18.995332

profiling microglia from ad donors and non-demented ... · 3/18/2020 · to examine the effect of...

Documents