Oligodendrocytes contain all components required to produce Aβ

To identify the cell types in the brain that are intrinsically capable of producing Aβ, we exploited four publicly available human single nucleus RNA sequencing (snRNA-seq) datasets [5,10–12] and examined expression levels of genes involved in the production of Aβ: the amyloid precursor protein (APP), beta secretase (BACE1), and components of gamma secretase [presenilin (PSEN) 1, PSEN2, nicastrin (NCSTN), APH1A, APH1B, PSENEN]. We found that, apart from PSEN2 (which is not essential for gamma secretase function when PSEN1 is present [13]), oligodendrocytes had elevated expression of all of these genes (Fig 1A–1D). Remarkably, many genes were expressed at a higher level in oligodendrocytes than in any other cell type in the brain, including notably neurons (Fig 1A–1D). With the exception of neurons and oligodendrocytes, no other cell type examined was found to highly express all of the genes required for Aβ production (Fig 1A–1D), suggesting that oligodendrocytes, uniquely among glia, have an intrinsic capacity to produce Aβ. To confirm that this is not limited to RNA, we exploited publicly available proteomics data derived from isolated mouse brain cells [14] and found that, at the protein level, oligodendrocytes indeed contain high levels of APP, BACE1, and PSEN1 (S1 Fig). We further confirmed the presence of APP and BACE1 protein in mouse oligodendrocytes by immunohistochemistry (Figs 1E, 1F, and S2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Components required to produce Aβ are expressed at high levels in oligodendrocytes, but not other glial cells.

Heatmaps showing the log₂ (norm count) z-score of genes of interest across different cell types [Excitatory neurons (Ex), Inhibitory neurons (In), Astrocytes (Ast), Microglia (Mic), Oligodendrocyte Precursor Cells (OPC), and Oligodendrocytes (Oli)], from 4 publicly available human single nucleus RNA sequencing datasets. APP, BACE1, and all components of γ-secretase (PSEN1, PSENEN, NCSTN, APH1A, APH1B) with the exception of PSEN2 (which is interchangeable with PSEN1) are expressed at high levels in oligodendrocytes, many at higher levels than any other cell type. (a) Data from Zhou and colleagues [10] was generated using tissue from the motor cortex of 36 subjects including controls, AD patients, and those carrying TREM2 variants. (b) Data from Bakken and colleagues [11] was generated using tissue from the motor cortex of 5 control subjects. (c) Data from Lake and colleagues [12] was generated using tissue from the frontal cortex of 6 control subjects. (d) Data from Mathys and colleagues [5] was generated using tissue from the prefrontal cortex of 48 subjects with varying degrees of AD-related pathology. (e) Representative immunofluorescent images showing APP (green), oligodendroglial marker Olig2 (red), and DAPI (nuclei; blue) in the cortex of a 4-month-old wild-type mouse. Scale bar = 10 μm. (f) Representative immunofluorescent images showing BACE1 (green), oligodendroglial marker Olig2 (red), and DAPI (nuclei; blue) in the cortex of a 4-month-old wild-type mouse. Scale bar = 10 μm.

https://doi.org/10.1371/journal.pbio.3002727.g001

Human AD brains possess more oligodendrocytes with the capacity to produce Aβ

To validate these findings, and to determine whether the capacity of oligodendrocytes to produce Aβ is altered in AD, we performed RNAscope in situ hybridization (ISH) on postmortem tissue from the brains of patients with sporadic AD (sAD) and controls for MBP (a gene expressed exclusively in oligodendrocytes in the central nervous system [15]), APP, and BACE1 (Fig 2A). We found that approximately 80% of oligodendrocytes express both APP and BACE1 in Layers 5/6 of both sAD patient and control brains, indicating that they are capable of producing Aβ (Fig 2B). Contrary to our expectation of oligodendrocyte loss associated with myelin loss in AD [16], we observed an increased number of oligodendrocytes in Layers 5/6 of the prefrontal cortex of brains from sAD patients (Fig 2C), which was not solely explained by age (S3A Fig) and was not seen in Layers 2/3 of the prefrontal cortex (S3B Fig). Combined with the equivalent proportion of oligodendrocytes expressing both APP and BACE1, this increase in oligodendrocyte density resulted in an increased number of Aβ-capable oligodendrocytes within the brains of sAD patients (Fig 2D). By examining all cells expressing both APP and BACE1, we observed an increase in the total number of APP⁺ BACE1⁺ cells in the brains of sAD patients (S3C Fig), which was driven by an elevated proportion of APP⁺ BACE1⁺ cells that are oligodendrocytes (S3D Fig). Notably, there was no change in the number of other cells (MBP^–) capable of producing Aβ (S3E Fig), which were almost exclusively neurons (labelled as RBFOX3⁺, a gene expressed in all neurons in the cortex; S3F–S3I Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Human sporadic AD brains have more oligodendrocytes capable of producing Aβ compared to controls.

(a) Fluorescence images from Layers 5/6 of control (top) and sporadic AD (sAD; bottom) postmortem human prefrontal cortex labelled for MBP (oligodendrocyte-specific gene; green), BACE1 (yellow), APP (red), Aβ (identified by 6E10-antibody; white), and DAPI (nuclei; blue). Aβ-capable oligodendrocytes (MBP⁺ BACE1⁺ APP⁺ nuclei) are marked with white arrowheads. Scale bar = 25 μm. (b) Quantification showing that approximately 80% of oligodendrocytes are capable of producing Aβ in both control and sAD brains. (c) Quantification showing an increase in the number of oligodendrocytes in Layers 5/6 of sAD brains. (d) Quantification showing significantly more Aβ-capable oligodendrocytes in sAD brains than controls. In (b–d), each data point represents a single brain (n = 4 control brains, n = 5 sAD brains) with bars representing mean ± SEM; unpaired t test: t(7) = 0.568, 3.058, 2.581 in (b–d), respectively. Source data are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3002727.g002

We additionally carried out densitometric analysis of APP and BACE1 spots within oligodendrocytes and neurons. On average, oligodendrocytes were found to express higher amounts of APP and BACE1 than neurons (S4A and S4B Fig), though neurons showed a greater variability in expression levels, with a subset of neurons showing very high expression yet others showing only minimal expression (S4C and S4D Fig). The observed increase in the number of Aβ-capable oligodendrocytes in sAD brains suggests that oligodendrocyte Aβ production may play an important, but hitherto unappreciated, role in the pathogenesis of the disease.

Human iPSC-derived oligodendrocytes produce Aβ

As well as expressing the components necessary for Aβ production, it is pathologically relevant to understand whether human oligodendrocytes do indeed produce Aβ. To determine this, we differentiated oligodendrocytes from human induced pluripotent stem cells (iPSCs) following a previously established protocol [17]. We used iPSCs derived from familial AD (fAD) patients as well as an isogenic control, to examine this question in well-established lines both with and without pathological mutations affecting Aβ production [18,19]; due to the rarity of iPSC lines with fAD mutations, we did not compare between different lines, but considered them together. After 30 days of maturation, these cultures contained 6.3% MBP⁺ oligodendrocytes and 87.3% oligodendrocyte precursor cells (OPCs)/pre-myelinating oligodendrocytes, but almost no other non-oligodendroglial cell types (Figs 3A and S5A–S5C), consistent with previous studies [17]. Across all lines tested, we found that human iPSC-derived oligodendrocytes indeed produced Aβ, and that this Aβ production was significantly reduced by administration of the BACE1 inhibitor NB-360 [20], indicating that Aβ production in oligodendrocytes is BACE1 dependent (Fig 3B). Notably, this reduction in Aβ production is similar to what has previously been shown in human iPSC-derived neurons upon administration of NB-360 or similar BACE inhibitors [21–24]; these data from human iPSCs are also consistent with, and translationally extend, previous findings from rat and human embryonic stem cell oligodendrocyte cultures [25,26].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Human oligodendrocytes produce soluble Aβ and aggregates at higher levels than neurons.

(a) Fluorescent image of human iPSC-derived oligodendrocyte culture immunolabelled for MBP (green), OLIG2 (marker of all oligodendroglia; red), and DAPI (nuclei; blue). This example shows multiple mature oligodendrocytes extending MBP⁺ myelin processes, while the majority of other cells are OLIG2⁺ MBP^- OPCs. Scale bar = 25 μm. (b) Quantification by ELISA showing a significant reduction in the amount of Aβ₄₀ produced (as a % of the amount produced prior to treatment) by human oligodendrocytes when treated with BACE1 inhibitor (NB-360) compared to vehicle control (DMSO). (c) ELISA data showing more Aβ₄₀ produced by oligodendrocytes than neurons derived from the same fAD human-iPSC lines. (d) Quantification by ELISA showing higher Aβ₄₂/Aβ₄₀ ratio produced by oligodendrocytes compared to neurons derived from the same fAD human-iPSC lines. (e) DNA-PAINT super-resolution images showing more Aβ aggregates in media from oligodendrocytes (right) compared to neurons (left). Scale bar = 1 μm. (f) Quantification showing oligodendrocytes produce a higher proportion of Aβ as aggregates compared to neurons derived from the same fAD human-iPSC lines. (g) Violin plots showing the length of aggregates produced by neurons and oligodendrocytes (mean ± standard deviation: Neurons 79.8 nm ± 58.7 nm, Oligodendrocytes 77.7 nm ± 58.5 nm; n = 18,550 aggregates from 3 neuron lines [3 independent inductions per line] and 24,719 aggregates from 3 oligodendrocyte lines [3 independent inductions per line]). In (b), each data point represents the average of 2 independent inductions from each of a different cell line (n = 3 cell lines), with bars showing mean ± SEM; paired t test: t(2) = 9.613. In (c, d, f), each data point represents the average of 4 (for PSEN1 WT line neurons in d) or 3 independent inductions from each of a different cell line (n = 3 cell lines), showing mean ± SEM with each cell line shown in a different colour to highlight pairing (blue: PSEN1 WT; green: PSEN1 int4del; red: PSEN1 R278I); paired t test: t(2) = 4.435, 5.411 in (c, d), respectively; ratio paired t test: t(2) = 7.117 in (f). Un-pooled data for (b, c, d, f) are shown in S6 Fig. Source data are available in S1 Data. fAD, familial AD; iPSC, induced pluripotent stem cell; MBP, myelin basic protein; OPC, oligodendrocyte precursor cell.

https://doi.org/10.1371/journal.pbio.3002727.g003

Next, to understand how oligodendrocyte Aβ production compares to neuronal Aβ production, we generated human iPSC-derived cortical neurons from the same cell lines following a previously established protocol [27] (S5D–S5G Fig), and normalised the amount of Aβ produced to cell number. Under the experimental conditions tested, we found that oligodendrocytes produce more Aβ than neurons derived from the same cell line, both in lines from fAD patients with PSEN1 mutations and an isogenic control (Fig 3C). We additionally quantified Aβ production by human iPSC-derived OPCs, microglia and astrocytes (S7A–S7F Fig) and found that these cells produce only low amounts of Aβ (S7G Fig), consistent with our analyses of snRNA-seq data (Fig 1A–1D).

We next compared the Aβ species produced between the cell types and found that oligodendrocytes had an increased Aβ 42:40 ratio relative to neurons from the same line (Fig 3D). This observation suggests that oligodendrocytes could potentially generate a relatively higher proportion of longer Aβ species, which are more prone to aggregation [28] and, consequently, could be more pathogenic than shorter Aβ species. We therefore used a single-molecule pull-down (SiMPull) assay with DNA-PAINT super-resolution imaging [29] to analyse soluble Aβ aggregates, of a size consistent with oligomers and protofibrils, in the media from these cells (Figs 3E, S8A, and S8B). This technique relies on dual binding of a single epitope antibody and clustering analysis requiring close proximity of at least 2 of these super-resolved assemblies to ensure specific detection of soluble aggregates, and has previously been validated to detect stable soluble aggregates with a high sensitivity [29–31] (see Methods for details). Intriguingly, our analysis revealed that oligodendrocytes produced a greater quantity of these soluble Aβ aggregates than neurons, exceeding the levels that would be anticipated based on the increased amount of Aβ produced (Fig 3E and 3F). Furthermore, we found that the majority of aggregates produced by both oligodendrocytes and neurons were between 20 nm and 200 nm in length (Figs 3G and S8C), consistent with the reported size of synaptotoxic Aβ aggregates from human AD brains [32]. These results suggest that Aβ produced by oligodendrocytes has a higher propensity for aggregation than neuron-derived Aβ.

Oligodendrocyte-derived Aβ contributes to plaque formation in vivo

To directly investigate whether these features of oligodendrocyte-produced Aβ are consequential for Aβ plaque formation in vivo, we generated App^NL-G-F knock-in mice [33] with BACE1 knocked out specifically in either oligodendrocytes (BACE1^fl/fl;PLP1-Cre/ERT^+/-;App^NL-G-F) or neurons (BACE1^fl/fl;Thy1-Cre/ERT2,-EYFP^+/-;App^NL-G-F). We elected to use the App^NL-G-F knock-in mouse model for this study as it expresses App under its endogenous promoter in a physiological manner with regards to both cell type and amount, and avoids the issue of APP overexpression and abnormal expression patterns. To avoid developmental and early postnatal effects of BACE1 knockout on myelin [34], we used tamoxifen inducible Cre lines, with tamoxifen administered between the ages of 4 and 8 weeks (after the majority of developmental myelination has taken place [35]), according to standard protocols [36]. Indeed, several control experiments confirmed that this approach had no significant effect on the amount of MBP (S9A and S9B Fig), myelin sheath lengths (S9C and S9D Fig), the number of oligodendrocytes (S10A–S10C Fig), or oligodendrocyte autophagy (S11A–S11F Fig).

We assessed Aβ plaque load within the visual, retrosplenial, and motor cortex of these mice at 4 months of age, when there is already widespread plaque distribution within the cortex of App^NL-G-F mice [33]. We found that oligodendrocyte-specific knockout of BACE1 led to approximately 25% reduction in the number of plaques across the combined area of these cortical regions compared to unmodified App^NL-G-F control mice (Fig 4A and 4B), while knockout of BACE1 specifically in neurons led to a near elimination of plaques within the cortex (Fig 4A–4C). Notably, oligodendrocyte-specific knockout of BACE1 led to a greater reduction in plaque area in cortical Layers 5/6 as compared to Layers 2/3 specifically within the retrosplenial cortex, which is known to be a selectively vulnerable site of early plaque deposition and functional impairment in AD [37] (S12A and S12E Fig), albeit not within the motor cortex (S12B and S12F Fig). A similar trend in 20% to 30% plaque reduction was also observed in the CA1 area of the hippocampus and the corpus callosum, a major white matter tract (S12C, S12D, S12G, and S12H Fig). These results are consistent with our analysis of snRNA-seq data, where oligodendrocytes had high expression of genes required to produce Aβ in all of the datasets from different brain regions (Fig 1A–1D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Genetic suppression of oligodendrocyte Aβ production reduces Aβ plaques in the App^NL-G-F mouse model of AD.

(a) Immunofluorescent images showing Aβ (6E10 antibody; green) and DAPI in the retrosplenial cortex of App^NL-G-F control mice (left), App^NL-G-F mice with BACE1 knocked out (KO) specifically in oligodendrocytes (middle; Oligo-KO), and App^NL-G-F mice with BACE1 KO specifically in neurons (right; Neuron-KO). Images show all layers of the cortex, with Layer 6 at the bottom. Scale bar = 100 μm. (b) Quantification of the number of Aβ⁺ plaques across the visual, retrosplenial, and motor cortical areas, showing a 25% reduction in Oligo-KO mice compared to App^NL-G-F, and an elimination of plaques in Neuron-KO mice. (c) Quantification of the total area of Aβ⁺ plaques across the visual, retrosplenial, and somatomotor cortical areas, showing approximately 25% reduction in Oligo-KO mice compared to App^NL-G-F and an elimination of plaques in Neuron-KO mice. In (b and c), data points represent individual mice (n = 8 App^NL-G-F, 9 Oligo-KO, 4 Neuron-KO) with bars showing mean ± SEM. One-way ANOVA with Dunnet’s post hoc tests: F(2,18) = 25.38(b), 13.24(c); p < 0.0001(b), p = 0.0003(c). Source data are available in S1 Data.

https://doi.org/10.1371/journal.pbio.3002727.g004

Suppression of oligodendrocyte-derived Aβ rescues neuronal dysfunction in App^NL-G-F mice in vivo

To determine the consequences of BACE1 knockout in oligodendrocytes on neuronal dysfunction in vivo, we used high-density Neuropixels probes to record ongoing neuronal action potential firing in the retrosplenial cortex of 3-month-old awake mice. We found that that oligodendrocyte-specific knockout of BACE1 abolished the early abnormal neuronal hyperactivity phenotype that is present in the App^NL-G-F mice, which has been shown to be dependent on soluble Aβ [38–40] (Fig 5A and 5B). Notably, in App^NL-G-F mice with an oligodendrocyte-specific knockout of BACE1, we observed that not only were the levels of neuronal firing restored to those seen in WT controls, but so too also the temporal structure, as indicated by the analysis of the variability in mean firing rates (MFRs) and inter-spike intervals (S13 Fig). We also analysed the power spectrum of cortical local field potentials (LFPs) in the BACE1 knockout mice but found no significant changes compared to controls (S14 Fig). In addition, analyses of multiunit neuronal responses in retrosplenial cortex to sharp-wave ripple (SWR) events in CA1 during awake rest [41] did not reveal any notable impairments in the timing or magnitude of cortical neuronal responses in the BACE1 knockout animals compared to controls (S15 Fig). Together, these findings indicate that BACE1 knockout in oligodendrocytes effectively ameliorates abnormal neuronal hyperactivity in App^NL-G-F mice without disrupting the electrophysiological integrity of the cortical-hippocampal network.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Genetic suppression of oligodendrocyte Aβ production rescues neuronal dysfunction in the App^NL-G-F mouse model of AD, while oligodendrocyte-derived media promotes neuronal dysfunction in vivo.

(a) Raster plots from Neuropixels recordings showing spontaneous neuronal firing in 20 randomly selected cortical neurons/units from 3-month-old awake WT (top), App^NL-G-F (middle) and oligodendrocyte BACE1 KO (bottom) mice, illustrating rescue of hyperactivity phenotype in oligodendrocyte BACE1 KO mice to WT levels. Units are sorted from high MFRs to low (top to bottom) and plots show a 6s resting state period. (b) Quantification of the MFR showing significantly reduced activity in cortical neurons of both Oligo-KO and Neuron-KO mice compared to App^NL-G-F mice, with firing rates returning to levels observed in WT controls (WT vs. Oligo-KO: p > 0.9999; WT vs. Neuron-KO: p > 0.9999). In (b), data points represent individual neurons/units (n = 82 WT, 134 App^NL-G-F, 165 Oligo-KO, 75 Neuron-KO) across 4 (WT) or 3 mice per group with the shaded area representing smoothed distribution. Median is shown by a thick black line, and quartiles indicated with grey lines. Kruskal–Wallis test (H(3, n = 456) = 15.33, p = 0.0016) with Dunn’s post hoc tests. (c) Raster plots showing neuronal firing in the same 20 cortical neurons/units at baseline (left) and during injection of oligodendrocyte conditioned media containing soluble Aβ aggregates (right) in a 4-month-old WT mouse, illustrating the strong increase in neuronal activity upon exposure to oligodendrocyte-derived Aβ. (d) Quantification showing an increase in neuronal firing rates upon local injection of oligodendrocyte conditioned media (Aβ₄₀ concentration by ELISA: 133 pM) into retrosplenial cortex, compared to injection of either the same media which had been immunodepleted of Aβ (Aβ₄₀ concentration by ELISA: 24 pM) or media from BACE1 inhibitor-treated oligodendrocytes (Aβ₄₀ concentration by ELISA: 18 pM). In (d), data points represent individual mice (n = 5 oligodendrocyte conditioned media, 4 other groups) with mean ± SEM shown. One-way ANOVA (F(2,10) = 11.23, p = 0.0028) with Tukey’s post hoc tests. Source data are available in S1 Data. AD, Alzheimer’s disease; KO, knockout; MFR, mean firing rate.

https://doi.org/10.1371/journal.pbio.3002727.g005

Our results that targeted suppression of Aβ production in oligodendrocytes can rescue neuronal hyperactivity, even without the complete elimination of plaques, indicated that soluble Aβ species produced by oligodendrocytes may in fact contribute to early neuronal dysfunction in the AD brain. To directly test this hypothesis, we injected human fAD iPSC-derived oligodendrocyte conditioned media containing soluble Aβ aggregates into the retrosplenial cortex of WT mice while recording ongoing neuronal action potential firing using Neuropixels in vivo. As controls, we administered the same media immunodepleted of Aβ using 6E10 antibody, or media from oligodendrocytes which had been treated with BACE1 inhibitor in order to suppress Aβ production. Indeed, we found that neuronal firing rates were markedly increased during injection of the oligodendrocyte conditioned media compared to baseline, which was not observed with control media (Fig 5C and 5D). These results suggest that oligodendrocyte-derived Aβ is sufficient to promote neuronal hyperactivity, in the absence of Aβ from any other cellular source.

Animals

All experimental procedures were conducted in accordance with the Animal (Scientific procedures) Act 1986, approved by the Animal Welfare and Ethical Review Body at University College London (UCL), and performed under an approved UK Home Office project licence at UCL (PP8718352 to MAB). Mice were maintained in a 12-h light/dark cycle with food and water supplied ad libitum. App^NL-G-F mice [C57BL/6-App<tm3(NL-G-F)Tcs> (No.RBRC06344)] were provided by the RIKEN BRC through the National BioResource Project of the MEXT, Japan [33]. BACE1^fl/fl mice (C57BL/6-Bace1tm1.1mrl) were previously generated and provided by R.V. [36]. PLP1-Cre/ERT mice [B6.Cg-Tg(Plp1-cre/ERT)3Pop/J; JAX:005975] and Thy1-Cre/ERT2,-EYFP mice [B6.Tg(Thy1-cre/ERT2,-EYFP)HGfng/PyngJ; JAX:012708] were purchased from The Jackson Laboratory. Three mouse cohorts were generated: (1) BACE1^fl/fl;PLP-Cre/ERT^+/-;App^NL-G-F; (2) BACE1^fl/fl;Thy1-Cre/ERT,-EYFP^+/-;App^NL-G-F; and (3) BACE1^fl/fl;App^NL-G-F (i.e., the Cre/ERT^-/- littermates from the other 2 cohorts). All 3 cohorts were treated with tamoxifen between 4 and 8 weeks of age: tamoxifen (MP Biomedicals; 156738) was dissolved in corn oil at 25 mg/ml. Animals were injected with tamoxifen intraperitoneally at a dose of 100 mg/kg daily for 5 days followed by a 2 day break, repeated 4 times. For Neuropixels recordings, C57BL/6J mice (originally purchased from Charles River Laboratories) were used as wild type (WT) controls. For acute recordings, animals were aged between 4 and 5 months. All other animals were aged between 3 and 4 months at the time of Neuropixels recordings or perfusion for fixed tissue analyses. Data presented shows the means of both male and female animals.

Immunohistochemistry (IHC)

Brains were extracted from animals after intracardiac perfusion of 1× phosphate-buffered saline (no calcium, no magnesium; PBS; Gibco), post-fixed for 24 h in 4% paraformaldehyde (PFA) in PBS (Alfa Aesar), cryopreserved through increasing sucrose concentrations, embedded in optimal cutting temperature compound (OCT; CellPath), and frozen on dry ice. Amyloid plaque and oligodendrocyte analyses: 10-μm thick sagittal sections were cut using a Leica cryostat. Sections were blocked in blocking solution (10% donkey or goat serum, 0.3% triton in PBS) for 3 h then incubated with Alexa Fluor 594 anti-β-amyloid (Clone 6E10; 1:1,000; BioLegend; 803019), Olig2 (1:500; Millipore; MABN50), ASPA (1:1,000; Sigma-Aldrich; ABN1698), APP (Y188; 1:500; Abcam; ab32136), and/or BACE1 (1:250; Abcam; ab108394) antibodies in block overnight at 4°C. Where appropriate, sections were washed and incubated with secondary antibodies [Goat anti-Mouse Alexa Fluor 488 (1:1,000; Invitrogen; A11001), Goat anti-Rabbit Alexa Fluor 594 (1:1,000; Invitrogen; A11012), Goat anti-Mouse Alexa Fluor 594 (1:1,000; Invitrogen; A11005), Goat anti-Rabbit Alexa Fluor 488 (1:1,000; Invitrogen; A11008)] in block for 2 h at room temperature (RT). Sections were then labelled with DAPI and mounted with ProLong Gold. Myelin sheath analysis: 30-μm thick coronal brain sections were cut between bregma +1.1 mm and +0.8 mm. Free-floating coronal sections were blocked in 10% fetal calf serum, 0.1% triton in PBS for 1 h then incubated with rat anti-MBP antibody (1:500; Abcam; ab7349) in blocking solution overnight at 4°C. Sections were washed in PBS before adding Donkey anti-Rat Alexa Fluor 488 secondary antibody (1:1,000; Invitrogen; A21208) for 1 h at RT. Sections were washed and stained for nuclei before being mounted onto slides. Oligodendrocyte autophagy analysis: 40-μm thick free-floating coronal sections were washed in PBS, blocked in M.O.M. blocking solution (Vector laboratories) for 90 min, then incubated in primary antibodies [Olig2 (1:500; Millipore; MABN50), LC3 (1:1,000; Sigma-Aldrich; L7543), Lamp2 (1:2,000; Hybridoma Bank; ABL-93), p62 (1:500; Abcam; ab91526), Ctsd (ThermoFisher; PA5-47046)] overnight at 4°C. Sections were then washed in PBS followed by incubation with secondary antibodies (Donkey anti-Mouse 488, Donkey anti-Mouse 568, Donkey anti-Rabbit 657, Donkey anti-Rat 488, Donkey anti-Goat 568; all at 1:1,000; Invitrogen) for 2 h at RT. Sections were then washed, labelled with DAPI and mounted onto slides.

Western blotting

Brains were extracted from animals after intracardiac perfusion of 1× PBS, the forebrain was dissected and protein extracted in RIPA buffer (Pierce). Protein in each sample was quantified using a BCA Assay (Pierce) and 20 μg of protein was loaded per well of a NuPAGE 4% to 12% Bis-Tris gel. Samples were run for 1.5 h at 150 V, before transfer to a nitrocellulose membrane using an iBlot 2 transfer system (Thermo Fisher Scientific). Membranes were blocked for 1 h in 1% casein (Bio-Rad) in PBST (0.1% Tween-20 in PBS), then incubated with MBP (1:1,000; Bio-Rad; MCA409S) antibody overnight at 4°C. Membranes were washed in PBST, incubated with Goat anti-rat ECL antibody (1:10,000; Amersham; NA935) for 1 h, then imaged using Pierce ECL substrate and an Amersham Imager 680. Membranes were subsequently stripped (Pierce) and reprobed for GAPDH [(1:1,000; Sigma-Aldrich; MAB374); secondary: Sheep anti-mouse ECL (1:10,000; Amersham; NA931)]. Blots were analysed using the Gel Analyzer tools in the FIJI build of ImageJ (open source). Intensity of MBP bands was normalised to the GAPDH band for each lane, and all lanes were normalised to the average of the App^NL-G-F values.

Neuropixels recordings and analysis

Human tissue

The study was carried out in accordance with the Declaration of Helsinki. Postmortem human brain tissue from prefrontal cortex (Brodmann’s Area 10) was provided as precut frozen tissue sections by the Queen Square Brain Bank (who obtained the tissue with written informed consent from the donor, next of kin or person with power of attorney) with ethical committee approval (reference UCLMTA4-20), and the tissue is stored under licence from the Human Tissue Authority (licence number 12198). AD tissue is from donors with a clinical diagnosis of sporadic AD and a high level of AD pathological change (A3, B3, C2, or C3). Control tissue is from donors with no clinical diagnosis of neurological disease and no or minimal AD-related pathology. Age: control, 90 ± 7 years (mean ± SD) and AD, 75 ±10; postmortem delay: control, 69 ± 25 h and AD, 59 ± 28; percentage male: control, 50% and AD, 80%.

RNAscope in situ hybridization (ISH)

ISH was carried out using RNAscope Multiplex Fluorescent Kit v2 [Advanced Cell Diagnostics (ACD)] according to the manufacturer’s instructions. Briefly, frozen 10-μm thick human tissue sections were dried (4 min at 40°C then 10 min at RT) before fixation in 4% PFA in PBS (30 min at RT). Sections were then dehydrated through increasing ethanol concentrations (5 min each in 50%, 70%, 100%, 100% ethanol) before incubation with hydrogen peroxide (10 min at RT) followed by incubation with Protease IV (20 min at RT). Hybridization of the probes [Hs-APP (418321; ACD), Hs-BACE1-C2 (422541-C2; ACD), and Hs-MBP-C3 (411051-C3; ACD) or Hs-RBFOX3-C3 (415591-C3; ACD) at a ratio of 50:1:1] was carried out at 40°C for 2 h. Sections then underwent amplification (30 min at 40°C for each probe sequentially) followed by sequential detection of each probe [15 min at 40°C with probe-specific HRP then 30 min at 40°C with Opal dye (1:500; C1: Opal 570, C2: Opal 620, C3: Opal 520; Akoya Biosciences)]. Following ISH, sections underwent IHC for Aβ: Sections were blocked for 1 h in blocking solution (10% goat serum, 0.1% triton in PBS), incubated with primary antibody in block (anti-β-amyloid clone 6E10; 1:2,000; BioLegend; 803001) overnight at 4°C, then incubated with secondary antibody in block [Goat anti-Mouse Alexa Fluor 647 (1:1,000; Invitrogen; A21235)] for 2 h at RT. Sections were incubated with TrueBlack (1X; Biotium; 23007) for 30 s and counterstained with DAPI (1:5,000) prior to mounting with ProLong Gold (Invitrogen).

Induced pluripotent stem cells (iPSCs)

iPSCs were previously obtained or generated, and validated, by CA and SW [18,19]. iPSCs were cultured in Essential 8 media on Geltrex and passaged using 0.5 mM EDTA (all reagents from Thermo Fisher). The following cell lines were used in this study: APP V717I 1 (APP V717I mutation), PSEN1 R278I (PSEN1 R278I mutation), PSEN1 int4del (intron 4 deletion in PSEN1), PSEN1 WT (PSEN1 int4del line CRISPR edited to correct the mutation [19]).

iPSC-derived cortical neurons

Cortical neurons were derived from iPSCs as previously described [27]. Briefly, iPSCs underwent neural induction in N2B27 media [1:1 DMEM-F12 Glutamax (Gibco): Neurobasal (Gibco) with 0.5X B27 supplement (Gibco), 0.5X N2 supplement (Gibco), 0.5X Non-essential amino acids (Gibco), 50 μM 2-mercaptoethanol (Gibco), 2.5 μg/ml insulin (Sigma-Aldrich), 50 U/ml Penicillin (Gibco), 50 μg/ml Streptomycin (Gibco), 1 mM L-glutamine (Gibco)] supplemented with 1 μM Dorsomorphin (Tocris) and 10 μM SB-431542 (Generon) for 10 days. Cells were then expanded over 12 to 18 days in neural maintenance media, and split every 6 days using dispase (Gibco) to remove non-neuronal cell types. Cells were then singularised using accutase (Sigma-Aldrich) before final plating in poly-L-ornithine (Sigma-Aldrich) and laminin (Sigma-Aldrich)-coated wells. Mature cortical neurons begin to appear from 60 days after final plating, and media which had been on the cells for 24 h was collected for Aβ quantification between 120 and 200 days after final plating. To normalise for cell number, cells were lysed at the end of the experiment, total genomic DNA extracted and quantified, and cell number calculated based on each cell containing 6 pg of DNA.

iPSC-derived oligodendrocytes and OPCs

Oligodendrocytes were derived from iPSCs as previously described [17]. Briefly, neural precursor cells (NPCs) were first derived from iPSCs as previously described [27,62]: iPSC colonies were grown in N2B27 media supplemented with 1 μM Dorsomorphin (Tocris), 10 μM SB-431542 (Generon), 3 μM CHIR 99021 (MedChemExpress), and 0.5 μM purmorphamine (Peprotech). After 4 days, Dorsomorphin and SB-431542 were withdrawn and 150 μM ascorbic acid was added. Cells were plated on Geltrex (Gibco) and grown for 30 days, with passaging every 5 to 6 days, to expand and remove contaminating cell types. NPCs were then plated on Geltrex at a density of 100,000 cells/well and expression of SOX10, OLIG2, and NKX6.2 was driven by transduction of the cells with SON lentivirus (plasmid kindly provided by Prof. Tanja Kuhlmann, University of Münster; virus constructed by VectorBuilder) at an MOI (multiplicity of infection) of 5, together with 5 μg/ml protamine sulfate (Sigma-Aldrich) for 24 h. Cells were then grown in glial induction medium [GIM; DMEM-F12 glutamax with 0.5X B27 minus vitamin A supplement, 0.5X N2 supplement, 100 U/ml Penicillin, 100 μg/ml Streptomycin, 2 mM L-glutamine, 1 μM SAG (Cayman Chemical), 10 ng/ml PDGF (Peprotech), 10 ng/ml NT3 (Peprotech), 10 ng/ml IGF-1 (Peprotech), 200 μM ascorbic acid, 0.1% Trace Elements B (Corning), 10 ng/ml T3 (Sigma-Aldrich)]. After 4 days, cells were changed into differentiation media [DM; DMEM-F12 glutamax with 0.5X B27 minus vitamin A supplement, 0.5X N2 supplement, 100 U/ml Penicillin, 100 μg/ml Streptomycin, 2 mM L-glutamine, 60 ng/ml T3, 10 ng/ml NT3, 10 ng/ml IGF-1, 200 μM ascorbic acid, 0.1% Trace Elements B, 100 μM dbcAMP (Sigma-Aldrich)]. After 7 to 10 days of differentiation, cells were singularised with accutase and replated on Geltrex at a density of 250,000 cells/well (12-well plate) or 150,000 cells/well (glass coverslips in 24-well plate), and 35 days after glial induction, almost all cells were oligodendroglia (OLIG2⁺ and/or MBP⁺) and cells with mature MBP⁺ sheets were visible. Media which had been on the cells for 24 h was collected for Aβ quantification between 40 and 120 days after glial induction. For immature OPCs, cultures were maintained in GIM, and media which had been on the cells for 24 h was collected for Aβ quantification at day 5 after glial induction. To normalise for cell number, cells were lysed at the end of the experiment, total genomic DNA extracted and quantified, and cell number calculated based on each cell containing 6 pg of DNA.

iPSC-derived astrocytes

iPSC-derived astrocytes were generated using an established protocol [63]. Briefly, glial progenitor cells were enriched from cortical neuronal cultures (see above) in a proliferative phase [N2B27 media plus 10 ng/ml FGF2 (Peprotech)] until day 150 of differentiation. During this phase, glial progenitors were routinely passaged using 0.5 mM EDTA and cultured on Geltrex substrate (Thermo Fisher). A final two-week maturation phase consisted of 10 ng/ml LIF (Sigma) and 10 ng/ml BMP4 (Peprotech), at the end of which media which had been on the cells for 24 h was collected for Aβ quantification. To normalise for cell number, cells were lysed at the end of the experiment, total genomic DNA extracted and quantified, and cell number calculated based on each cell containing 6 pg of DNA.

iPSC-derived microglia

iPSC-derived microglial-like cells were generated using established protocols [64,65]. Briefly, embryoid bodies were generated from 10,000 iPSCs and haematopoietic differentiation was initiated via 3 days of 50 ng/ml BMP4, 50 ng/ml VEGF, and 20 ng/ml SCF followed by maintenance in myeloid media consisting of x-vivo media with 25 ng/ml IL3 and 100 ng/ml MCSF. Microglial-like cells were harvested and exposed to a two-week final maturation in 25 ng/ml MCSF, 100 ng/ml IL34, 5 ng/ml TGF-β and further supplemented with CX3CL1 and CD200 for the final 2 days (both 100 ng/ml), at the end of which media was collected for Aβ quantification. All growth factors Peprotech unless stated. To normalise for cell number, cells were lysed at the end of the experiment, total genomic DNA extracted and quantified, and cell number calculated based on each cell containing 6 pg of DNA.

BACE1 inhibitor treatment

BACE1 inhibitor NB-360 (a kind gift from Novartis) was prepared at a concentration of 10 μM in differentiation medium (DM), from a stock solution of 10 mM prepared in dimethyl sulfoxide (DMSO; Sigma-Aldrich). Cells between 40 and 60 days after glial induction which had been in their medium for 24 h had their medium collected (pretreatment) and medium containing 10 μM NB-360 was added. After 24 h of treatment, the medium was collected (posttreatment) and replaced with regular DM. As vehicle controls, cells were in parallel treated with DM containing 0.1% DMSO. Aβ concentrations were quantified in both pretreatment and posttreatment media samples to calculate the change in Aβ production for each well of cells.

DNA extraction

DNA was extracted using the QIAGEN DNeasy Blood and Tissue Kit (69504) according to the manufacturer’s instructions. Briefly, cells were collected and suspended in PBS with 10% Proteinase K. Lysis buffer was added and the samples were incubated at 56°C for 10 min. Ethanol was added and the lysate was added to a DNeasy Mini spin column where the DNA was bound to a membrane. The DNA was twice washed before being eluted. DNA concentrations were measured using a Nanodrop One (Thermo Fisher).

Immunocytochemistry (ICC)

Cells were fixed in 4% PFA in PBS for 10 min, washed in PBS, then incubated with blocking solution (10% goat serum, 2% bovine serum albumin, 0.2% triton in PBS) for 2 h. Cells were then incubated in primary antibodies [MBP (1:50; Abcam; ab7349), OLIG2 (1:100; Atlas Antibodies; HPA003254), GFAP (1:500; Invitrogen; 14-9892-82), anti-Tubulin beta 3 (clone TUJ1; 1:1,000; BioLegend; 801202), IBA1 (1:500; Wako; 019-19741)] in block overnight at 4°C, washed in PBS, then incubated in secondary antibodies [Goat anti-Rat Alexa Fluor 647 (1:1,000; Invitrogen; A21247), Goat anti-Mouse Alexa Fluor 488 (1:1,000; Invitrogen; A11001), Goat anti-Rabbit Alexa Fluor 594 (1:1,000; Invitrogen; A11012), Goat anti-Rabbit Alexa Fluor 488 (1:1,000; Invitrogen; A11008)] for 1 h at RT. Cells were then labelled with DAPI.

qPCR

Total RNA was extracted from iPSC and iPSC-derived cortical neuron cells using a RNeasy Mini kit (QIAGEN; 74104), including a DNase treatment step, according to the manufacturer’s instructions. RNA was reverse-transcribed using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Carlsbad, California, United States of America) according to the manufacturer’s instructions. Quantitative analysis was carried out using qPCR with gene-specific SYBR green forward (Fwd) and reverse (Rev) primers (TBR1: Fwd AGCAGCAAGATCAAAAGTGAGC, Rev ATCCACAGACCCCCTCACTAG; CTIP2: Fwd CTCCGAGCTCAGGAAAGTGTC, Rev TCATCTTTACCTGCAATGTTCTCC; RPL18A: Fwd CCCACAACATGTACCGGGAA, Rev TCTTGGAGTCGTGGAACTGC) using a Roche LightCycler 480 II. Gene expression in all samples was quantified using the ΔΔCt method normalised first to the housekeeping gene RPL18A, then to the average for all iPSC samples.

ELISA

ELISAs for human Aβ40 and human Aβ42 were carried out using commercial kits (Aβ40: Invitrogen KHB3481; Aβ42: Invitrogen KHB3441) according to the manufacturer’s instructions. Briefly, samples were added to the wells with capture antibody prebound, together with detection antibody, and incubated for 3 h. Wells were then washed before being incubated with anti-rabbit IgG HRP for 30 min. Wells were again washed, incubated with stabilised chromogen for 30 min, and the reaction stopped. Absorbance at 450 nm was measured using a FLUOstar Omega microplate reader.

Immunodepletion of Aβ

Oligodendrocyte-conditioned media was incubated with anti-Aβ antibody 6E10 (1:500; BioLegend; 803001) for 2 h at 4°C. Samples were then incubated with prewashed Sepharose Protein A/G beads (50 μl; Rockland; PAG50-00-0002) for 2 h at 4°C. Beads with bound Aβ and antibody were then removed by centrifugation for 30 s at 12,500 RPM.

Quantification of soluble Aβ aggregates

The technique used to quantify soluble aggregates has previously been validated in multiple studies and has been shown to specifically detect aggregates, and not monomers [29–31]. It relies on the binding of 2 copies of the same antibody (6E10, which only has a single epitope [66]), ensuring that what are detected are at least dimers. The clustering analysis used requires close proximity of at least 2 of these dimers, thus further ensuring that what are detected are aggregates. This method allows quantification of stable assemblies or aggregates at picomolar concentrations, which is a much higher sensitivity than bulk biochemical methods.

Single-molecule pull-down coverslip preparation was performed as previously described [67]. Briefly, neutravidin (0.2 mg/ml) in TBS with 0.05% Tween 20 (TBST) was added to glass coverslips covalently mounted with polyethylene glycol (PEG) and biotin for 10 min, followed by 2 wash steps with TBST and once with TBS with 1% Tween 20 (1%T). Afterwards, biotinylated 6E10 antibody (10 nM; BioLegend; 803007) was added for 15 min, followed by 2 wash steps with TBST and once with 1%T. The media samples were added and incubated overnight at 4°C followed by 2 wash steps with TBST and once with 1%T. The coverslips were then incubated with 6E10 antibody (500 pM; BioLegend; 803020), labelled with single-strand DNA [68] (ACCACCA) for 45 min, followed by 2 wash steps with TBST and once with 1%T. After washing, TetraSpeck microspheres (1:7,000 in TBS, 10 μl, Thermo Scientific, Cat. T7279) were introduced to each well for 10 min. The TetraSpeck solution was then removed, followed by 2× wash with TBST, and a second PDMS gasket (Merck, GBL-103250-10EA) was stacked on the coverslip before introducing 3 μl of complimentary imaging strand (TGGTGGT- cy3B; atdbio) in TBS. Finally, the coverslip was sealed with another coverslip on top of the second PDMS gasket. Then, the coverslips were imaged on a purpose built [67] TIRF microscope using a 520 nm laser, with 100 milliseconds of exposure for 4,000 frames for each field of view. Three fields of view were acquired from each well. Acquired frames were stacked, reconstructed, drift corrected, and analysed for the number of aggregates in each field of view, as well as aggregate length using the ACT software [29]. Images of individual aggregates were generated using the ASAP software [69].

Analysis of single nucleus RNA sequencing data

Human single nucleus RNA sequencing data from Zhou and colleagues [10] were obtained via the AD Knowledge Portal (https://adknowledgeportal.org) study snRNAseqAD_TREM2. Human single nucleus RNA sequencing data from Bakken and colleagues [11] were obtained from the Human Protein Atlas (proteinatlas.org). Human single nucleus RNA sequencing data from Mathys and colleagues [5] and Lake and colleagues [12] were obtained from The Myeloid Landscape 2 portal (http://research-pub.gene.com/BrainMyeloidLandscape) [70]. Analysis of Zhou and colleagues dataset: Data analysis was performed using R (4.1.2) and packages Seurat (4.0.6), dplyr (1.0.7), ggplot2 (3.3.5), pheatmap (1.0.12), and wesanderson (0.3.6). For analysis, we retained all cells included in the labelled metadata accompanying the study, without further quality filter. We performed data normalisation, variable gene identification, data scaling, dimensionality reduction with PCA, and clustering analysis using functions implemented in the package Seurat. Subgroups Ex0 and Ex1 were considered together for excitatory neurons and subgroups Oli0 and Oli1 were considered together for oligodendrocytes. The z-score was calculated from the log₂ (normalised counts + 1) and plotted as a heat map for each dataset. Analysis of other 3 datasets: Data for each gene of interest were obtained as normalised counts averaged for each cell type. The z-score was calculated from the log₂ (normalised counts + 1) and plotted as a heat map for each dataset.

Analysis of proteomics data

Protein expression data of isolated mouse brain cells was generated by Sharma and colleagues [14] using Magnetic-Activated Cell Sorting (MACS) to isolate microglia, astrocytes, oligodendrocytes, and neurons from the brains of young (P8) C57/BL6 mice (3 biological replicates), analysed by mass spectrometry. These data were obtained from the journal website. Data for each protein of interest were obtained as log₂ LFQ intensity averaged for each cell type. The z-score was calculated from the log₂ LFQ intensity and plotted as a heat map.

Imaging and analysis

IHC: For oligodendrocyte analysis, sections were imaged using a Zeiss LSM 800 confocal microscope using a 40× magnification (field of view (FOV): 159.73 × 159.73 μm; 2,101 × 2,101 pixels), 4 to 6 FOV from Layers 5/6 of the cortex were captured across 2 to 3 sections, and cells were quantified manually using the FIJI build of ImageJ (open source) after blinding the experimenter to animal genotype. For APP and BACE analysis, sections were imaged using a Zeiss LSM 980 confocal microscope using a 40× magnification, 3 to 4 FOV from the cortex were captured, and cells were quantified manually using the FIJI build of ImageJ (open source). For plaque analysis, sections were imaged using a Zeiss AxioScan slide scanner, capturing the whole of 3 sagittal sections per animal, with a maximum resolution of 0.325 μm/pixel. Images were first manually segmented to select regions of interest [all visible cortical areas (comprising of visual, retrosplenial, and M2 motor areas), Layers 2/3 of motor cortex (M2), Layers 5/6 of motor cortex (M2), Layers 2/3 of retrosplenial cortex, Layers 5/6 of retrosplenial cortex, corpus callosum (including splenium), area CA1 of hippocampus] and remove regions with postmortem tissue damage or poor imaging quality. The boundary between Layers 2/3 and Layers 5/6 was identified based on the sharp reduction in the density of cell nuclei. Analysis was then automated using the FIJI build of ImageJ software (open source) to segment, count, and measure Aβ⁺ plaques. Myelin sheath lengths were analysed as previously described [71,72]: images were captured using a Leica TCS SPE confocal microscope at 20× magnification with a z-step size of 2 μm. Four images were collected from Layer 2/3 of the primary somatosensory cortex from each of 3 sections per animal. Myelin sheaths were measured using the simple neurite trace plug-in for the FIJI build of ImageJ (open source). Only complete myelin sheaths that started and ended within the 30-μm section were traced and measured, and 120 sheaths were measured per mouse. Imaging and analysis was performed blind to genotype/condition. Oligodendrocyte autophagy was analysed as previously described [73,74]: Images were captured using a Zeiss LSM 980 confocal microscope using a 63× magnification (pixel size 0.132 μm) with a z-step size of 0.5 μm, 4 FOV were captured per animal from Layers 5/6 of the cortex, with 2 to 3 cells analysed per FOV. Cells were segmented based on Olig2 staining, the nuclear mask was dilated to capture the cytoplasm, and the fluorescent intensity normalised to surface was measured. Imaging and analysis was performed blind to genotype/condition.

RNAscope: Sections were imaged using a Zeiss LSM 880 or Zeiss LSM 980 confocal microscope using a 40× magnification (FOV: 212.55 × 212.55 μm; 1,024 × 1,024 pixels). Ten FOV were captured from each brain, 5 corresponding to Layer 5/6 of the cortex from different parts of the tissue section and 5 corresponding to Layer 2/3. The number of cells expressing different genes was quantified manually using the FIJI build of ImageJ software (open source) after blinding of the experimenter to the donor ID and diagnosis. Densitometry was carried out using a custom macro in the FIJI build of ImageJ software to segment cell nuclei (rather than algorithmically inferring cell boundaries, which is liable to introduce bias when comparing cell types which greatly differ in size), classify them as MBP⁺ or RBFOX3⁺, and then measure the number and area of BACE1 and APP spots in each cell.

ICC: Cells were imaged using a Zeiss Cell Discoverer 7 imaging system with an LSM 900 confocal imaging module using a 20× magnification (FOV: 202.45 × 202.45 μm; 1,024 × 1,024 pixels). For each line, a minimum of 3 FOV were captured from each of 2 wells. Cell number was quantified manually using a custom macro in the FIJI build of ImageJ software, while cell positivity was assessed manually.

Statistics

Graphs were plotted and statistical tests were carried out using GraphPad Prism software or Matlab. Statistical tests were chosen for each experiment based on the normality of data and sample matching. Actual statistical tests used for each experiment are stated in the figure legends. All statistical tests used were two-sided unless otherwise stated; n values stated refer to true biological replicates—depending on the experiment an individual n is a human brain (comprising 5 FOV imaged), a cell line (comprising 1 to 4 independent inductions, each of which is the average of 2 to 5 wells), a mouse (comprising 2 to 3 whole sagittal sections imaged or 3 to 12 FOV), or a neuron/unit. Actual n values and what they refer to for each experiment are stated in the figure legends.