Strange IndiaStrange India

Generation of iPS cells

iPS cells were generated from control human early neonatal dermal fibroblasts (Gibco C0045C) with an apparently normal karyotype (checked at Cell Guidance Systems, according to CellGS fixed sample protocol) via episomal reprogramming using plasmids coding for OCT3/4, SOX2, KLF4 and TP53 shRNA61,62. Plasmids (Addgene #27077, #27078 and #27080; 1 µg each) were electroporated into 1 million fibroblasts using 100 µl Neon tips (Invitrogen MPK10025) on the Neon Transfection System (Pulse Voltage: 1,650 V, Width: 10 ms, Number: 3; Invitrogen MPK5000). After electroporation, cells were grown on gelatin-coated plates (EmbryoMax 0.1% Gelatin Solution (Millipore ES-006-B)) in fibroblast medium (DMEM (Sigma D5671); 20% FBS (Gibco 10270-106); 1× GlutaMAX (Gibco 35050-061); 1× NEAA (Gibco 11140-050); 1× Penicillin/Streptomycin (Sigma P4333-100ML)). On day 3 post-electroporation (and then until day 12), 0.5 µg ml−1 of sodium butyrate (Sigma B5887) was added to the culture medium. The fibroblasts were then sub-cultured onto mouse embryonic fibroblasts (MEFs) (Global Stem GSC-6001G or Millipore PMEF-CFX) on the day 5 in iPS cell medium (KnockOut DMEM/F12 (Gibco 12660012); 20% KnockOut Serum Replacement (Gibco 10828028), 1× GlutaMAX (Gibco 35050-061); 1× NEAA (Gibco 11140-050); 80 ng ml−1 bFGF (Gibco PHG0261); 1× EmbryoMax 2-Mercaptoethanol (Millipore ES-007-E)). The first iPS cell colonies were observed around day 11. Positive live alkaline phosphatase staining (Invitrogen A14353) was performed on day 14. iPS cell colonies were individually manually picked on day 14–21, resuspended and re-plated onto fresh MEF 24-well plates. iPS cell clones were then expanded and the ones that displayed best proliferation and morphology (NiPS 8, 9 and 10) were banked at passage 10/11 in iPS cell freezing medium (iPS cell medium + 10% DMSO (Sigma D2650-100ML)). Upon further expansion to P12-15, cells were prepared for karyotype check at Cell Guidance Systems (according to CellGS fixed sample protocol) and based on their apparently normal karyotyping results, NiPS 10 was selected for further experiments.


Similarly to our previous study on human embryonic stem cells6, iPS cell colonies were manually picked and partially dissociated into smaller clumps and transferred into non-adhesive culture dishes and induced to form embryoid bodies in the presence of EB medium (iPS medium without bFGF). After 5–7 days, embryoid bodies were transferred onto poly-l-ornithine (Sigma P4957-50ML; 20 µg ml−1 in sterile water (Gibco 10977035), 1 h at 37 °C followed by 3 washes with PBS (Gibco 10010015) and laminin (Gibco 23017-015); 5 µg ml−1 in PBS, 1 h at 37 °C)-coated dishes and left to adhere in NSC medium (DMEM/F12 (Gibco 21331046); 0.5× B27- supplement (Gibco 12587-010), 0.5× N2 supplement (Gibco 17502-048); 1× GlutaMAX (Gibco 35050-061); 25 ng ml−1 bFGF (Gibco PHG0261)). Formation of neural rosettes was observed within 4–10 days. Neural rosettes were manually dissected and picked under EVOS XL Core Imaging System (LifeTech AMEX1000), and after dissociation re-plated onto fresh poly-l-ornithine and laminin-coated dishes in NSC medium. After 2–5 days, new and smaller rosettes appeared (R1 rosettes) with the presence of heterogeneous contaminating cells. The R1 rosettes were then manually dissected, picked and dissociated into smaller clumps and re-plated onto fresh poly-l-ornithine and laminin-coated dishes. After further 2–5 days, new rosettes (R2) with minimal contaminating cells appeared. R2 rosettes were then routinely monitored to identify small groups of radially organized cells that were present outside of the neural rosettes and represented an independent ‘clone-like population’, the iCoMoNSCs. These small patches of iCoMoNSCs were then manually picked and transferred onto freshly poly-l-ornithine and laminin-coated 24-well plates. Clones that showed clear radial and consistent morphology, good attachment, survival and proliferation upon transfer were enzymatically detached using 0.05% Trypsin (Gibco 15400-054 diluted in PBS), which was blocked by 1× Defined Trypsin Inhibitor (Gibco R-007-100) and expanded for numerous passages and banked in NSC freezing medium (NSC medium + 10% DMSO (Sigma-Aldrich D2650)). Upon further expansion to P9-13, iCoMoNSCs lines were prepared for karyotype check at Cell Guidance Systems (according to CellGS fixed sample protocol). iCoMoNSCs clone 10/80 with apparently normal karyotype was used in the study. iCoMoNSCs are available upon request.

Differentiation of iCoMoNSCs into iNets

iCoMoNSCs were plated in NSC medium at 75,000 cells per cm2 onto Matrigel-coated (Corning 354234; ~0.15 mg ml−1 diluted in cold DMEM/F12 Gibco 11330032 and incubated at least 1 h at 37 °C) 6-well plates and left to recover and proliferate to reach ~95% confluency. At this point, NSC medium was switched to D3 differentiation medium (DMEM/F12 (Gibco 11330032); 0.5× B27+ supplement (Gibco 17504-044), 1× N2 supplement (Gibco 17502-048); 1× GlutaMAX (Gibco 35050-061); 1× Penicillin/Streptomycin (Sigma P4333-100ML)) supplemented with 5 µM Forskolin (Cayman AG-CN2-0089-M050), 1 µM synthetic retinoid Ec23 (Amsbio AMS.SRP002-2), 500 nM Smoothened agonist SAG (Millipore 5666600) for the first 5 days. On the days 6–10, Ec23 was increased to 2 µM. On days 11–25, Ec23 was decreased to 10 ng ml−1, SAG to 50 nM and BDNF (PeproTech 450-02), GDNF (PeproTech 450-10) and CNTF (Alomone labs C-240) were added at 20 ng ml−1. At day 26 and onwards, medium was switched to maturation medium (1:1 DMEM/F12:Neurobasal (Gibco 21103049) mix; 1× B27+ supplement, 1× N2 supplement; 1.5× GlutaMAX, 5 µM forskolin, BDNF, GDNF, CNTF, NT-3 (PeproTech 450-03) and IGF-1 (Stem Cell 78022) all at 20 ng ml−1 and 10 µM cAMP (Sigma-Aldrich D0260)). Medium was changed daily and almost completely at days 0–10, whereas from this point on only two-thirds of the medium was changed 3 times a week.

For all experiments (except scRNA-seq where the experiment was performed in the original 6-well plate), young to middle stage iNets were dissociated into single-cells suspension using Papain Dissociation System (Worthington LK003150), passed through a 70-µm cell strainer (Falcon 07-201-431), resuspended in maturation medium (for imaging experiments, cells were passed through another 40 µm cell strainer (Falcon 352340)) and re-plated into the corresponding vessel at approximately 0.5 million cells per cm2 as counted by CASY Cell Counter (Innovatis AG). 96-well imaging plates used: Greiner Bio-One 655090 or ibidi µ-Plate 96 Well Black 89626.

Preparation of human SNSs

iNets were washed with PBS and homogenized as they were scraped off with the SNS buffer containing 0.35 M sucrose pH 7.4, 10 mM 4-(2 hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES; Biosolve 08042359), 1 mM ethylenediaminetetraacetic acid (EDTA; VWR 0105), 0.25 mM dithiothreitol (DTT; ThermoFisher Scientific R0861), 30 U ml−1 RNAse inhibitor (Life Technologies N8080119) and cOmplete-mini EDTA free protease inhibitor cocktail (Roche 11836170001; at least 2500 µl volume of buffer per one full 6-well plate of iNets). Homogenates were passed sequentially through three 100-µm Nylon net filters (Millipore NY1H02500), followed by one 5 µm filter (Millipore SMWP013000). The final filtrate was resuspended in 3 volumes of SNS buffer without sucrose and centrifuged at 2,000g, for 15 min at 4 °C to yield a pellet containing SNSs. The pellet was resuspended in 100 µl of SNS buffer, which was used for electron microscopy (60 µl) and western blot analysis (40 µl).

Cloning of the all-in-one monocistronic TetON cassette, lentiviral vector preparation and transduction

Using NEBuilder HiFi DNA Assembly Cloning Kit (NEB E5520; HiFi kit), human wild-type, full-length TDP-43 with a C-terminal HA tag from a published pcDNA5 plasmid containing the human TDP-43 cDNA sequence41,63 was inserted into our all-in-one monocistronic TetON (mTRE; build based on Markusic et. al. 37), which was previously inserted into a pLVX lentiviral transfer vector (Clontech 632164), while deleting CMV-PGK-Puro, generating mTRE-TDP-43–HA lentiviral transfer vector. mTRE cassette was build using both NEBuilder and Q5 site directed mutagenesis (NEB E0554S; Q5 kit) kits and consists of the Tet-responsive promoter Ptight, consisting of seven tet operator sequences followed by the minimal CMV promoter (sequence source pCW57.1; Addgene #41393), driving the inducible expression of the downstream TDP-43–HA, followed by downstream IRES2 sequence (sequence source Addgene #60857), which is immediately followed by T7 tag fused to SV40 NLS, which was fused to rtTA-Advanced (sequence source Addgene #41393), which made the rtTA predominantly nuclear, making it readily available for the system, while the T7 tag made rtTA a useful, independent marker of the transgenic cells. See Extended Data Fig. 5a,b for more details. To generate mTRE-GFP-FUS, GFP-FUS sequence from Hock et al. 27 was substituted into mTRE-TDP-43–HA via HiFi Kit.

mTRE-TDP-43–HA and all other lentiviruses were then packaged into lentivirus via Lipofectamine 2000 (Invitrogen 11668019; 8.3 µl per ml final volume) co-transfection with CMV-Gag-Pol (Harvard dR8.91) and pVSV-G (Clontech, part of 631530) plasmids into production HEK293T cells (ATCC CRL-3216; gift from the Greber laboratory) adapted to grow in serum-free conditions (OHN medium; based on Opti-MEM (ThermoFisher 11058-021), supplemented with 0.5% B27- (ThermoFisher 12587-010), 0.5% N2 (ThermoFisher 175020-01), 1% GlutaMAX (ThermoFisher 35050038) and bFGF (25 ng ml−1; ThermoFisher PHG0261)), which reduces the expression of the gene of interest from the transfer vector (that is, it eliminates traces of tetracyclines in the FBS) as well as eliminates serum-carry over into the lentivirus supernatant. Medium was changed the following morning and supernatants were then collected 48 h post transfection (36 h post-medium change), centrifuged (500g, 10 min, 4 °C), filtered through Whatman 0.45-µm CA filter (GE 10462100) and concentrated using Lenti-X Concentrator (Takara 631232) according to the producer instructions (overnight incubation). The resulting lentiviral pellets were then resuspended in complete neuronal maturation medium to achieve 10x concentrated lentivirus preparations, which were titrated using Lenti-X GoStix Plus (Takara 631280). For scRNA-seq and immunofluorescence in Fig. 3b and Extended Data Fig. 5f–i, 10× concentrate of mTRE-TDP-43–HA lentivirus was then used at 1,600 ng (of lentiviral p24 protein as per GoStix Value (GV)) per well of a 6-well plate of differentiated iNets (around 2 months old) along with 3 µg ml−1 of polybrene (Sigma-Aldrich TR-1003-G), pipetting the lentivirus concentrate directly onto the culture (drop-wise). Complete neuronal maturation medium was then added to reach 750 µl total. For all other experiments, all lentiviruses were used at 500 ng (GV) ml−1 of medium in a total of 1,000 µl in sub-cultured 6-well plates, 350 µl in 12 well plates, 80 (Greiner) or 110 µl (ibidi) for imaging 96-well plates. Medium was exchanged completely the following day. For mTRE vectors, TDP-43–HA or GFP–FUS expression was induced by 1 µg ml−1 of doxycycline (DOX; Clontech 631311) when needed.

Cloning of shRNA and constitutive expression vectors

To generate shRNA-expressing vectors, our previously build pSHE lentivirus transfer vector64 was modified using Q5 site directed mutagenesis kit (NEB E0554S; Q5 kit) to substitute the shRNA sequence for either NPTX2 or TARDBP-targeting shRNAs (see Supplementary Table 11 for primer sequences) and using NEBuilder HiFi DNA Assembly Cloning Kit (NEB E5520; HiFi kit) to substitute the EGFP sequence with either TDP-43–HA (resulting in pshTDP vectors (used in NPTX2 rescue experiments) carrying hU6-driven NPTX2 (or NT controls) shRNA and inducible mTRE-TDP-43–HA) or HaloTag (resulting in psHalo vectors (used in TDP-43-knockdown experiments) carrying hU6-driven TARDBP shRNA and inducible mTRE-HaloTag). Similarly, pshTDP vectors for dual luminescence experiments were generated so that they carry hU6-driven endo-TARDBP shRNA (or NT control) and inducible mTRE-TDP-43–HA (with or without mutations in the RNA-recognition motif; shRNA-resistant (exploiting silent mutations present in our TDP-43–HA construct)).

To generate lentivirus transfer vectors for constitutive EF‐1α-driven expression of TDP-43–HA (with or without 3′ UTR), NPTX2–HA and HA–FUS; using HiFi Kit, first the EF‐1α promoter from custom synthetic sequence from GenScript was cloned into mTREAuto-TDP-43–HA vector substituting TRE promoter, generating pLVX-EF‐1α-TDP-43–HA vector. Then either NPTX2 ORF (Origene SC122629) or wild-type FUS27 was cloned into pLVX-EF‐1α-TDP-43–HA vector, substituting TDP-43–HA, generating pLVX-EF‐1α-NPTX2–HA and pLVX-EF‐1α-HA-FUS vectors.

Then the full-length human TARDBP 3′ UTR (cloned out from cDNA in-house generated from HEK293T cells) was cloned using HiFi Kit directly downstream to the TDP-43–HA, generating pLVX-EF‐1α-TDP-43–HA-3′ UTR vector.

SH-SY5Y culture and shRNA testing

SH-SY5Y human neuroblastoma cell line (Sigma 94030304) was cultured in DMEM/F12 (Gibco 11330032) supplemented with 1× GlutaMAX (Gibco 35050-061), 1× Penicillin/Streptomycin (Sigma P4333-100ML) and 15 % heat-inactivated FBS (Gibco 10270-106).

Cells (200,000 per well of a 6-well plate) were plated. Cells were transduced the day after and each lentivirus was mixed with spent medium (500 ng (GV) ml−1, 1 ml total). Three wells were transduced individually for each shRNA-coding lentivirus (2–4 shRNA sequences per target or 1–2 per nontargeting shRNA were tested; see cloning for details). Medium was changed the day after transduction and cells were then collected 4 days post-transduction in RIPA buffer (500 µl per well) containing 2 tablets of cOmplete EDTA free protease inhibitor cocktail (Roche 11873580001) and 1 tablet of PhosSTOP (Roche 0490684500) per 20 ml of buffer.

Dual luminescence assay and 3′ UTR reporter plasmids

The in-house generated full-length human TARDBP 3′ UTR or human NPTX2 3′ UTR (Origene SC213552) were cloned into psiCHECK2-let-7 wild type (Addgene 78260), substituting the let-7 wild-type site (using HiFi Kit) to generate psiCHECK2 3′ UTR reporters used in the dual luminescence assay.

HEK293T cells were plated at 7,500 cells per well in OHN medium (details above) into a white well/bottom Nunc 96-Well plates (Thermo Scientific 136102). The day after, the cells were transduced either with psHalo-shTDP-43 (or NT shRNA) for TDP-43 knockdown or with mTRE-TDP-43–HA (ON or OFF) for TDP-43 overexpression or with pshTDP-sh-endo-TARDBP (or NT shRNA) to knock down endogenous TDP-43 and temporarily overexpress wild-type TDP-43–HA or RRMm TDP-43–HA (see vector details above). Twenty-four hours later, 20 ng per well of psiCHECK 3′ UTR pDNA reporters were transfected using Lipofectamine 2000 (Invitrogen 11668019; 0.5 µl per well final) while simultaneously inducing the mTRE-driven TDP-43–HA variants by 1 µg ml−1 of DOX (Clontech 631311) or leaving them OFF as a control. Medium was exchanged completely the following day adding 78 µl of fresh OHN medium (with or without DOX). Finally, the following day, all conditions were subjected to a lytic Dual-Glo Luciferase Assay System (Promega E2920) according to the manufacturer instructions, luminescence was read using Tecan Infinite M Plex reader and Renilla:firefly luminescence ratios were calculated for each well.

Statistical analysis was performed in Prism (GraphPad San Diego, CA, USA) and unpaired t-test for each lentivirus and 3′ UTR reporter pair was applied on the datasets.

Transmission electron microscopy

SNS pellet was prepared as mentioned above and submitted to the imaging facility (ZMB) of UZH. In brief, SNS pellets were resuspended in 2× fixative (5% glutaraldehyde in 0.2 M cacodylate buffer) and fixed at room temperature for 30 min. Samples were then washed twice with 0.1 M Cacodylate buffer before embedding into 2% Agar Nobile. Post-fixation was performed with 1% Osmium for 1 h on ice, washed three times with ddH2O, dehydrated with 70% ethanol for 20 min, followed by 80% ethanol for 20 min, 100% for 30 min, and finally propylene for 30 min. Propylene:epon araldite at 1:1 was added overnight followed by addition of epon araldite for 1 h at room temperature. Sample was then embedded via 28 h incubation at 60 °C. The resulting block was then cut into 60-nm ultrathin sections using an ultramicrotome. Ribbons of sections were then put onto the TEM grid and imaged on the TEM-FEI CM100 electron microscope.

Fixation, immunofluorescence and imaging of iNets

iNets were fixed with pre-warmed 16% methanol-free formaldehyde (Pierce 28908) pipetted directly into the culture medium, diluted to 4% final formaldehyde concentration, and incubated for 15 min at room temperature. Cells were then washed once with PBS (Gibco 10010015) for 10 min, once with PBS with 0.2% Triton X-100 (Sigma T9284) washing buffer for 10 min and then blocked with 10% normal donkey serum (Sigma-Aldrich S30-M) and 0.2% Triton X-100 in PBS blocking buffer filtered via stericup (Millipore S2GPU02RE) for 30 min at room temperature. Primary antibodies (Supplementary Table 10) were then diluted in blocking buffer and left incubating overnight at 4 °C on an orbital shaker. Cells were then washed 3 × 15 min in washing buffer at room temperature and secondary antibodies (Supplementary Table 10) were then diluted in blocking buffer and incubated for 1.5 h at room temperature. Cells were then again washed 3 × 15 min in washing buffer at room temperature with DAPI (Thermo Scientific 62248) diluted to 1 μg ml−1 in the final washing buffer wash. Cells were finally washed 1 × 15 min in PBS at room temperature, and PBS was then added to the wells to store the stained cells at 4 °C.

Stained iNets were imaged using GE InCell Analyzer 2500 HS wide-field microscope (40× air objective; 2D acquisition; 182 fields of view per well; 50 µm separation to avoid counting cells twice) or high content scanner MD ImageXpress Confocal (40× Water Apo LambdaS LWD objective; 50 µm spinning disk; 2,048 × 2,048 pixels; 30 z-steps per stack and 0.3 µm step size) for quantification or with Leica SP8 Falcon inverted confocal for high-power, high-resolution microscopy (63× oil objective; 2,096 × 2,096 pixels at 0.059 µm per pixel, approximately 20–30 z-steps per stack and 0.3 µm step size). Laser and detector settings were kept the same for each staining combination and all imaged conditions. Huygens professional (Scientific Volume Imaging) was then used to deconvolute the stacks (from SP5 and SP8 confocals) and the deconvoluted images were further post-processed in Fiji65 to produce flattened 2D images (z-projection) for data visualization. Immunofluorescence experiments were repeated twice as independent experiments with biologically independent samples.

For quantification of NeuN- and Ki67-positive cells, images were acquired using Leica SP5 inverted confocal with 63× oil objective (1,024 × 1,024 pixels at 1.7× zoom) and cells were counted and normalized to the total number of cells (total number of DAPI-positive nuclei).

Wide-field image quantification was done using trained ilastik66 algorithms to segment the pixels (positive vs background) of TDP-43–HA, TDP-43p403/404 and DAPI staining. Similarly, ImageXpress acquired images (maximum projections of z-stacks) were segmented for TDP-43–HA, NPTX2, MAP2 and DAPI staining. Segmented pictures (182 images per well, per channel or 36 for ImageXpress) were then exported and the total number of DAPI-positive nuclei (and MAP2+ cells for ImageXpress) was quantified in Fiji65 via batch processing using a custom macro, which is available upon request. TDP-43–HA or TDP-43p403/404-positive neurons were counted manually. Nuclear intensity measurements of TDP-43–HA expressed from TRE or EF‐1α promoters (with or without 3′ UTR) were performed using a custom Cell Profiler (4.2.5) segmentation pipeline on flattened z-stacks acquired on ImageXpress. Statistical analysis was performed in Prism (GraphPad) and unpaired t-tests, one-way or two-way ANOVA followed by Tukey’s multiple-comparison test was applied on the datasets (see figure legends for details).

Biochemical analysis of protein aggregates using SarkoSpin

A modified SarkoSpin protocol14,67 was used. iNets were collected in 200 µl of cell lysis buffer (0.5% sarkosyl (Sigma L5125-100G) + 0.5 µl Benzonase (Millipore E1014-5KU)) and 2 mM MgCl2 in 1× HSI buffer (20 ml of 1× HSI: 10 mM TRIS, 150 mM NaCl, 0.5 mM EDTA, 1 mM DTT + 2 tablets of cOmplete EDTA free protease inhibitor cocktail (Roche 11873580001) and 1 tablet of PhosSTOP (Roche 0490684500)). Resuspended samples were transferred into low-protein-binding Eppendorf tubes. From the obtained total cell lysate, 30 µl was used for total blots, 5 µl was used for protein quantification and the remaining sample was further processed and analysed by SarkoSpin. To 170 µl of the remaining cell homogenate, 178 µl 2× HSI with 4% sarkosyl and 52 µl 1× HSI (to obtain final sarkosyl concentration of 2% in total 400 µl volume) was added. The samples were incubated at 37 °C, 600 rpm for 45 min (Thermomixer, Eppendorf), and after increasing the volume by addition of 200 µl 1× HSI, the samples were vortex mixed and centrifuged at 21,200g on a benchtop centrifuge (Eppendorf) for 30 min at room temperature. Supernatants (~450 µl) were transferred to a new tube and the remaining supernatant was carefully removed to leave the pellets completely dry. The latter were then resuspended in 80 µl of 0.5% sarkosyl 1× HSI and analysed by immunoblot as described below. SarkoSpin was independently performed twice using biologically independent samples.

SDS–PAGE and western blotting

For total cell lysates (both SH-SY5Y and iNets) and lysates of SNS pellets (of iNets), protein concentration was adjusted using Pierce 660 nm Protein Assay Reagent (22660, all reagents from ThermoFisher unless otherwise stated) and 5–10 and 20 µg (total lysates and SNS pellets, respectively) of total protein per lane was used for immunoblots. Samples were resuspended in final 1× LDS loading buffer (NP0007) with 1× final Bolt sample reducing agent (B0009), denatured at 70 °C for 10 min, and loaded on Bolt 12% (SNS) or 4–12% (total and SarkoSpin iNets fractions) Bis-Tris gels (NW04122BOX, NW04127BOX, NW00122BOX, NW00125BOX) using Bolt MES running buffer (B0002) and Bolt Antioxidant (BT0005) and run at 115 V. Gels were transferred onto nitrocellulose membranes using iBlot 2 Transfer NC Stacks (IB23001) with iBlot 2 Dry Blotting System (IB21001) using P0 transfer method. Membranes were then blocked with 5% w/v non-fat skimmed powder milk in 0.05% v/v Tween-20 (Sigma P1379) in PBS (milk PBST) and probed with primary antibodies (Supplementary Table 10) overnight in PBST with 1% w/v milk, washed three times with PBST, followed by incubation with HRP-conjugated secondaries (Supplementary Table 10) in 1% milk PBST. After three washes in PBST, immunoreactivity was visualized by chemiluminescence using SuperSignal West Pico or Femto Chemiluminescent Substrate (PierceNet 34077, 34096) on Amersham Imager 600RGB (GE Healthcare Life Sciences 29083467) or Fusion FX6 EDGE imager. SarkoSpin western blots were independently performed twice using biologically independent samples. All uncropped images of western blots shown in main and extended data figures can be found in Supplementary Figs. 4 and 5.

Two-photon calcium imaging

iNets were bolus loaded with the AM ester form of Oregon Green BAPTA-168 by application of 5 µM solution in HBSS with Ca2+/Mg2+ prepared from 500 µM stock solution69 according to manufacturers’ instructions. Shortly, iNets that had been in differentiation for 12 weeks were briefly washed twice with pre-heated HBSS and then the 5 µM of Oregon Green BAPTA-1 solution was added and incubated for 1 h at 37 °C. Cells were then washed briefly twice with pre-heated HBSS and subsequently allowed to stabilize at room temperature for 30 min before imaging. Two-photon calcium imaging was performed on a Scientifica Hyperscope with a Ti:sapphire laser (Coherent; ~120 fs laser pulses) tuned to 840 nm. Fluorescence images of 256 × 256 pixels at 4.25 Hz were collected with a 16× water-immersion objective lens (Nikon, NA 0.8). Data acquisition was performed by ScanImage.

Calcium imaging data were imported and analysed using custom-written routines in ImageJ and MATLAB, which are available upon request. First, fluorescence image time series for a given region were concatenated. The concatenated imaging data was then aligned using TurboReg to correct for small xy drift (alignment on OGB-1 image series, NIH ImageJ; based on Thévenaz et al. 70). As a next step, the background was subtracted as the bottom first percentile fluorescence signal of the entire image. Average intensity projections of the imaging data were used as reference images to manually annotate regions of interest (ROIs) corresponding to individual neurons. Calcium signals were expressed as the mean pixel value of the relative fluorescence change ΔF/F = (F – F0)/F0 in a given ROI. F0 was calculated as the bottom 10th percentile of the fluorescence trace.


Patch-clamp recordings were made on a SliceScope 2000 microscope (Scientifica) using a HEKA EPC10/2 USB amplifier (HEKA Elektronik). Data were low-pass filtered at 10 kHz and sampled at 200 kHz. Patch pipettes were pulled from borosilicate glass (Science Products) using a P-97 Puller (Sutter Instruments). Pipettes had open-tip resistances of 4–8 MΩ when filled with intracellular solution containing (in mM): potassium gluconate 150, NaCl 10, HEPES 10, magnesium ATP 3, sodium GTP 0.3, BAPTA 0.2. A liquid junction potential of +13 mV was corrected for. The extracellular solution contained (in mM): NaCl 135, HEPES 10, glucose 10, KCl 5, CaCl2 2, MgCl2 1. Series resistance ranged between 7–22 MΩ and was compensated online by 50–80% with 10 µs lag. All recordings were performed at room temperature (23–25 °C). Resting membrane potential was measured in current-clamp after establishing the whole-cell configuration. To examine action potential firing and membrane properties, current injection protocols with incrementing amplitude (duration, 500 ms; step, ±10 pA) were applied from a holding voltage of approximately −80 mV (achieved by tonic current injection). Input resistance was calculated from the steady-state voltage deflection elicited by hyperpolarizing current injections of –10 pA. Membrane time constant was estimated from exponential fits to the voltage decay.

To record voltage-dependent ionic currents, cells were voltage-clamped at a holding potential of –80 mV. Currents were elicited by 100-ms voltage steps from –80 mV to +40 mV in steps of 10 mV and corrected for leak and capacitance currents using the P/5 method. For analysis of currents, cells were included if remaining series resistance was <10 MΩ. Data were analysed with custom-written routines in Igor Pro (Wavemetrics), which are available upon request.


CMOS-based HD-MEAs30 were used to record the extracellular action potentials of iCoMoNSC-derived human neural networks. The HD-MEA featured 26,400 electrodes, organized in a 120 × 220 grid within a total sensing area of 3.85 × 2.10 mm2. The electrode area was 9.3 × 5.45 µm2, and the centre-to-centre electrode distance (pitch) was 17.5 µm, which allowed for recording of cell electrical activity at subcellular resolution. Up to 1,024 electrodes could be simultaneously recorded from in user-selected configurations. The HD-MEA featured noise values of 2.4 µVrms in the action potential band of 0.3–10 kHz and had a programmable gain of up to 78 dB. The sampling frequency was 20 kHz.

HD-MEA recordings

The recording setup was placed inside a 5% CO2 cell culture incubator at 37 °C. Recordings were performed using the ‘Activity scan assay’ and ‘Network assay’ modules, featured in the MaxLab Live software (MaxWell Biosystems), as previously described32. The spontaneous neuronal activity across the whole HD-MEA was recorded using 6,600 electrodes at a pitch of 35 µm in 7 electrode configurations for 120 seconds. The most “active” 1,024 electrodes were then used to record network electrical activity for 300 seconds. Active electrodes were identified based on their firing rate, and, among those, the 1,024 electrodes featuring the highest firing rates were selected.

HD-MEA metrics

We used metrics similar to those described in Ronchi et al. 32 to characterize and compare the neuronal cultures; we used network, single-cell and subcellular-resolution metrics. As network metrics (Extended Data Fig. 3c) we used the burst duration (BD), inter-burst interval (IBI), inter-burst interval coefficient of variation (IBI cv)32. As single-cell metrics (Extended Data Fig. 3d) we used the mean firing rate (MFR), mean spike amplitude (MSA), and the inter-spike interval coefficient of variation (ISI cv)32. Additionally, we included the following extracellular waveform metrics (Extended Data Fig. 3e), extracted from SpikeInterface33, an open-source Python-based framework to enclose all the spike sorting steps:

  1. (1)

    Half width half maximum (HWHM), half width of trough of the action potential wave at half amplitude.

  2. (2)

    Peak-to-trough ratio (PTr), ratio of peak amplitude with respect to amplitude of trough.

  3. (3)

    Peak to valley (PtV), time interval between peak and valley.

  4. (4)

    Repolarization slope (RepS), slope between trough and return to baseline.

  5. (5)

    Recovery slope (RecS), slope after peak towards recovery to baseline.

As subcellular-resolution metrics, we extracted the action potential propagation velocity32 (Vel) and the axon branch length (BL).

The percentage of active electrodes (AE) was also computed to measure the overall number of electrodes that could detect action potentials32.

HD-MEA data analysis

Data analysis was performed using custom-written codes in MATLAB R2021a and Python 3.6.10, which are available upon request.

Spike sorting was performed to identify single units in the extracellular recordings. We used the Kilosort271 software within the SpikeInterface33 framework and the corresponding default parameters. We automatically curated the spike sorting output using the following parameters:

  1. (1)

    Inter-spike interval violation threshold (ISIt) = 0.5. The ISIt takes into account the refractory period, which follows every action potential. The assumption is that if two action potentials occur within a too short time interval, they most probably come from two different neurons.

  2. (2)

    Firing rate threshold (FRt) = 0.05. The FRt sets the minimum firing rate of a neuron to be considered as a ‘good’ unit.

  3. (3)

    Signal-to-noise ratio threshold (SNRt) = 5. The SNRt takes into account the ratio between the maximum amplitude of the mean action potential waveform and the noise characteristics of the specific channel.

  4. (4)

    Amplitude cutoff (ACt) = 0.1. The ACt takes into account the false-negative rate—that is, the fraction of spikes per unit with an amplitude below the detection threshold.

  5. (5)

    Nearest-neighbours hit rate (NNt) = 0.9. After computing the principal component for a unit, the NNt is used to check on the fraction of the nearest neighbours that fall into the same cluster.

HD-MEA statistical analysis

Statistical comparisons to compare samples of more than two populations were performed using the Kruskal–Wallis H test. In case the null hypothesis was rejected, we performed a post-hoc Dunn test with Sidák correction for multiple comparisons (Dunn–Sidák multiple-comparison test).

Statistical analysis was performed in MATLAB R2021a.

HD-MEA plating

HD-MEAs were sterilized for 40 min in 70% ethanol. HD-MEAs were washed three times with sterile deionized water (dH2O) and then dried. After sterilization, HD-MEAs were soaked in 20 µl of poly-d-lysine solution (P6407, Sigma-Aldrich) (diluted to 50 mg ml−1 in dH2O) for 1 h at room temperature to render the surface more hydrophilic; each HD-MEA was then washed three times with sterile dH2O and dried. Thereafter, HD-MEAs were coated with 10 µl Matrigel (354234, Corning), previously diluted in a plating medium (see below) at a 1:10 ratio and incubated at 37 °C for 2 h. iNets dissociated at 1.5, 3 and 7.5 months of age were centrifuged at 188g for 5 min, supernatant was aspirated, and plating medium was added to the cell pellet to reach the desired cell density (200,000 cells per HD-MEA). Matrigel was aspirated from HD-MEAs and cells were plated in a 10 µl drop on the HD-MEAs. Medium (0.9 ml) was added after 2 h of incubation at 37 °C. 50% of plating medium was exchanged one day post plating and subsequently twice a week. Plating and culture medium was almost identical to maturation medium except that it used BrainPhys (05790, Stem Cell Technologies) as a base medium and was further supplemented with 1% penicillin/streptomycin (P4333-100ML, Sigma-Aldrich) to provide protection to cells that had to be periodically moved from culture incubator to recording culture incubator.

Primary mouse hippocampal neuronal cell culture

Primary neuronal cell cultures were prepared from mouse embryos (E16/17). All animal experimentation, including mouse housing and breeding was done in accordance with the Swiss Animal Welfare Law and in compliance with the regulations of the Cantonal Veterinary Office, Zurich (current approved license ZH169/2022, valid until 16.02.2026). Pregnant C57BL/6 females were delivered from Janvier Labs (France) at day 12 of pregnancy and housed in accordance with the Swiss Animal Welfare Law and in compliance with the regulations of the Cantonal Veterinary Office of Zurich at LASC Irchel, Zurich. In detail, mice were housed in IVC type 2 long (T2L) cages with an enriched environment that consisted of bedding, a red mouse house, tissues and crinklets (sizzled paper). The room temperature was between 21 °C and 24 °C with a humidity level between 35% and 70%. A 12 h light/12 h dark cycle was used. Pregnant females were euthanized in their home cage with CO2 prior to extracting embryos. CO2 euthanasia was chosen over other methods because it is considered rapid, highly successful, safe and easy for the operator and, in our case, compatible with downstream experimental procedures (no physical damage is inflicted to any organs of interest). The fetuses were then extracted and euthanised by decapitation using scissors, which is considered the most rapid and effective method for fetuses. Five embryos per litter were used. Hippocampi were isolated on ice from each embryo’s brain, pooled together and digested with trypsin ((Gibco 15400-054) 0,5% w/v supplemented with 4% w/v d-glucose (Sigma G8769)) for 15 min at 37° degrees. Following digestion, trypsin was quenched with horse serum (Sigma H1138) and the tissue was manually dissociated until clumps were not visible anymore. Cells were spun at 120g for 5 min and plated on poly-d-lysine (Sigma P6407) coated glass-bottom 8-well-IBIDI chamber slide (80807; 4 wells per experimental condition) in Neurobasal medium (Gibco 21103049) supplemented with GlutaMAX (ThermoFisher 35050061), 2% B27+ supplement (Gibco 17504-044), 100 U penicillin-streptomycin (Sigma P4333-100ML) and 4% w/v d-glucose (Sigma G8769). To transduce the primary neuronal cultures, TDP-43–HA or GFP–FUS mTRE lentivirus vectors were used at 500 ng (GV) per ml of spent culture medium in a total 100 µl at DIV 11 and the medium was fully exchanged the following day (DIV 12), supplemented with 1 µg ml−1 of DOX (Clontech 631311), thereby inducing the expression of transgenes. Cells were fixed at DIV 17.

Mycoplasma check

All cycling cells used in the study were routinely checked for possible mycoplasma contamination using EZ PCR Mycoplasma Detection Kit (Sartorius 20-700-20). Young iNets (with some residual KI67+ cells) were sporadically checked. No mycoplasma contamination was detected.

Patient post-mortem brain immunofluorescence

Formalin-fixed, paraffin-embedded hippocampal, frontal or primary motor cortex patient (ALS, FTLD-TDP-A, FTLD-TDP-C, FTLD-FUS, FTLD-tau, Alzheimer’s disease) sections (see Supplementary Table 9) were used. All FTLD and Alzheimer’s disease tissue samples were donated to Queen Square Brain Bank for Neurological Disorders at UCL Queen Square Institute of Neurology with full, informed consent. Anonymized autopsy ALS sample was collected by the Institute of Neuropathology at UZH. According to Swiss law, anonymized autopsy tissues do not fall within the scope of the Human Research Act and may be used in research. Sections were deparaffinized in three Xylene rounds (5 min each) and rehydrated in decreasing ethanol washes (2× 100% for 10 min; 2× in 95% for 5 min; 80, 70 and 50% for 5 min each) and finally submerged in MilliQ water for 10 min. Antigen retrieval was then performed by microwave heat treatment in sodium citrate buffer (0.01 M, pH 6.0). Sections were then cooled down on ice for 10 min and once quickly washed in PBS, followed by 3x PBS washes for 5 min at room temperature before blocking with blocking buffer (5% normal donkey serum (Sigma-Aldrich S30-M), 3% BSA (Sigma A4503) and 0.25% Triton X-100 (Sigma T9284) in PBS) for 30 min at room temperature. Primary antibodies (Supplementary Table 10) were then diluted in blocking buffer and 300 µl of the antibody mix was evenly put per slide/section. Slides were put into a wet and dark incubation chamber and incubated overnight at room temperature. Slides were rinsed once in PBS and then washed 3× 5 min in PBS on a shaker. Secondary antibodies (Supplementary Table 10) diluted in blocking buffer, centrifuged for 30 min at 15,000g at room temperature and 500 µl of the antibody mix was evenly put per slide/section and put into the incubation chamber to incubate for 2.5 h at room temperature. Slides were then rinsed once in PBS, washed 1× 5 min in PBS on a shaker and to stain the nuclei (DNA), 500 µl of DAPI solution (Thermo Scientific 62248; diluted to 1 μg ml−1 in PBS) was evenly added onto the slide/section and incubated for 10 min at room temperature in the incubation chamber. Slides were rinsed once in PBS, washed 2×5 min in PBS on a shaker, followed by Sudan Black (0.2% in 70% ethanol; Sigma 199664) autofluorescence quench for 10 min at room temperature on shaker. Slides were rinsed 6 times in PBS and left washing in PBS until mounted and coverslipped in ProLong Diamond Antifade Mountant (ThermoFisher P36961) and then left to dry in the chemical cabinet at room temperature in the dark. Mounted sections were stored at 4 °C in the dark.

Stained patient brain sections were imaged using Leica SP8 Falcon inverted confocal for high-power, high-resolution microscopy (63× oil objective; 1.7 or 3× zoom; 2,096 × 2,096 pixels at 0.059 µm per pixel or 1,848 × 1,848 pixels at 0.033 µm per pixel, respectively, approximately 20 z-steps per stack at 0.3 µm). White laser and HyD detector settings were kept the same for each staining combination and all imaged conditions. Huygens professional (Scientific Volume Imaging) was used to deconvolute the stacks and the deconvoluted images were further post-processed in Fiji65 to produce a 3D projection for data visualization (the first image of the 3D projection is shown in figures). Post-processing settings were kept the same for all images of all sections from all donors (except for DAPI and MAP2 channels for certain sections when higher brightness settings were used to reach intensity that allowed proper visualization).

Cell isolation for scRNA-seq

Duplicates (except for TDP-43–HA OFF and TDP-43–HA 2 weeks samples) of iNets (young (1.5 months), middle stage (3 months) and old (7.5 months)) and TDP-43–HA experiment samples at middle stage, were dissociated into single-cells suspension using Papain Dissociation System (Worthington LK003150), passed through 70-µm and 40-µm cell strainers (Falcon 07-201-431 and 07-201-430), and resuspended in HIB++ medium (Hibernate-E Medium medium (Gibco A1247601) supplemented with EDTA (1 mM final; Invitrogen AM9260G), HEPES (10 mM final; Gibco 15630080), with 1× B27+ supplement (Gibco 17504-044), 1× N2 supplement (Gibco 17502-048); 1× GlutaMAX (Gibco 35050-061), BDNF (PeproTech 450-02), GDNF (Alomone labs G-240), CNTF (Alomone labs C-240), NT-3 (PeproTech 450-03) and IGF-1 (Stem Cell 78022) all at 20 ng ml−1) to 1,000 cells per µl using CASY Cell Counter (Innovatis).

scRNA-seq using 10X Genomics

The quality and concentration of the single-cell preparations were evaluated using an haemocytometer in a Leica DM IL LED microscope and adjusted to 1,000 cells per µl. Ten-thousand cells per sample were loaded in to the 10X Chromium controller and library preparation was performed according to the manufacturer’s indications (Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 protocol). The resulting libraries were sequenced in an Illumina NovaSeq sequencer according to 10X Genomics recommendations (paired-end reads, R1 = 26, i7 = 10, R2 = 90) to a depth of around 50,000 reads per cell. The sequencing was performed at Functional Genomics Center Zurich (FGCZ).

Single-cell data analysis

Data was processed with CellRanger for demultiplexing, read alignment to the human reference genome (GRCh38) and filtering to generate a feature-barcode matrix per sample. Cell doublets were removed with scDblFinder72 and outlier cells were detected and filtered with the scater R package73. In short, cells with more than 3 median-absolute-deviations away from the median number of UMIs, the number of features and the percentage of mitochondrial genes were removed. For the 7.5-month-old samples, we additionally filtered cells with fewer than 2,000 UMIs and fewer than 1,500 detected features. For the TDP-43 overexpression experiment, we additionally filtered cells with fewer than 5,000 UMIs and fewer than 2,500 detected features.

Seurat v374 was used for log-normalization and to identify the top 2,000 highly variable genes per sample. Louvain clustering was always performed with resolution 0.4 based on a shared nearest-neighbour graph constructed from the top 20 principal components. The UMAP75 cell embeddings were computed from the top 20 principal components. Cell cycle scores were estimated with Seurat using a list of G2/M and S phase markers from Kowalczyk et al. 76.

Marker genes that are upregulated in one cluster compared to any other cluster were identified with the findMarkers function from the scran R package77, which runs pairwise t-tests and combines the results into a ranked list of markers for each cluster. Heat maps with marker gene expression show scaled mean log counts of all cells in each cluster. Cluster 12 (TDP-43–HA overexpression experiment) was profiled with the same method against other neuronal clusters (0, 2 to 5, 7 to 11). We note that clustering scRNA-seq data followed by looking for differences between clusters uses the same data twice (that is, ‘double dipping’)78. We provide a ranked list of differential expression statistics to help cluster interpretation, but we are aware that P values are difficult to interpret in this setting.

The two iCoMoNSC samples were integrated with Seurat using canonical-correlation analysis79, the data was scaled and number of UMIs and the percentage of mitochondrial UMIs was regressed out before clustering.

The scRNA-seq data from three different NSC lines7 were downloaded from ArrayExpress (accession number E-MTAB-8379) and integrated with our iCoMoNSC data using Seurat; the data were scaled and the number of UMIs was integrated out before clustering.

The human organoid (409b2 and H9) scRNA-seq data8 was downloaded from ArrayExpress (accession number E-MTAB-7552). Data from the iCoMoNSC and iNets (1.5, 3 and 7.5 months old) were integrated with data from cells from human brain organoids using Seurat, and scaled and the number of UMIs and the percentage of mitochondrial UMIs were regressed out before clustering. Batch effects were assessed with the smoothed cms mixing metric from CellMixS80 using k = 500 neighbours and 10 dimensions in principal component analysis space.

In the four samples of the TDP-43 overexpression experiment, we quantified the expression of the TDP-43–HA construct components, including TDP-43–HA, the long terminal repeats, and the rtTA, as well as the endogenous TARDBP using CellRanger. The counts were added to the filtered data from the whole transcriptome CellRanger output. The total TDP-43 log2 fold change between cluster 12 and all other neuronal clusters was computed using the summary.logFC metric from scran’s findMarkers.

Pseudotime analysis was carried out with monocle3 v1.0.081, commit 004c09682, rooting the trajectories in our annotated NSC cluster.

To assess the replicability of independent iNets, we reused the TDP-43 overexpression experiment cells and their clustering and cluster annotations (TDP-43–HA OFF only) and similarly the cells from the time series (middle stage), using their canonical genes (that is, excluding TDP-43–HA). We run the same downstream analysis as described above (top 2,000 highly variable genes, 20 principal components, UMAP cell embedding), without integrating the two datasets. Independently, we ran MetaNeighbor v1.8.083 to measure cell-type replication across experiments using the annotated clusters (with at least 10 cells) from the aforementioned joint TDP-43–HA OFF and time series middle stage cells. We report the MetaNeighbor area under the receiver operating characteristic (AUROC) values pairwise comparing clusters. These AUROC scores can be read as a probability. An AUROC of 1 depicts a perfect cluster:cluster match (so both clusters are composed by cells from the same cell type), an AUROC of 0.5 is as good as random (unlikely the two clusters belong to the same cell type), and an AUROC of 0 can be read as a strong non-match (so both clusters are clearly not the same cell type).

Bulk RNA-seq

Middle stage iNets in 12-well plates were transduced with shRNA to either knock down TDP-43 (or control shRNA) or overexpress TDP-43–HA (or non-induced control; see ‘Lentivirus preparation’) in 4 replicates per condition (individually transduced wells). Two weeks later, RNA was extracted and isolated using RNeasy Plus Mini Kit (Qiagen 74134) according to manufacturer’s instructions and an additional DNA digestion step was performed by RNase-Free DNase Set (Qiagen 79254). RNA QC was performed by RNA ScreenTape, revealing that the samples had a RINe value in a range of 9.1–9.8. Novaseq 6000 (Illumina) was used for cluster generation and sequencing according to standard protocol. Sequencing configuration was paired-end 150 bp, 150 million reads depth.

Raw reads were processed with ARMOR84. In brief, reads were aligned and counted to the human genome (GRCh38 assembly and Gencode release 43) with salmon v1.4.0 and with STAR 2.7.7a. We modelled the salmon-generated count data with quasi-likelihood (QL) negative binomial generalized log-linear models and ran differential expression analysis with edgeR v3.36.0.


We re-analysed CLiP data from GSE27201 Polymenidou 201111 (mouse) including accessions SRR107031 and SRR107032; from E-MTAB-530 Tollervey 201113 (human) accessions ERR039847, ERR039848, ERR039843, ERR039842, ERR039844, ERR039845, ERR039846 with the nf-core/clipseq workflow v1.0.085 with default parameters and against the GRCh38 or GRCm38 reference genomes, as appropriate. Polymenidou 201111 CLiP binding sites were transformed from GRCm38 to GRCh38 coordinates using liftOver from Kent utils v39086.

Re-analysis of CLiP and iCLIP data

FASTQ files of TDP-43 human brain iCLIP13 (three controls and three patients with FTLD) were downloaded from ArrayExpress (E-MTAB-530)87. The data were re-analysed using the GRCh38 reference genome with the nf-core/clipseq85 for pre-processing, mapping and crosslink site identification. Similarly, FASTQ files of TDP-43 CLiP in mouse were retrieved from GSE27201 and processed using nf-core/clipseq and the GRCm38 genome. Source code is available at

Re-analysis of RNA-seq data from sorted nuclei

We downloaded the RNA-seq gene count table as a supplementary data file (GSE126542_NeuronalNuclei_RNAseq_counts.txt.gz) from GEO accession GSE126542. A differential gene expression analysis between TDP-43 negative and positive nuclei was performed in R version 4.0.5 with edgeR88 taking into account the paired nature of the data using the quasi-likelihood framework89 (empirical Bayes quasi-likelihood F-test). Reported false discovery rate (FDR) values are adjusted for multiple comparisons using the Benjamini–Hochberg method. Source code is available at

Splicing analysis to identify skipped exons

The human reference genome (GRCh38) and annotation (version 108) were obtained from ENSEMBL90. Illumina Trueseq adapters were removed from RNA-seq reads using cutadapt (version 4.1) with the parameters -q 25 -m 2591. The processed reads were mapped to the reference genome using STAR aligner (version 2.7.10b) with default parameters92. Subsequently, differential splicing analysis at the event level was performed using rMATS (version 4.1.2) with the parameters -t paired –readLength 100 –variable-read-length93. Significant differentially spliced events were defined by FDR < 0.05 and 15% change in absolute value of IncLevelDifference and intersected with a human skipped exon dataset43 to check for potential skipped exons in our datasets. The results were processed using tidyverse package (v2.0.0) in R94.

Cell line identification

Cell lines (SH-SY5Y Sigma 94030304 and HEK293T ATCC CRL-3216) used in the study were checked against cross-contaminated or misidentified cell lines ( No cross-contaminated or misidentified cell lines were identified. Other cells used in this study were control human early neonatal dermal fibroblasts (Gibco C0045C)—the source cells used to generate iPS cells, iCoMoNSCs and iNets in this paper; and primary mouse hippocampal neurons.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Source link


Leave a Reply

Your email address will not be published. Required fields are marked *