Cerebral organoid proteomics reveals signatures of dysregulated cortical development associated with human trisomy 21

Human trisomy 21 (Down syndrome) is the most common genetic cause of intellectual disability, and is associated with complex perturbations in protein expression during development. Brain region-specific alterations in neuronal density and composition originate prenatally in trisomy 21 individuals, and are presumed to underlie the intellectual disability and early onset neurodegeneration that characterizes Down syndrome. However, the mechanisms by which chromosome 21 aneuploidy drives alterations in the central nervous system are not well understood, particularly in brain regions that are uniquely human and thus inaccessible to established animal models. Cerebral organoids are pluripotent stem cell derived models of prenatal brain development that have been used to deepen our understanding of the atypical processes associated with human neurobiological disorders, and thus provide a promising avenue to explore the molecular basis for neurodevelopmental alterations in trisomy 21. Here, we employ high-resolution label-free mass spectrometry to map proteomic changes over the course of trisomy 21 cerebral organoid development, and evaluate the proteomic alterations in response to treatment with harmine, a small molecule inhibitor of the chromosome 21 encoded protein kinase DYRK1A. Our results reveal trisomy 21 specific dysregulation of networks associated with neurogenesis, axon guidance and extracellular matrix remodeling. We find significant overlap of these networks show significant overlap with previously identified dysregulated gene expression modules identified in trisomy 21 fetal brain tissue. We show that harmine leads to partial normalization of key regulators of cortical development, including WNT7A and the transcription factors TBR1, BCL11A, and POU3F2, pointing to a causative role for DYRK1A over-expression in neurodevelopmental effects of human trisomy 21.


SUMMARY
Human trisomy 21 (Down syndrome) is the most common genetic cause of intellectual disability, and is associated with complex perturbations in protein expression during development. Brain region-specific alterations in neuronal density and composition originate prenatally in trisomy 21 individuals, and are presumed to underlie the intellectual disability and early onset neurodegeneration that characterizes Down syndrome. However, the mechanisms by which chromosome 21 aneuploidy drives alterations in the central nervous system are not well understood, particularly in brain regions that are uniquely human and thus inaccessible to established animal models. Cerebral organoids are pluripotent stem cell derived models of prenatal brain development that have been used to deepen our understanding of the atypical processes associated with human neurobiological disorders, and thus provide a promising avenue to explore the molecular basis for neurodevelopmental alterations in trisomy 21. Here, we employ high-resolution label-free mass spectrometry to map proteomic changes over the course of trisomy 21 cerebral organoid development, and evaluate the proteomic alterations in response to treatment with harmine, a small molecule inhibitor of the chromosome 21 encoded protein kinase DYRK1A. Our results reveal trisomy 21 specific dysregulation of networks associated with neurogenesis, axon guidance and extracellular matrix remodeling. We find significant overlap of these networks with previously identified dysregulated gene expression modules identified in trisomy 21 fetal brain tissue. We show that harmine leads to partial normalization of key regulators of cortical development, including WNT7A and the transcription factors TBR1, BCL11A, and POU3F2, pointing to a causative role for DYRK1A over-expression in neurodevelopmental effects of human trisomy 21.

INTRODUCTION
Trisomy of chromosome 21 (HSA21), the cause of Down syndrome (DS), is the most common viable human aneuploidy with an estimated prevalence of ~8 in 10,000 live births (1).
DS is characterized by a complex spectrum of clinical manifestations that includes intellectual disability, cardiovascular and craniofacial abnormalities, early onset dementia with Alzheimer's neuropathology (2), increased risk of childhood leukemias, and reduced risk of solid tumors (3).
Among the most consistent neuroanatomical phenotypes in DS is a reduction in the surface area of specific regions of the cerebral cortex (4). These alterations have been attributed to disruption in neurogenesis and synaptogenesis programs during fetal brain development (5)(6)(7)(8)(9).
Hypocellularity in fetal (6, 10) and adult DS neocortex (11), and defective lamination in the developing neocortex (12) suggest that atypical neuroanatomical features in adult DS brain originate early in gestation during the critical phase of cortical expansion.
The precise mechanisms by which specific HSA21 genes contribute to the neuroanatomical and neurobehavioral phenotypes in DS remain unclear. Aneuploidies cause stoichiometric imbalances in macromolecular complexes, signaling pathways, and gene regulatory networks, and can lead to proteotoxic stress (13). For a subset of genes encoded on an aneuploid chromosome, protein expression levels can be buffered to diploid levels by various post-transcriptional mechanisms (14, 15). For example, in trisomy 21 cells, HSA21 encoded proteins are more likely to be expressed at diploid levels if they participate in macromolecular complexes (16). However, a proportion of genes will not be buffered, resulting in protein levels that increase in proportion to the dosage increase associated with the aneuploidy (15).
The HSA21-encoded protein kinase DYRK1A is essential for normal brain development from flies to humans (17). Loss of function mutations in the Drosophila ortholog minibrain impair post-embryonic neurogenesis, causing a reduction in brain size (18). In humans, haploinsufficiency of DYRK1A underlies the neurodevelopmental disorder MRD7 (OMIM 614104), which is associated with intellectual disability, microcephaly, and autism (19-26). In DS brain tissue, DYRK1A protein levels are expressed at ~1.5-fold higher levels in proportion to HSA21 copy number increase (27). Increased DYRK1A expression has been implicated in disrupting aspects of neurodevelopment and neurodegeneration in DS (28-31). Multiple lines of evidence from mouse models support a causative role for DYRK1A dosage alterations in atypical brain development and intellectual disability (31-36), and point to DYRK1A kinase activity as a potential therapeutic target in DS-associated neuropathologies (28, 37-39).
Despite some success in using mouse models of trisomy 21 to interrogate potential interventions in Down syndrome, anatomical and developmental differences between rodents and humans limit the utility of the mouse in the study of human neurological disorders (40). For example, the rodent central nervous system develops along fundamentally different trajectories involving neural stem cell induction mechanisms distinct from those found in hominids (41, 42).
Mouse and rat brains are lissencephalic (smooth), whereas human brains are gyrencephalic (convoluted), with a high surface area thought to originate from proliferative expansion of specific neural progenitor cell subtypes (43, 44) that are poorly represented in rodents (45). The recent failure of a phase II clinical trial to meet primary and secondary endpoints for RG1662, a GABAA inverse agonist (46) that showed promise in a DS mouse model (47, 48), highlights the urgent need for new model systems that more faithfully recapitulate DS neurobiology.
Human stem cell derived cerebral organoids (COs) provide a tractable alternative model system that is amenable to pharmacological and genetic manipulations not possible in humans (49,50). COs develop into three dimensional structures that mimic important features of the in 6 vivo cytoarchitecture found in developing human neocortex (51). Many CO culture protocols begin with the formation of embryoid body aggregates (EBs) that develop regions of radially expanding neuroepithelium upon neural induction, a process that mimics aspects of neurogenesis during human cortical development (52). Many neurodevelopmental disorders arise during these nascent stages of corticogenesis, making COs a useful platform to model complex and uniquely human neurobiological processes and disorders, and to evaluate candidate therapeutics for neuropathologies previously considered untreatable (53).
Large-scale transcriptome and epigenome profiling studies reveal that COs recapitulate many of the key epigenetic and transcriptional programs found in developing human forebrain (54, 55). Post-transcriptional regulatory mechanisms play crucial roles in both neocortex development (56-58) and dosage compensation in DS cells (16). Consequently, trisomy 21 is predicted to have complex effects on proteome remodeling in the developing brain not revealed by steady state mRNA measurements. While comprehensive transcriptome maps of human DS brain tissue in neurotypical and DS individuals have been produced at unprecedented spatiotemporal detail (55, 59, 60), similar efforts to systematically map proteome alterations in DS brain tissue and CO model systems are lacking.
To identify potential mechanisms underlying altered cortical development in DS, we performed ultra-deep, label-free quantitative proteomics at four developmental stages during CO development and differentiation, using human euploid and trisomy 21 induced pluripotent stem cell (iPSC) lines. We also evaluated the proteome-wide effects of the DYRK1A inhibitor, harmine, on the trisomy 21 differentiation trajectory (61). We found that growth in 300 nM harmine led to widespread alterations of the trisomy 21 CO proteome, and partial normalization of functional gene signatures on the protein level for key processes in human cortical 7 development. Integrative analysis of these proteomic changes with a previously described human DS brain mRNA developmental trajectory revealed functional signatures of trisomy 21 dysregulated neuronal differentiation that were partially normalized by harmine treatment.

Cell Culture and Maintenance
The iPSC line I50-2 (euploid female), referred to in this paper as D21, was previously described (62), and exhibited a normal karyotype (supplemental Fig. 1A). The iPSC line C2-4 8 (trisomy of HSA21, female) (63), referred to in this paper as T21, was a generous gift from Gretchen Stein and Chris Link (University of Colorado, Boulder). The C2-4 line exhibited trisomy 21 in karyotype analysis (supplemental Fig. 1B). Genetic stability of the iPSCs was assessed by G-banding karyotype analysis performed at the University of Colorado Molecular Pathology Shared Resource Cytogenetics Core Facility. Metaphase spreads were captured using CCD camera and karyotyping was carried out using the BandView software (Applied Spectral Imaging Inc).
All cell lines were maintained under feeder-free conditions (E8 media with supplements), grown on Geltrex-coated polystyrene 6-well tissue culture plates. For routine passaging, media was aspirated from wells and cells were rinsed briefly with room-temperature, protease-free 0.5 mM EDTA in Dulbecco's phosphate-buffered saline (DPBS) and then incubated with proteasefree 0.5 mM EDTA in DPBS for 5 min at 37 o C. Cells were detached by pipetting media up and down 3-5 times with a 5 mL pipette before being rinsed off the plate and diluted into an equivalent volume of E8 media. Cell suspensions were centrifuged at 400 x g for 2 min, and the cell pellets were resuspended in room temperature E8 media prior to plating. Cells were typically passaged 1:3 when they reached approximately 80% confluence. Cells were treated with 300 nM harmine in 0.003% DMSO, or 0.003% DMSO as vehicle control. To generate the cerebral organoids, we employed the method detailed by Lancaster and Knoblich (52) with some modifications as detailed below. Samples for proteomic analysis were collected as follows.
Samples of iPSC cultures were collected after 3 days of culture, prior to EB induction. We initiated EB formation at day in vitro (DIV) 0, and collected EBs at DIV5 prior to neural induction (Fig. 1A). We refer to embryoid bodies that were grown in neural induction media for 5 days as "neurospheres", which were collected at DIV10. Neurospheres (NS) were embedded in 9 Matrigel at DIV10 for 15 days, followed by 21 days in rotating suspension culture. The COs were collected at DIV45, pooling approximately 10 organoids per replicate.
Embryoid bodies were seeded as follows, identical to a previously described protocol (64). iPSC colonies were collected as for routine passaging, followed by trituration using a firepolished Pasteur pipette in complete E8 media supplemented with 50 μM Y-27632 and 4 ng/mL bFGF to obtain a homogenous cell suspension. Cells were counted in a hemacytometer and were seeded at a density of 9,000 cells/well in a volume of at least 100 μl. Cells were fed every other day for 5 days, and monitored visually for successful induction into embryoid bodies and neurectodermal differentiation, as indicated by clearing surface tissue. On the 6 th day, embryoid bodies were transferred en masse into neurosphere induction media (DMEM/F12 supplemented with 1x MEM-NEAA, 1x N2 supplement, 1 μg/mL heparin sulfate). The cells were fed every other day for an additional 5-7 days, until neuroectoderm differentiation was apparent.
Neurospheres were embedded in Matrigel by putting 10 μl droplets of cold (~4 o C) Matrigel onto a 10 cm tissue culture dish on a cold surface. Single neurospheres were collected in 2-3 μl of media and pipetted directly into the cool Matrigel droplet. The plate containing rows of embedded neurospheres was placed into the 37ºC cell culture incubator for 10 min to allow the Matrigel to solidify before being covered with 10 mL of primary CO media (1:1 DMEM/F12:Neurobasal, supplemented with 1x B27 [minus Vitamin A -Gibco 12587-010], 1x N2, 1x MEM-NEAA, 3.5 μl/L β-mercaptoethanol, 1x antibiotic/antimycotic). COs were monitored during stationary culture for 10-14 days. COs that exhibited successful neural induction, as indicated by the appearance of multiple neuroectodermal lobes, were transferred to rolling bottle cultures containing mature organoid media (using B27 supplement containing vitamin A -Gibco 17504-044). Media was replaced every 3-5 days until harvest.
Immunocytochemistry iPSC growing on Geltrex-coated 1cm diameter round coverslips had media aspirated and were gently washed 1x with 2 mL warm (37 o C) DPBS, then fixed with 4% MeOH-free formaldehyde in PBS for 30 min at room temperature. Fixed cells were washed 2x with 1 mL PBS, permeabilized with the addition of 1 mL PBS+0.1% TX100 and gently rocked for 15 min.
For cells labeled with EdU (1 μM), fluorescent conjugation was performed by incubating permeablized cells with AlexaFluor-488-azide (10 μM) in the presence of 4 mM CuSO4 and 100 mM ascorbic acid for 30 min at room temperature with gentle agitation. Samples were blocked with 3% IgG-free BSA in PBS+0.1% TX100 for a minimum of 30 min prior to the addition of primary antibodies. All primary antibodies were added at a concentration of 1 μg/mL in fresh 3% BSA/PBS solution and incubated overnight at 4 o C. The following morning, primary antibody solution was removed, the samples washed 2x with 1 mL PBS+0.1% TX100, then re-blocked with 3% IgG-free BSA and 1% normal goat serum in PBS+0.1% TX100 for a minimum of 30 min at room temperature. Secondary antibodies were added to the samples at a 1:250 dilution in fresh 3% BSA/1% NGS/PBS solution and incubated for 2 hr at room temp with gentle rocking.
Secondary antibody solution was removed and cells were washed twice with 1 mL PBS+0.1% TX100 for 15 min; Hoechst 33342 was added to a final concentration of 10 μg/mL during the first washing step to label cell nuclei. Following the final wash, coverslips were mounted onto glass slides with ~50 μl of Fluoromount G and allowed to dry in the dark overnight before imaging.

Confocal Microscopy
one microgram of trypsin (Promega), rocking overnight at room temperature. Tryptic peptides were removed from the ultrafiltration device by centrifugation and stored at -80 o C.

Tryptic peptide fractionation
To reduce peptide sample complexity, 30-50 µg of tryptic peptides were fractionated on a high pH reversed-phase C18 column using a Waters M-class Acquity UPLC. Mobile phases used were (A) 10 mM ammonium formate, pH 10 in LCMS water and (B) 10 mM ammonium formate, pH 10 in 80% (v/v) LCMS acetonitrile. Samples were loaded and washed to baseline on a custom fabricated C18 column (1.8 µm bead size, 120 Å pore size, UChrom C18 (nanolcms) 0.5 mm x 200 mm) at a flow rate of 15 µl/minute equilibrated with 5% B. Peptides were eluted with a gradient from 5% B to 100% over 160 min. A 'flow-through' fraction was collected and offline desalted, and fractions were collected for 1 min each, concatenating throughout the gradient for a total of 12 mixed fractions of nearly equal complexity. Fractions were dried by vacuum centrifugation and stored at -80 o C.

Mass spectrometry and data analysis
All fractions for mass spectrometry were suspended in 5% (v/v) acetonitrile, 0.1% (v/v) trifluoroacetic acid in LC/MS-grade water and analyzed by direct injection on a Waters M-class Acquity column (BEH C18 column, 130 Å, 1.7 µm, 0.075 mm x 250 mm) at 0.3 µL/minute using a nLC1000 (Thermo Scientific). Peptides were gradient eluted from 3% ACN, 0.1% formic acid to 20% ACN in 100 min, 20 to 32% ACN in 20 min, and 32 to 85% ACN in 1 min. Mass spectrometry analysis was performed using an Orbitrap Fusion (Thermo Scientific). MS1 were collected at 120,000 FWHM resolution from 380-1500 m/z, 2 x 10 5 automatic gain control (AGC), 50 ms max fill time, with 20 s dynamic exclusion +/-10 ppm using the data dependent mode "Top Speed" for 3 s on the most intense ions. MS2 was performed in the linear ion trap 13 with a 1.6 m/z isolation window, 35% HCD collision energy, 1x10 4 AGC, 35 ms max fill time.
All Thermo raw files were processed using MaxQuant (version 1.5.6.5)(66), with MS/MS spectra searched using the Andromeda search engine (67) against a Uniprot human database (reviewed version downloaded April 1, 2015) supplemented with sequences from the CRAPome database (68). The search was limited to peptides with a minimum length of seven amino acid residues, and a maximum of two missed cleavages with trypsin/P specificity, and included variable modifications for methionine oxidation and protein N-terminal acetylation, with carbamidomethyl Cysteine as a fixed modification, with search tolerances of 20 ppm for MS1 precursor ions, and 0.5 Da for MS/MS peaks. Protein and peptide level false discovery rate (FDR) thresholds were set at 1%. The MaxQuant method for label-free quantification (LFQ) was used to generated normalized relative protein abundances, with the match between runs option on, and protein groups quantified using unique and razor peptides. The Perseus software platform (69) (version 1.5.5.3) was used to annotate proteins and filter contaminants from the MaxQuant proteinGroups.txt file. Only protein groups with at least two valid LFQ values (with MS/MS identified MS1 features) over all samples were considered for analysis.

Data availability
All mass spectrometry raw files, and MaxQuant analysis files (parameters.txt, summary.txt, evidence.txt, proteinGroups.txt and Perseus annotated proteinGroups.txt) are uploaded to the MassIVE repository at UCSD, under the ID "MSV000080549".

Experimental Design and Statistical Rationale
Microscopy data from measurements taken in ImageJ were analyzed and graphed in SigmaPlot v13.0 and SPSS v23, using unpaired student's t-test, ANOVA, reporting 95% confidence intervals. All parametric statistics passed a Shapiro-Wilk test of normality and Brown-Forsyth test of equal variance; values for some measures were log-transformed prior to analysis to ensure normality where appropriate. Data are reported as mean values with error bars depicting the standard error of the mean (SEM). For measurements of phenotypes visualized as individual cells in a larger population, we considered "n" to equal mean values from three random observations of a single coverslip (representing a unique cell culture environment). For observations of intact units made up of multiple cells (embryoid bodies, neurospheres) we considered "n" to be a discrete measurement from each individual multicellular unit. Data points were excluded from analysis where obvious preparation failures were evident (i.e. lack of specific staining, cells imaged were not fixed in mounting media). For proteomic profiling, biological replicates were defined as separate plates at the iPSC stage, and independent pools of multiple spheroids at the EB, NS and CO stages. Figure 2 Principal components analysis (PCA) of the relative protein abundance values (MaxQuant LFQ values) was performed using the R package FactoMineR (70). The proteins most highly correlated and anticorrelated with PC1 and PC2 were selected using the dimdesc function (p < 0.01 and absolute value of Pearson correlation coefficient > 0.7). For enrichment analysis, the UniProt identifiers were converted to gene symbols, removing duplicate identifiers.

Methods for
The Fisher exact test was used to test for significant enrichment of these sets against four libraries downloaded from the EnrichR and ARCHS4 projects (71)(72)(73): ARCHS4_Tissues, GO_Molecular_Function_2017, Jensen_COMPARTMENTS, and Reactome_2016. We used custom-built R code to perform the enrichment analysis to enable control of the set of genes used as the enrichment background, which avoids bias introduced by whole genome background gene sets used by the online EnrichR tool. The background used for performing the Fisher exact test 15 was the complete set of gene symbols that mapped to proteins in the PCA analysis. The heatmap of proteins correlated with PC1 and PC2 in Fig. 2B was generated using the morpheus R interface package (MIT Broad Institute). Figure 3, Supplemental Figure 4 and Table I Protein LFQ values were considered for differential analysis if the total number of missing intensities for that protein group in each stage was less than three. Differential analysis of the LFQ protein abundances for the two comparisons, T21 vs. D21 and T21+DMSO vs.

Methods for
T21+harmine, was performed by using the empirical Bayes modeling framework in the R package, limma (74,75), with the resulting p-values corrected for multiple-hypothesis testing using the Benjamini-Hochberg method (76). Gene set enrichment analysis was performed for the proteins showing differential expression in the T21 vs. D21 and T21 +/-harmine comparisons (FDR < 0.1). Differential analysis statistics and annotations for all quantified proteins in the trajectory are reported in supplemental Table 2, and CO stage differential analysis for the set of markers shown in Fig. 4B are reported in supplemental Table 3. Figure 4 Functional enrichment using the Wilcoxon rank sum test was assessed independently for both the T21 vs. D21 effect (x-axis) and the harmine effect (y-axis), using the functional gene set modules identified by Olmos-Serrano et al. 2016 in post-mortem human brain tissue from Down syndrome individuals and age-matched euploid controls (60). The moderated t-statistics from the empirical Bayes analysis in Fig. 3 Table 4.

Methods for Supplemental Figure 5,6
Enrichment analysis was performed with the Fisher exact test as in Fig. 2B, using cell type specific gene sets assembled using supplemental tables from previous studies (54, 77), and the indicated libraries downloaded from the EnrichR and ARCHS4 projects (71,72,78).
Complete enrichment results are listed in supplemental Tables 5 and 6. Figure 5 To generate the network, we used Cytoscape (v3.5.1), and performed a STRING protein query through the "Import Network from Public Databases" dialog, entering all differentially expressed proteins in harmine treated T21 COs (q < 0.1), using confidence score cutoff equal to 0.55 and maximum additional interactors equal to zero. The log2 ratios for the T21 vs. D21 and T21+harmine effects were then mapped to node border color and node fill color, respectively. This network was then clustered using MCODE (79) within the clusterMaker app (80). MCODE parameters were as follows: include loops, unchecked; Degree cutoff, 2; Haircut, checked, Fluff, checked; Node score cutoff, 0.2; k-Core, 2; Max Depth, 100; Create new clustered network, checked. The resulting networks were then exported as an svg file. Functional enrichment analysis was used to annotate each cluster with the most significant gene ontology function. The edge thickness and grayscale indicates the confidence in interaction (81).

Harmine effects on mitotic index and embryoid body size for euploid and trisomy 21 lines.
We used two human iPSC lines grown in feeder-free culture for proteomic and phenotypic analyses: one line was derived from a female individual with Down syndrome (C2-4 line, referred to hereafter as T21) (63) and a second was derived from an unrelated female euploid individual (I50-2 line, referred to hereafter as D21) (Experimental Procedures). To modulate DYRK1A activity, we used the β-carboline alkaloid, harmine, based on reports that DYRK1A inhibition and knockdown was reported to rescue impaired neurogenesis in DS iPSCderived neural progenitor cell cultures (82). Harmine has also been shown to increase proliferation in euploid neural progenitor cells (83), and remains one of the most selective DYRK1A inhibitors to date (84).
To identify an optimal concentration of harmine for proteomic analysis of the T21 organoid differentiation trajectory, we examined EB size after continuous exposure to a range of harmine concentrations, from the iPSC to the EB stage. In the absence of drug, EBs produced from T21 iPSCs were ~61% (s.e.m. = 17%) smaller relative to D21 EBs (supplemental Fig. 2A).
Below 1 µM, harmine produced a dose-dependent increase in T21 EB size, with no significant change for D21 EBs. At concentrations above 1 µM, harmine led to reduced EB size in both T21 and D21 lines, indicating a negative effect on viability and/or growth rate. Based on these observations, we tested the effect of 300 nM harmine on proliferation in the iPSC lines. The T21 iPSC line exhibited a lower mitotic index relative to D21; treatment with 300 nM harmine resulted in an increased mitotic index to the level found in untreated D21 cells (supplemental by 30-50% (85). Submicromolar harmine has been shown to inhibit DYRK1A-dependent phosphorylation events in cells (61), and to attenuate axonal growth in mouse primary neurons overexpressing Dyrk1a (86). Based on these results, we used 300 nM harmine, present throughout all differentiation stages, from iPSC to CO, for the proteomic analysis of the CO developmental trajectory (Fig. 1A). To identify proteins that change in abundance during the differentiation process, and show differential expression with harmine treatment, we generated quantitative proteomes at multiple stages along the CO developmental trajectory (Fig. 1A) (Experimental Procedures). All lines were grown in parallel to match culture conditions (T21 + DMSO, T21 + 300 nM harmine, and D21 + DMSO), using a modified version of the Lancaster protocol (52) similar to that used in a recent Zika virus study in COs (64) (Experimental Procedures). We collected three biological replicates at the iPSC stage and two biological replicates at the spheroid stages (EB, neurosphere, CO), pooling multiple spheroids in each of the last three stages to obtain sufficient material for high-depth label-free proteomic profiling. Biological replicates in this experiment refer to independent cultures grown in parallel. We found that 300 nM harmine treatment led to the systematic failure of euploid (D21) iPSC-derived CO formation; consequently, D21 harminetreated samples were not collected for any stage in this experiment.
We used an optimized offline peptide fractionation method to simplify peptide mixtures for LC/MS/MS proteomic analysis on a quadrupole/linear ion trap Orbitrap mass spectrometer

20
To gain further insight into the biological processes associated with these transitions, we performed functional enrichment analysis on the proteins most highly correlated and anticorrelated with the first two principal components, PC1 and PC2 (Fig. 2B, C). We tested for nonrandom overlap of these proteins with gene set libraries from the EnrichR project (71)(72)(73), which includes libraries related to molecular function (89-91), subcellular compartments (92), and tissue-specific gene sets from the ARCHS4 libraries from EnrichR. The ARCHS4 libraries contain gene sets derived from large-scale co-expression analysis that identifies highly covarying sets of mRNAs over thousands of different perturbations and genetic backgrounds in human cell lines and tissues. Enrichment analysis against these libraries can be used as an unbiased method to infer gene function, protein-protein interactions, and human tissue expression patterns (73) (Experimental Procedures).
Proteins positively correlated with PC1 were characterized by low expression in the iPSCs, EBs and NSs, and high expression in COs (Fig. 2C). In this group, we found enrichment for terms related to human brain specific expression (  Table 1). We also observed significant enrichment in proteins found in postsynaptic density structures (not shown, adj. p-value = 2.3 x 10 -4 ), and other neuronal subcellular structures and functions (Fig. 2B). Proteins enriched in this category included DCX, which is important for neuronal migration and cortical lamination, and NCAM1, which is involved in neurite extension (93). Significant enrichments for subcellular structures important during neurogenesis were consistently found. These included cell-cell junctions, which are critical for the regulation of neuroepithelial cell polarity during early forebrain development (94). Intriguingly, highly significant enrichment was found for functions related to primary cilium assembly. The term "Assembly of the primary cilium" from Reactome showed a high degree of overlap (56 out of 187) with PC1 correlated proteins (adj. p-value = 2 x 10 -11 ). This category included 15 intraflagellar transport complex proteins, four BBSome subunits (BBS2, BBS7, BBS9 and LZTFL1), and PAFAH1B1. These proteins are involved in primary cilia assembly and trafficking, and have essential roles in morphogen signaling and patterning during brain development. Disruption of many of these genes cause severe neurodevelopmental defects in humans (95). Primary cilia play an important role in the proliferation and differentiation of neuroepithelial cells and apical radial glial cells, which undergo dramatic proliferative expansion during brain development, and are present in COs (96).
Both of these cell types have a primary cilium anchored in the apical plasma membrane that projects into the ventricular lumen. Primary cilia are involved in the establishment and maintenance of apical-basal polarity of radial glial cells in the developing brain (97). Expression in this group of proteins sharply increased in COs, suggesting that cell types with primary cilia increase in number during organoid maturation.
The PC1 anti-correlated proteins, which show high expression in the first three stages and low expression in the CO stage, are enriched in functions related to RNA binding, rRNA processing and mitochondrial translation. Consistent with loss of pluripotency and commitment toward neuronal identities over the course of the developmental trajectory, we found highly significant enrichment in markers of human embryonic tissue (Fig. 2B) and neuronal progenitor cells (adj. p-value = 2.3 x 10 -10 , ARCHS4_Tissue term "NEURONAL PROGENITOR CELLS", supplemental Table 1). Many of the PC1 anti-correlated proteins that map to this progenitor cell type term are involved in regulating the cell cycle and mitotic spindle (e.g. PLK1, CENPF, CCNB2 and AURKA), and include ASPM, which is associated with microcephaly in humans when both copies are disrupted (98). We found two other microcephaly-associated genes that were associated with the set of PC1 anti-correlated proteins, WDR62 and CENPJ, although these were not found to be enriched in these same functional categories. Together, ASPM, WDR62, and CENPJ are involved in spindle orientation in apical progenitors (99), and regulate centriole duplication and the timing of apical progenitor delamination (100).
Proteins correlating with PC2 show differential expression patterns specific to the To summarize, we found that the greatest variation in protein abundance is associated with the last two developmental transitions (NS and CO). During these stages, neurogenic precursor cells undergo proliferative expansion, and early born neurons begin differentiation and maturation, leading to a diversification of multiple cell types within the organoid. Functional analysis of changes in proteins at the neurosphere stage revealed alterations in chromatin structure, cell polarity, and cell-cell interactions, suggesting important roles for these processes 23 during neural induction. At the CO stage, we found coordinated upregulation of proteins with functions related to neuronal differentiation and primary cilia assembly, and downregulation of proteins involved in maintaining pluripotency.

Stage-specific differential expression analysis.
Having established the primary sources of proteome variation and associated functional signatures, we next performed a differential expression analysis to identify proteins that showed differences between the T21 and D21 lines and responded to harmine treatment at each differentiation stage. We used an empirical Bayes moderated t-test method that provides greater statistical power to detect differentially expressed genes than the unmoderated Student's t-test, particularly when sample numbers are low (75). This method has been validated for proteomics data (103), and has been successfully used in large-scale differential proteomics experiments (104-106). Differential analysis was performed using custom R scripts that employ the LIMMA modeling framework (74) (Experimental Procedures).
We found that proteins encoded on HSA21 were expressed, on average, at 30-50% higher levels in T21 across the four stages, relative to the euploid control, consistent with the increased gene dose associated with the extra copy of chromosome 21 ( Interestingly, at the iPSC, EB and NS stages, we detected only two differentially expressed proteins with harmine treatment, that is, in the T21+DMSO vs. T21+harmine comparisons, hereafter referred to as the T21+harmine effect (Table I). In contrast, harmine 24 treatment led to large-scale changes in protein abundance in T21 COs, resulting in 1498 differentially expressed proteins for the CO T21+harmine effect (q < 0.1) (Fig. 3C, Table I).
Examination of the log2 ratio distributions for the T21+harmine effect, at each stage in the T21 line, revealed that the iPSC, EB and NS log2 ratios exhibit much less dispersion around zero, compared to the CO T21+harmine effect (supplemental Fig. 4B, D, F, G). This suggests that harmine treatment resulted in fewer large-magnitude effects on protein expression at earlier stages, where cell type heterogeneity is lower compared to the CO stage.
We next examined the proteins with differential expression specific to the CO stage, where harmine treatment exhibited the most dramatic effect on abundance (Fig. 3C). We found that a number of the proteins with the lowest expression in T21 COs, relative to D21 COs, were Wnt ligands (e.g. WNT7A, WNT7B, WNT1) and transcription factors involved in the specification of laminar identity in post-mitotic neurons (e.g. TBR1, BCL11B)(107-109).
Strikingly, harmine treatment increased the expression for these proteins in T21 COs to near D21 levels ( Fig. 3C). Conversely, many extracellular matrix proteins were expressed at higher levels in T21 COs, relative to D21 COs. For example, Reelin (RELN), an extracellular glycoprotein involved in regulating neuronal migration during cortical lamination, was ~5-fold higher in T21 COs, and decreased by 2.6-fold with harmine treatment. Other ECM proteins that exhibited this trend (higher T21 vs. D21 CO expression and reduction by harmine) included COL1A2, DCN, and COL5A1, proteins recently found to be uniquely expressed in a mesenchymal cell type identified in human fetal brain and COs by single cell RNA-seq (54). In this same study, COL5A1 was found in the periphery of CO cortical regions, colocalized with ECM embedded collagen fibers, suggesting that T21 COs contain an increased population of mesenchymal cell types and/or upregulated ECM secretion that is normalized by harmine treatment. Dysregulated 25 extracellular matrix remodelling would be expected to disrupt neuronal migration and selfrenewal of neurogenic precursor cells in the developing brain (110).
Overall, we found a large number of proteins with differential expression in the T21 vs. D21 comparison that changed in abundance toward D21 expression levels upon harmine treatment. Of the 1,444 proteins that were differentially expressed in the T21 vs. D21 CO comparison (Table I), an unexpectedly large number, 425 (29%), changed significantly with harmine (p < 2.2 x 10 -16 , odds ratio = 3.1, Fisher exact test). Of the 946 proteins that were high in the T21 COs relative to D21 COs, 3.3-fold more proteins decreased with harmine than increased (226 vs. 68) Conversely, of the 498 proteins that were low in the T21 COs, 1.4-fold more proteins increased than decreased (77 vs. 54) (supplemental Fig. 5A-C). To test whether this "harmine-normalization effect" was evident on a global level, we performed a correlation analysis between the two differential expression effects, using the LFQ values in the  Table 3).

Concordance between human DS brain signatures and dysregulated CO protein expression.
To clarify the potential functional ramifications of differential expression in the CO comparisons, we mapped differentially expressed proteins onto a recently described transcriptome analysis of developmental stages for normal and Down syndrome human brains, spanning fetal to adult stages (60). In that study, the WGCNA algorithm (111) was used to identify gene modules with highly correlated temporal mRNA expression patterns, many of which were discordant between euploid and Down syndrome brains. Our analysis assumes that discordant mRNA expression will be reflected on the protein level, and should therefore capture dosage-sensitive proteomic changes related to trisomy 21 that are not under post-transcriptional control. In this way, we can determine whether the impacts of trisomy 21 observed in the proteome are consistent with dysregulated signatures observed in human DS brain.
We tested whether the proteins that map to a given DS brain module have significantly increased or decreased expression in each of the two CO comparisons used in this study, T21 vs. D21 and T21+harmine, adapting a previously described statistical method for threshold-free annotation enrichment (112). The method uses a two-sided Wilcoxon rank sum test on the empirical Bayes t-statistics from the differential analysis used in Figure 3, producing a position score and an adjusted p-value for each DS brain module that indicates whether there is significantly increased or decreased protein expression within the module (Experimental Procedures). This analysis enabled us (i) to identify proteins in COs with functional overlap with the dysregulated gene modules in human DS brain, and (ii) to visualize module-specific biases in the two treatment effects (Fig. 4C), analogous to the two axes of the scatter plot in Figure 4A.
We found many DS brain modules with significant shifts in median protein log2 ratios (for both CO effects) that mirrored the correlation structure of the protein log2 ratio scatter plot in Figure 4A (Fig. 4C). That is, many co-expressed gene modules found in DS brain tissue were associated with significant expression differences in the T21 vs. D21 and T21+harmine comparisons. For example, the M4 and M45 modules, enriched in genes involved in neuron 27 differentiation and development, showed significant expression shifts for the T21 vs. D21 and the T21+harmine comparisons, with an overall shift toward lower T21 expression, relative to D21, and increased T21 expression upon harmine treatment (Fig. 4C, supplemental Table 4). A complementary enrichment method, using the differentially expressed protein sets (q < 0.1) across all stages, confirmed that the CO stage showed the most highly significant enrichment in DS brain modules (supplemental Fig. 6A).
The M45 module mapped to 36 differentially expressed proteins that were up-regulated by harmine in the T21 COs, including BCL11A, BCL11B, CHD4 and CHD3 (  Table 4). BCL11A and BCL11B are paralogous subunits of the BAF complex that are critical for neuron subtype specification (113). Loss of function mutations in BCL11A cause intellectual disability, which has been attributed to transcriptional dysregulation during cortical development (114). CHD4 and CHD3 are mutually exclusive subunits of the chromatin remodeling NuRD complex, which regulates neural progenitor proliferation and layer specification of differentiated neurons during cortical development (115).
The M4 module mapped to 33 differentially expressed proteins that were upregulated by harmine in the T21 COs; these included WNT7A, TBR1 and FOXG1 (Fig. 4B). WNT7A was one of the largest magnitude changes induced by harmine treatment, and plays a crucial role in forebrain development and synaptic organization (116). WNT7A mRNA expression was previously reported with low expression in DS iPSC-derived neural progenitors, relative to an isogenic euploid control (82). In the same study, knockdown of DYRK1A expression rescued the low WNT7A mRNA expression in DS cells, increasing it to euploid levels, suggesting a potential causal relationship between altered Wnt signaling, neurogenesis, and the over-expression of DYRK1A protein.

28
The M50 and M54 modules exhibited higher T21 expression, relative to D21, and decreased T21 expression with harmine treatment (Fig. 4C). Consistently, expression of mRNAs in these two modules was elevated in multiple DS brain regions (60). Both modules contain genes with previously annotated functions related to immune response that appear to be dysregulated in DS (117). Four interferon receptor subunits, IFNAR1, IFNAR2, IFNGR2, and IL10RB, are encoded on HSA21, and have been shown to be upregulated by ~1.5-fold on the protein level in DS cells (118). Upregulated interferon receptor expression in DS is strongly implicated in the hyperactive response to interferon stimulation found in skin fibroblasts, lymphoblastoid lines, primary monocytes, and T cells (118). A recent large-cohort proteomics study demonstrated that DS individuals exhibit higher circulating levels of pro-inflammatory cytokines (119).
Altogether, these results reveal that T21 dysregulated protein expression in COs maps to specific human DS brain gene co-expression patterns that are discordant between DS and euploid individuals, particularly during prenatal developmental stages. Moreover, a number of these T21discordant modules are enriched in proteins with functions related to neurogenesis, neuronal differentiation, and immune response, processes that are known to be disrupted in DS. Proteins within these modules respond to harmine treatment, leading to normalization of expression levels toward D21 levels.

Harmine upregulated proteins are enriched in a diverse array of functions, cell typespecific markers and transcription factor targets.
We next performed a large-scale enrichment analysis on differentially expressed proteins across all stages using the EnrichR libraries (Experimental Procedures). In agreement with the module analysis, we found that harmine upregulated proteins in T21 COs are enriched in markers of neurogenic precursors found in human fetal brain, and markers associated with human cortical brain structures (supplemental Fig. 6B To visualize functionally-related differentially expressed proteins that may participate in protein complexes or regulatory interactions, we mapped the harmine-modulated differentially expressed proteins (q < 0.1) found in T21 COs onto a STRING interaction network that combines experimentally verified physical interactions with functional interactions found in the literature (81). We then clustered the resulting network to identify densely connected and functionally related subnetworks, and visualized the expression ratios for the T21 vs. D21 effect (node border color) and the T21+harmine effect (node color) (Fig. 5A). One of the larger subnetworks ( Fig. 5B) contained proteins enriched in functions related to neurogenesis (q-value = 7 x 10 -12 ) and WNT signaling (q-value = 3 x 10 -10 ) (for enrichment results of all clusters, see supplemental Data File 1). A number of these proteins are subunits of multiprotein complexes involved in chromatin remodeling. For example, we found four subunits of the NuRD complex (CHD3, HDAC2, MBD3, GATAD2B), which is required for layer specification in the developing cortex (Fig. 5B)(115). Nodes for these subunits were connected to differentially expressed proteins NOTCH1, GLI3, BCL11B and TCF7L2, suggesting important functional relationships between chromatin remodeling and Notch, sonic hedgehog and Wnt signaling in the harmine response of T21 COs. The number of proteins in the two subnetworks in Figure 5B that showed a "harmine normalization" effect was more than 2-fold greater than non-normalized (95 vs. 45 out of 139 proteins total). While harmine leads to only a partial normalization of T21discordant expression, our evidence suggests that harmine inhibition of a critical target, potentially DYRK1A, results in widespread normalization of protein abundance in T21 COs, particularly for central regulatory transcription factors and morphogens involved in neurogenesis and neuronal differentiation.

DISCUSSION
We describe a high-depth, quantitative proteomics study of human COs grown from trisomy 21 and euploid iPSC lines, in which we evaluate the proteome-wide effects of a DYRK1A inhibitor, harmine, at four stages of development, starting with iPSC and culminating at COs. On a phenotypic level, we observed a decreased mitotic index phenotype in the T21 iPSC line that agrees with the previous characterization of the C2-4 line (63). Reduced proliferative capacity in human DS lines has been documented previously, notably for iPSCderived neural precursor cells (82) and primary fibroblasts (120). Consistently, we found that this phenotype could be reversed by long-term treatment with sub-micromolar doses of the DYRK1A inhibitor harmine.
The proteome depth of our organoid trajectory is comparable to a recent comprehensive brain-region resolved proteomics study of mouse brain that identified 13,061 proteins across all regions and cell types (121). We found that the dominant variation in protein expression over the developmental trajectory occurred during the NS to CO transition, when neuroepithelial cells undergo asymmetric neurogenic divisions to form radial glia and early born neurons(94), a process that has been modeled in COs (122). By comparing proteome profiles from euploid COs to T21 COs grown with and without 300nM harmine, we identified a number of cell fatedetermining pathways and functional gene signatures that were dysregulated in T21 COs, and normalized by harmine treatment. These include proteins involved in chromatin remodeling and alternative splicing, both of which play critical roles in regulating neurogenesis and forebrain development (123). Most striking was the large number of differentially expressed proteins upon harmine treatment in the T21 COs compared to the earlier iPSC, EB, and NS stages, particularly for proteins associated with early and late differentiated neurons, synaptogenesis, and ECM proteins with known roles in neuronal migration. This stark difference in molecular responses to harmine in the different stages point to the crucial importance of cellular and developmental context, and highlights a distinct advantage of modeling human neurodevelopmental disease states in COs for evaluating potential therapeutic interventions.