Single Cerebral Organoid Mass Spectrometry of Cell-Specific Protein and Glycosphingolipid Traits

Cerebral organoids are a prolific research topic and an emerging model system for neurological diseases in human neurobiology. However, the batch-to-batch reproducibility of current cultivation protocols is challenging and thus requires a high-throughput methodology to comprehensively characterize cerebral organoid cytoarchitecture and neural development. We report a mass spectrometry-based protocol to quantify neural tissue cell markers, cell surface lipids, and housekeeping proteins in a single organoid. Profiled traits probe the development of neural stem cells, radial glial cells, neurons, and astrocytes. We assessed the cell population heterogeneity in individually profiled organoids in the early and late neurogenesis stages. Here, we present a unifying view of cell-type specificity of profiled protein and lipid traits in neural tissue. Our workflow characterizes the cytoarchitecture, differentiation stage, and batch cultivation variation on an individual cerebral organoid level.


Chemicals and Reagents
Synthetic isotopically labeled (SIL) peptide standards (SpikeTides_L crude) were from JPT Peptide Technologies Inc. (Acton, MA, USA). Isotopically-labeled ganglioside (GS) internal standards (GM1 and GM3) were synthesized by Lukas Opalka (Faculty of Pharmacy, Charles University). Pierce BCA Protein Assay Kit reagents were from Ther-moFisher Scientific, Waltham, MA, USA. The ultrapure water was prepared in the purification system (arium® Comfort System, Sartorius).
For qPCR assays and Western blotting (WB), COs were harvested at representative time points (D50, D85, and D110), and 5-7 pooled COs were analyzed per each time point.

RNA isolation, cDNA synthesis, and Real-Time quantitative PCR (qPCR) assay
Total RNA was isolated from undifferentiated iPSCs on day 0 (D0) and differentiating COs using the RNA Blue (Top-Bio, Prague, Czechia). The concentration and purity of isolated RNA were determined using NanoDrop 1000 (ThermoFisher Scientific). According to the manufacturer's instructions, the RNA was transcribed to cDNA using Transcriptor First Strand cDNA Synthesis Kit (Roche, Basel, Switzerland). qPCR was performed from the cDNA samples using LightCycler® 480 SYBR Green I Master kit (Roche) on LightCycler 480 II (Roche). Samples were analyzed in technical triplicates, and results were normalized to respective glyceraldehyde 3-phosphate dehydrogenase (GAPDH) gene values. Primers are listed in Supplementary Tab. S12.
Immunoblotting assay for cell-specific markers WB was performed as described previously in Fedorova et al., 2019 25 . Briefly, harvested organoids were lysed in 50mM Tris-HCL with 1% SDS and 10% glycerol. Protein concentrations were measured using DC Protein Assay (Bio-Rad, Hercules, CA, USA) and adjusted to the same level. Proteins were separated using a 10% SDS-PAGE and transferred onto the PVDF membrane (Merck Millipore, Darmstadt, Germany). Membranes were blocked in 5% skimmed milk in 4mM Tris-buffered saline with 0.05% Tween 20 (TBS-T) and incubated with primary antibodies (4 °C; overnight), followed by incubation with horseradish conjugated secondary antibodies (ambient temperature; 1 h). Proteins were visualized by ChemiDoc (Bio-Rad) using Amersham ECL Prime western blotting Detection Reagent (GE Healthcare Life Sciences, Chicago, IL, USA). Following primary antibodies were used in the immunoblotting analysis: DCX (sc-271390, Santa Cruz Biotechnology, Dallas, TX, USA); S100B (ab11178, Abcam, Cambridge, GB); SOX2 (4900S, CST, Danvers, MA, USA); TUBB3 (5568S, CST); MAP2 (8707T, CST); NEFL (2837S, CST); SYN1 (5297S, CST); GFAP (12389P, CST) and ACTB (3700S, CST). The performance highly depends on the antibodies' quality. The whole membrane should be incubated for accurate results with the primary antibody to reveal protein isoforms, post-translational modifications, or antibody non-specificity. In our laboratory, a horseradish peroxidase-conjugated secondary antibody was used for detection, so only three specific proteins can be analyzed in a 30 µg of organoid protein. This method can be modified by using fluorescent secondary antibodies, whereby two (or more) primary antibodies derived from different animals can be applied to the membrane; in this case, up to six proteins can be detected from the 30 µg of organoid protein. Another commonly used approach to spare the protein sample is to cut the membrane according to the size of the protein and incubate the different parts of the membrane in antibodies individually. It is necessary to examine proteins with different molecular weights and have well-validated primary antibodies, and then 6-9 different proteins can be detected.
Chemical synthesis of 13 C18 labeled ganglioside GM3 GM3 The amount of 5.3 mg (0.0085 mmol) of lactosyl sphingosine was mixed with 3.3 mg (0.0109 mmol) of 13 C18 stearic acid and 3.5 mg (0.0259 mmol) of 1-hydroxybenzotriazole, dried on the vacuum, dissolved in 0.5 mL of dry DCM and 0.75 mL of dry MeOH and cooled with ice to 0 °C. 3 uL (0.0170 mmol) of N-(3dimethylaminopropyl)-N′-ethylcarbodiimide were added, and the temperature was allowed to increase to RT. The reaction mixture was stirred at RT under argon atmosphere for 24 hours. After 24 hours, the resulting suspension was dissolved in excess of MeOH and the solution was evaporated with silica. The column chromatography CHCl3/MeOH/2.5M NH4OH 60:40:10 provided 7.4 mg (96%) of 13 C18 labeled lactosylceramide as a white solid. The product identity was confirmed using MS: C3013C18H91NO13, [M+Na]+ calculated 930.70, measured 930.72; [2M+Na]+ calculated 1838.41, measured 1838.84. In total, 7.4 mg (0.0081 mmol) of lactosylceramide from the previous reaction was mixed with 5.2 mg (0.0082 mmol) of cytidine-5′-monophospho-N-acetylneuraminic acid sodium salt and dissolved in 2 mL of TRIS buffer (pH 8, 100 mM). 4.3 mg (0.0082 mmol) of sodium taurodeoxycholate was added into the resulting suspension, and the suspension was shortly sonicated to yield a cloudy solution. Using TRIS buffer, the enzyme α-2,3-sialyltransferase from Pasteurella multocida (1 UN) was added into the solution, followed by the addition of 1 mg (10.5 µmol) of MgCl2 and 100 µL of the alkaline phosphatase from bovine intestinal mucosa in TRIS buffer (the equivalent of 100 DEA units). The reaction mixture was stirred in the incubator at 37 °C and monitored using MS. After 6 hours, the MS showed complete consumption of lactosylceramide, and the reaction was quenched. The solvent was carefully evaporated, the residue was dissolved in MeOH and evaporated with silica. The product was purified using column chromatography CHCl3/MeOH/2.5M NH3 100:45:10 to yield 2.9 mg (30 %) of the 13 C18 labeled GSs GM3 as a white solid. Chemical synthesis of 13 C18 labeled ganglioside GM1 GM1 The amount of 5 mg (0.0032 mmol) of monosialoGSs GM1 (NH4+ salt) was dissolved in 1 mL of acetate buffer (pH 5.8, 100 mM), 11 mg (0.099 mmol) of CaCl2 and 4 mg (0.0077 mmol) of sodium taurodeoxycholate were added followed by the addition of sphingolipid ceramide N-deacylase from Pseudomonas sp. (0.25 UN). The reaction mixture was stirred at 37 °C in the incubator for 48 hours. After 48 hours, the resulting suspension was centrifuged (6000 RPM, 4 min) and the supernatant was collected. The solid residue was washed with 300 uL of water and centrifuged again. The supernatant was collected and added to the previous one. The combined supernatants were applied on the prefilled RP18 column and washed first with water, then with 65% MeOH and finally with 85% MeOH. This purification provided 1.8 mg (44%) of the lysoGSs GM1 as a white solid. The product identity was confirmed using MS: C55H97N3O30, [M-H]-calculated 1278.61, measured 1278.70. In total, 3 mg (0.0099 mmol) of 13 C18 stearic acid were mixed with 1.7 mg (0.0149 mmol) of N-hydroxysuccinimide and dried on the vacuum. 1 mL of dry DCM was added, and the reaction mixture was cooled with ice to 0 °C. 3.5 uL (0.0198 mmol) of N-(3dimethyl aminopropyl)-N′-ethyl carbodiimide was added, and the temperature was allowed to increase to RT. The reaction was stirred at RT under an argon atmosphere for 48 hours. After 48 hours, the reaction mixture was diluted with 2 mL of DCM, evaporated on silica, and purified by column chromatography Hex/EtOAc 3:1 with 0.5% AcOH to obtain 2.6 mg (66%) of succinimide-1-yl 13 C18 stearate. The product was used in the next reaction without further characterization. In total, 1.8 mg (0.0014 mmol) of lysoGSs GM1 was mixed with 1 mg (0.0025 mmol) of succinimide-1-yl 13 C18 stearate and dried on the vacuum. 250 uL of dry DMF and 1.2 uL (0.0070 mmol) of DIPEA were added, and the reaction mixture was stirred at RT under argon atmosphere for 48 hours. After 48 hours, DMF was evaporated and the residue was dissolved in MeOH and evaporated on silica. The crude product was purified using column chromatography, first CHCl3/MeOH 60:40, then CHCl3/MeOH/2.5M NH4OH 60:40:10 to yield 1.7 mg (77%) of 13 C18 labeled GSs GM1. The product identity was confirmed using MS: C5513C18H131N3O31, [M+H]+ calculated 1564.94, measured 1564.95.

Ganglioside assay validation
We prepared a ten-point matrix-matched calibration curve adding the isotope-labeled internal standards to the pooled lipid extract in concentrations from 3.e -3 to 50 µM for GM3 and from 3.e-5 to 0.6 µM for GM1. The four low-concentrated dilutions were analyzed in technical quadruplicates, and the six highconcentrated dilutions in triplicates, given biological material availability (Tab. S5a). The calibration curves for positive and negative ion detection modes are in Fig. S7. The linear regression correlation coefficients were >0.98. LOD was defined as S/N >3, and LOQ as the lowest concentration with CV <20 %. LOD and LOQ for individual GSs were corrected using respective response factors LODGS = (LODGM3*/RFGC). The linear response range was 8.e-3 to 0.6 µM for GM1 and GM2 and from 8.e-3 to 50 µM for other GSs; Precision on QC samples (n=5) was between 1.8 to 12.1 % of the CV (Tab. S5b). Recoveries of GM3 and GM1 were 82.3 % and 105.1 %, respectively. Matrix effects determined at concentration 0.3 µM were negligible 108.9 % and 110.7 % for 13C_GM3 and 13C_GM1, respectively (Tab. S6).

Protein assay validation
We prepared a ten-point calibration in the range of expected sample concentrations determined in the QC sample (Tab. S7a). The calibration curves for quantifier peptides in positive ion mode are in Fig. S8. We calculated sample matrix LOD and LOQ for each protein using the standard deviation (STDEV) of first reproducibly (CV<20 %) detected quantifier transition of ST peptide (Tab. S7b):

LOD=3*STDEV/SLOPE(calibration curve) LOQ=10*STDEV/SLOPE(calibration curve)
We assessed the QC samples' matrix effects, comparing ST peptide responses in the sample matrix after SPE with corresponding responses in the neat solution. The matrix effect was calculated: ST(matrix) peak area/ST(solution) peak area*100. The matrix effects were moderate on average 32 %, except for severe 83 % for S100B -peptide AMVALIDVFHQYSGR and 79 % for SOX2 -peptide LLSETEK (Tab. S3). However, quantifier peptides' responses were reproducible in neat solution (CV<18 %) and the sample matrix (CV<11 %) (Tab. S3). We compared total protein concentrations in IPA extracted QC sample protein pellets with non-extracted QC sample protein pellets. In parallel, we analyzed the IPA extracts to detect potential protein losses. We tested the reproducibility of the protein pellet solubilization. To assess the protein extraction reproducibility, we dried equivalent aliquots of the QC sample (n=2) and performed the entire proteolytic protocol with an average CV<12 % for individual protein marker levels. The overall proteomics protocol reproducibility (CV, n=8), in QC samples without additional dry-down and reconstitution step, was on average, 11 %, from 3 % (GAPDH) to 21 % (NEFM) (Tab. S13).

Heterogeneity in individual cerebral organoids
To identify abnormal COs, we sorted them based on three parameters: (i) a sum of all protein marker levels (PML), (ii) a sum of all lipid marker levels (LML), and (iii) the ratio of neuronal protein marker (sum of DCX, TUBB3, MAP2, and NEFM) to the glial marker levels (sum of FABP7 and S100B) (NGR) (Tab. S9). We excluded GFAP from the ratio as it was highly abundant in later time points. Parameters were calculated for each CO, and median values were determined at each time point (TP). The acceptance cut-off was +/-35 % of the median (MED). Values within the 35% upper/lower limit of the median received a score of 0. Values outside the 35% interval received a score of 1 for PML/LML and 2 for the NGR: PML/LML =1; NGR =2 if MED(x levels at TP) ± 0.35*MED(x levels at TP) where x = PML, GM or NGR Total score = Score(PML) + Score(LML) + Score(NGR) We assigned a higher score to NGR as a critical parameter of cellular differentiation and compositional changes in CO, whereas PML and LML parameters identify fewer specific discrepancies in the differentiation efficiency. For instance, the glia-to-neuron ratio is used to count human brain cells 27 . We summed up each parameter's score, and COs with a total score >2 were removed as outliers (Tab. S9). We excluded an abnormal CO at each time point ( Fig. S16 and Tab. S9), reducing the coefficient of variation among biological replicates considerably (Tab. S8).

Major lipid characterization
The lipid extract was analyzed using a 1290 Infinity II UHPLC (Agilent) system coupled with the 6469 Triple Quadrupole mass spectrometer (Agilent). 1 µL of lipid extract was injected on the reverse phase microbore column (CSH, 1 mm *100 mm, 1.7µm, Waters), separated at 100µl/min flow rate over 15 min. For the gradient elution, mobile phase A was 10 mM ammonium formate in acetonitrile: water (60:40), and mobile phase B was 10 mM ammonium formate in Isopropanol: acetonitrile (90:10). The gradient elution program was : 0 min 15 % B, 1.86 min 30% B , 2.32 min 48%, 9.5 min 82% B, 12.5 to 13.5 min 99% B and 13.5 to 15 min column re-equilibration. The positive mode jet stream source parameters were gas temp 200 0C, gas flow 14 l/min, nebulizer pressure 45 psi, sheath gas temp 400 0C, and sheath gas flow 8 l/min capillary voltage 4 kV, nozzle voltage 500v, and unit resolution for Q1 and Q3. Data were acquired in positive mode using the dynamic multiple reaction monitoring (MRM) mode, 2 min retention time window for each transition. Raw data files were processed using Mass Hunter Quantitative analysis (B.07.00, Agilent Technologies) software. Mass spectrometric fragmentation behavior of different lipid classes is well established. For the multiple reaction monitoring (MRM) based on targeted lipid identification1-3, a list NeXtProt, prioritizing peptides with experimental data in PeptideAtlas. The top 4-5 transitions were selected to generate the SRM library applied to the QC sample for tentative peptide identification using the RT prediction model. Peptides with expected SRM signature and RT were utilized further for protein assays. Heavy-labeled synthetic peptide internal standards and a scheduled SRM acquisition mode were used for relative protein quantitation. Quantifier transitions were selected based on the low %CV of L/H ratios between technical replicates. AA = amino acids, QC = quality control pooled sample, RT = retention time, L = light peptide, H = heavy labeled peptide, CV = coefficient of variation Figure S3. Mass spectra for the 13C labeled GM3 and GM1 synthesized in-house Figure S4. Chromatograms for the (a) gangliosides analysed in the positive and negative mode along with the labeled standards (b) protein analysed with appropriate light peptide underneath the heavy peptide a S13 b S14 Figure S5. Trypsin digestion efficiency after 2, 4, and 16 hours of incubation S15 Figure S6. Analytical parameters of proteomic protocol. (a) Solid-phase extraction method (SPE) recovery of standard peptides was on average 87 % (n=2-4). (b) The standard peptide response in the cerebral organoid matrix was, on average, 32% lower relative to the standard peptide response in the neat solvent. (c) The optimal sample injection amount was equivalent to 6 µg of total protein onto the LC column for maximal sensitivity of the protein assay (n=2). On average, the response of synthetic peptide internal standards was reduced by 24 % for the 6 µg total protein equivalent injection, relative to the neat solvent. (d) Normalized concentrations of target proteins at three sample dilution levels, CV between the three calculated concentrations was, on average, 15 %. samples. Figure S11. The time-trends of NSCs markers (SOX2, GD3); neuron-specific markers (DCX, NEFL, TUBB3, MAP2, SYN1, GD1a, GD1b, and GT1b); RGCs marker (FABP7); and astrocyte-specific markers (S100B, GFAP, and GM2). Concentrations <LOD are highlighted in red (•), >LOD and <LOQ in grey (•), zero values with empty points (o). Protein marker and GS levels were normalized to house-keeping protein ACTB.
S22 Figure S12. Housekeeping proteins ACTB and GAPDH represent the total cell mass during cerebral organoid proliferation S23 Figure S13.   Figure S14. The time-trends of major lipids as observed in the cerebral organoids. The box plots are the sum of all lipid-species from the same class. The peak area sum has been normalized to to house-keeping protein ACTB.
S25 Figure S15. The cluster analysis of proteins and gangliosides during early and late neurogenesis in cerebral organoids. (a) The early neurogenesis (D48, D76, D95; n=9) is represented by principal clusters (Ia, Ib, II). Cluster Ia. associated neuronal and astrocytic gangliosides, neuronal protein traits (except for MAP2), and RGCs markers. Cluster Ib. associated astrocytic protein markers and MAP2. Cluster II. grouped peripheral gangliosides and NSCs protein markers (SOX2, GM3, GD3). (b) Five distinct clusters (IIIa; IIIb1; IIIb2; IVa; IVb) represent the late neurogenesis (D110, D135, D160; n=9). Cluster IIIa associates immature neuronal markers, cluster IIIb1 associates GM3 and TTR, and cluster IIIb2 associates mature neuronal markers with an RGCs marker (FABP7). Cluster IVa associates mature astrocytic markers GFAP and CD44 and cluster IVb astrocytic S100B, MAP2, SOX2, and astrocytic gangliosides (i.e., GM2, GD2). a b S26 Figure S16. Variability scores the sum of lipid markers levels, the sum of protein markers levels, and neurons to glia ratio of individually profiled cerebral organoids as listed in Table S9 Supplementary tables    Scoring parameters were: i) neurons to glia ratio (NGR) = the sum of DCX, TUBB3, MAP2, and NEFM to the sum of FABP7 and S100B; ii) sum of a11 protein markers and housekeeping proteins ACTB and GAPDH levels except for SYN1 (PML), and the sum of eight GSs levels (LML). The upper and lower cut-off range was determined using median value +/-35 %. COs with NGR, PML, or LML values below or above the cut-off range were marked (in red). NGR parameter outside the median +/-35 % received the score of two, PML or LML outside the range received the score one (Table S10). The highest scoring COs were excessively variable and eliminated from further interpretation.

S44
Table S10. A score to assess potential sources of heterogeneity in cerebral organoids

Sorting Score
Interpretation 0 Similar COs 1 Varying differentiation efficiency 2 Varying differentiation efficiency or cellular composition 3,4 Varying differentiation efficiency and cellular composition S45 Table S11. The list of correlation coefficients and p-values for protein and lipid markers across (a) early (D48, D75, D96) and (b) later (D110, D135, D160) time points.
Significant differences have been highlighted.