Temporal Proteomic Profiling of SH-SY5Y Differentiation with Retinoic Acid Using FAIMS and Real-Time Searching

Abstract

The SH-SY5Y cell line is often used as a surrogate for neurons in cell-based studies. This cell line is frequently differentiated with all-trans retinoic acid (ATRA) over a 7-day period, which confers neuron-like properties to the cells. However, no analysis of proteome remodeling has followed the progress of this transition. Here, we quantitatively profiled over 9400 proteins across a 7-day treatment with retinoic acid using state-of-the-art mass spectrometry-based proteomics technologies, including FAIMS, real-time database searching, and TMTpro16 sample multiplexing. Gene ontology analysis revealed that categories with the highest increases in protein abundance were related to the plasma membrane/extracellular space. To showcase our data set, we surveyed the protein abundance profiles linked to neurofilament bundle assembly, neuron projections, and neuronal cell body formation. These proteins exhibited increases in abundance level, yet we observed multiple patterns among the queried proteins. The data presented represent a rich resource for investigating temporal protein abundance changes in SH-SY5Y cells differentiated with retinoic acid. Moreover, the sample preparation and data acquisition strategies used here can be readily applied to any analogous cell line differentiation analysis.

Graphical Abstract

INTRODUCTION

Transformed neuron-like cell lines are often used to overcome the limitations of primary mammalian neurons, which can no longer be propagated once differentiated into mature neurons. As such, cell culture models are often used to study drug neurotoxicity and neurodegenerative diseases, for example, Parkinson’s or Alzheimer’s disease.^1-4 The human-derived cell line SH-SY5Y is commonly used as an in vitro model for human neurons.⁵ This cell line, which originated from a bone marrow biopsy of a patient with neuroblastoma, can be induced to differentiate into cells with a neuron-like phenotype. As such, SH-SY5Y is generally used as a model of neuronal function and differentiation in neuroscience studies as the cell line can reproduce many biochemical and morphological properties of neurons.⁶ Specifically, SH-SY5Y cells have been used to investigate neurotoxicity, neuroprotection, neuroblastoma tumorigenesis, neuronal differentiation and metabolism, as well as other aspects of neuroadaptive and neurodegenerative processes.^4,7

Retinoic acid (all-trans-retinoic acid, ATRA) is a well-established inducer of a neuronal phenotype in SH-SY5Y cells.^8,9 Retinoic acid is a vitamin A derivative with wide-ranging cellular differentiation-promoting properties.¹⁰ This metabolite’s method of action involves binding to a DNA-bound receptor, thereby affecting transcription factors, which may induce or repress the expression of nearby genes.¹¹ As such, retinoic acid induces profound changes in the cellular proteome. SH-SY5Y cells differentiate primarily to a cholinergic neuron phenotype in response to at least 7 days of retinoic acid treatment, typically at a concentration of 10 μM.¹² Specifically, neuronal differentiation involves the formation and extension of neuronal processes, induction of neuron-specific enzymes, and production of neurotransmitters.^13-15 Studies frequently use either undifferentiated^16-19 and retinoic acid-differentiated^20-23 SH-SY5Y cells. Some studies have compared undifferentiated and differentiated SH-SY5Y cells, but they highlight only specific proteins and pathways.⁸ Until now, the comprehensive analysis of proteome remodeling during differentiation of SH-SY5Y has not yet been assessed.

Sample multiplexing strategies, such as isobaric tags for relative and absolute quantitation (iTRAQ) and tandem mass tags (TMT), have a wide range of benefits for profiling protein abundance at the proteome level.^24,25 Isobaric tagging allows multiple samples to be analyzed simultaneously, thereby reducing costs and instrument time, permitting multiple comparisons in one experiment, and producing fewer missing values within a sample set. Sample multiplexing strategies are also valuable for time-course investigations.^26,27 Here, we applied such an approach to investigate SH-SY5Y cell differentiation.

We used a TMTpro-based workflow to quantitatively profile the differentiation of SH-SY5Y cells following treatment with retinoic acid. First, we performed an “end-point” experiment comparing SH-SY5Y cells following 7 days of retinoic acid treatment versus mock treatment in a TMT6-plex format and observed hundreds of significant protein alterations. Next, we expanded our study using TMTpro16 reagents²⁸ that enabled the temporal abundance profiling of over 9400 proteins in SH-SY5Y cells with single-day resolution during a 7-day retinoic acid treatment. We also highlighted proteins related to neurofilament bundle assembly, neuron projections, and neuronal cell body formation. These data, which were collected on a Orbitrap Eclipse mass spectrometer with FAIMS and real-time database searching (RTS), represent the largest temporal protein abundance profiling data set for retinoic acid differentiation in SH-SY5Y cells to date. With this data set, we provide a rich resource that is amenable to further data mining to study individual proteins, pathways, and interactions thereof related to retinoic acid-induced SH-SY5Y differentiation.

EXPERIMENTAL SECTION

Materials

Tandem mass tag (TMT) isobaric reagents were from Thermo Fisher Scientific (Waltham, MA). All-trans-retinoic acid (R2625) was purchased from Sigma (St. Louis, MO). Water and organic solvents were from J.T. Baker (Center Valley, PA). Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) was from Life Technologies (Waltham, MA). Trypsin was purchased from Pierce Biotechnology (Rockford, IL) and LysC from Wako Chemicals (Richmond, VA). Unless otherwise noted, all other chemicals were from Pierce Biotechnology (Rockford, IL). SH-SY5Y cell line (CRL-2266) was from the American Type Culture Collection (Manassas, VA).

Cell Growth and Harvesting

Methods of cell propagation followed techniques used previously.^29,30 In brief, upon ~90% confluency, we aspirated the growth media and washed the cells three times with ice-cold phosphate-buffered saline. For our end-point experiment, a single dish of cells was propagated into six dishes at low density. Three dishes were treated with 10 μM retinoic acid and three without for 7 days. Cells were washed and fresh media was added daily. For our time-course experiment, two separate dishes of cells were each divided into eight dishes. Cells were either harvested on the appropriate day or spent media was aspirated and fresh media was added. All experiments with retinoic acid were conducted in subdued light conditions. We noted neurite outgrowth, as expected, as SH-SY5Y cells exhibited typical neuronal morphology in the presence of retinoic acid after several days.⁸ At the appropriate time point, the cells were dislodged with a nonenzymatic reagent, harvested by trituration following the addition of 10 mL of phosphate-buffered saline, pelleted by centrifugation at 3000g for 5 min at 4 °C, and the supernatant was removed. Five hundred microliters of 200 mM EPPS, 8 M urea, pH 8.5 supplemented with 1× Pierce Protease Inhibitors were added directly to each pellet.

Cell Lysis and Protein Digestion

Cells were homogenized by 12 passes through a 21-gauge (1.25 inches long) needle and incubated at 4 °C with gentle agitation for 30 min. The homogenate was sedimented at 21 000g for 5 min. Protein concentrations were determined from the supernatant using the bicinchoninic acid (BCA) assay. Proteins were subjected to disulfide bond reduction with 5 mM tris (2-carboxyethyl) phosphine (15 min, room temperature) and alkylation with 10 mM iodoacetamide (20 min in the dark, room temperature) followed by quenching with 10 mM dithiotreitol (15 min in the dark, room temperature). Trichloroacetic acid (TCA) precipitation was performed prior to protease digestion. Samples, resuspended in 200 mM EPPS, pH 8.5, were digested for 14 h at room temperature with LysC (100:1 protein-to-protease ratio). Trypsin was then added also at a 100:1 protein-to-protease ratio and the samples were incubated at 37 °C for 6 h.

Tandem Mass Tag Labeling

For the end-point experiment, 100 μg (in 5 μL of acetonitrile) of TMT reagent was added to the peptides (50 μg) along with 15 μL of acetonitrile for a final acetonitrile concentration of ~30% (v/v). After incubating for 1 h at room temperature, we quenched the reaction with hydroxylamine to a final concentration of 0.3%. The TMT-labeled samples were pooled at a 1:1 ratio across the samples. For the time-course experiment, 120 μg (in 6 μL of acetonitrile) of TMTpro reagent was added to the peptides (50 μg) with 14 μL of acetonitrile to achieve a final concentration of ~30% (v/v). Following incubation for 1 h at room temperature, the reaction was quenched with hydroxylamine to a final concentration of 0.3%. TMTpro-labeled samples were pooled at a 1:1 ratio across all 16 samples. For each experiment, the pooled sample was vacuum centrifuged and subjected to C18 solid-phase extraction (Sep-Pak, Waters).

Off-Line Basic pH Reversed-Phase (BPRP) Fractionation

We fractionated the pooled, labeled peptide sample using BPRP HPLC.³¹ We used an Agilent 1200 pump equipped with a degasser and 220 and 280 nm detectors. Peptides were subjected to a 50 min linear gradient from 5 to 35% acetonitrile in 10 mM ammonium bicarbonate pH 8 at a flow rate of 0.6 mL/min over an Agilent 300Extend C18 column (3.5 μm particles, 4.6 mm ID, and 250 mm in length). The peptide mixture was fractionated into a total of 96 fractions, which were consolidated into 24 superfractions.³² Samples were subsequently acidified with 1% formic acid and vacuum centrifuged to near dryness. Each consolidated fraction was desalted via StageTip, dried again via vacuum centrifugation, and reconstituted in 5% acetonitrile, 5% formic acid for LC-MS/MS processing.

Liquid Chromatography and Tandem Mass Spectrometry

Mass spectrometric data were collected on an Orbitrap Lumos (end-point experiment) or Eclipse (time-course experiment) mass spectrometer coupled to a Proxeon NanoLC-1200 UHPLC. The 100 μm capillary column was packed with 35 cm of Accucore 150 resin (2.6 μm, 150 Å; Thermo Fisher Scientific). The scan sequence began with an MS1 spectrum (Orbitrap analysis, resolution 120 000, 350–1400 Th, automatic gain control (AGC) target 5 × 10⁵, maximum injection time 100 ms). Data were acquired for 90 min per fraction. SPS-MS3 analysis was used to reduce ion interference.^33,34 The MS2 stage analysis consisted of collision-induced dissociation (CID), quadrupole ion trap analysis, automatic gain control (AGC) 2 × 10⁴, normalized collision energy (NCE) 35, q-value 0.25, maximum injection time 120 ms, and isolation window at 0.7 Th. Following the acquisition of each MS2 spectrum, we collected an MS3 spectrum in which multiple MS2 fragment ions were captured in the MS3 precursor population using isolation waveforms with multiple frequency notches. MS3 precursors were fragmented by HCD and analyzed using the Orbitrap (NCE 65, AGC 1.5 × 10⁵, maximum injection time 150 ms, resolution was 50 000). The NCE at the MS3 stage for TMTpro samples was set at 45, as per previous recommendation.²⁸ For TMTpro16 data, we used the real-time search option using a human database and we limited MS3 scans to two peptides per protein per fraction.³⁵ For data acquisition including FAIMS, the dispersion voltage (DV) was set at 5000 V, the compensation voltages (CVs) were set at −40, −60, and −80 V, and TopSpeed parameter was set at 1 s per CV.

Data Analysis

Spectra were converted to mzXML via MSconvert.³⁶ Database searching included all entries from the Human UniProt Database (downloaded: August 2019) that was concatenated with that database with all protein sequences in the reversed order. Searches were performed using a 50 ppm precursor ion tolerance and a 0.9 Da product ion tolerance in conjunction with Comet searches and linear discriminant analysis.^37,38 TMT tags on lysine residues and peptide N-termini (+229.163 Da for TMT and + 304.207 Da for TMTpro) and carbamidomethylation of cysteine residues (+57.021 Da) were set as static modifications, while oxidation of methionine residues (+15.995 Da) was set as variable. Peptide-spectrum matches (PSMs) were adjusted to a 1% false discovery rate (FDR).^39,40 PSM filtering was performed using a linear discriminant analysis (LDA), as described previously³⁸ and then assembled further to a protein-level FDR of 1%.⁴⁰ Proteins were quantified by summing reporter ion counts across all PSMs, also as described previously.⁴¹ Reporter ion intensities were adjusted to correct for the isotopic impurities of the different TMT reagents according to manufacturer specifications. The signal-to-noise (S/N) measurements of peptides assigned to each protein were summed and normalized such that the sum of the signal for all proteins in each channel was equivalent and so we accounted for equal protein loading. Finally, each abundance measurement was scaled, such that the summed signal-to-noise for that protein across all channels equaled 100, generating a relative abundance (RA) measurement. Data analysis and visualization were performed in R or Microsoft Excel. We used Bioconductor⁴² and the time-course package,⁴³ which applies a multivariate empirical Bayes model to rank temporally changing proteins and calculate the Hotelling T² statistic.

Data Access

RAW files will be made available upon request. The data have been deposited to the ProteomeXchange Consortium via the PRIDE⁴⁴ partner repository with the data set identifier PXD020807. In the Supporting Information, we have included tables listing protein names, gene symbols, and TMT quantitation values for both the end-point (Supplemental Table 1) and time-course (Supplemental Table 2) data sets. We have also included lists of peptides, associated protein names, gene symbols, TMT quantification values, retention time, and isolation purity for the end-point (Supplemental Table 3) and time-course (Supplemental Table 4) data sets.

RESULTS AND DISCUSSION

Isobaric Tag-Based Proteomics Permitted the Relative Protein Abundance Profiling of SH-SY5Y Cell Line Differentiation Induced by Retinoic Acid Treatment

We investigated proteome remodeling in the SH-SY5Y cell line resulting from retinoic acid treatment using an isobaric tagging strategy. Our experimental design consisted of treating SH-SY5Y cells with 10 μM of retinoic acid for a 7-day period (Figure 1A). The samples were processed for mass spectrometry analysis using a modified SL-TMT protocol⁴⁵ (Figure 1B). Cells were harvested and syringe-lysed in 8 M urea. Cysteines were reduced and alkylated, and proteins were TCA precipitated and digested with LysC followed by trypsin. We used TCA for precipitation as protein concentration was <1 mg/mL, which limits the use of chloroform–methanol precipitation, as is the typical precipitation strategy in the SL-TMT protocol.⁴⁶ Peptides for each condition were labeled with TMT, pooled, fractionated by basic pH reversed-phase chromatography, and analyzed by SPS-MS3. In the initial experiment (the “end-point” experiment), we compared three biological replicates each of SH-SY5Y cells treated for 7 days with retinoic acid and mock-treated cells as a TMT6-plex experiment (Figure 1C). After observing considerable changes in the proteome, we designed a time-course experiment using TMTpro16 reagents.²⁸ TMTpro16 facilitated the arrangement of eight time points (Days 0–7) in biological duplicate.

Figure 1.

Experimental overview. (A) SH-SY5Y has been shown to differentiate when treated with all-trans-retinoic acid (ATRA) over a 7-day period and to acquire neuron-like properties. (B) Experimental workflow. Samples were processed using the SL-TMT protocol (45). (C) Layout of TMT6-plex end-point experiment. Triplicates of untreated and 7-day retinoic acid-treated SH-SY5Y cells were arranged in a 3 × 2 TMT6-plex experiment. (D) Layout of TMTpro16-plex time-course. SH-SY5Y cells treated with retinoic acid were harvested over the course of 7 days. Biological replicate samples from Day 0 to Day 7 were arranged in a TMTpro16-plex. ATRA, all-trans retinoic acid; BPRP, basic pH reversed-phase; TCA PPT, trichloroacetic acid precipitation.

We quantified a total of 8575 proteins for the TMT6-plex end-point experiment, representing one of the largest isobaric-tagged SH-SY5Y proteomes.¹⁶ These proteins were inferred from 73 827 peptides of which 60 135 were unique in sequence. These data were collected on a FAIMSpro-equipped^47,48 Orbitrap Lumos mass spectrometer using SPS-MS3. We acquired the time-course TMTpro16 experiment on the latest-generation Orbitrap Eclipse mass spectrometer equipped with a FAIMSpro interface and using real-time searching (RTS) to decide on-the-fly whether or not to collect an MS3 scan for a given MS2 scan.^35,49 This analysis enabled the quantification of 9418 proteins. These proteins were inferred from 93,380 peptides of which 64 578 were unique in sequence. This data set marked the largest number of SH-SY5Y-derived proteins quantified to date. These data are tabulated in Table 1.

Table 1.

Data Set Overview

				dysregulateda
	total peptides	unique peptides	proteins	up	down
end point	73 827	60 135	8575	399	191
time course	93 380	64 578	9418	n/a	n/a

^aFold change > ∣2∣ and Benjamini–Hochberg corrected p-value < 0.01, not applicable.

Several Hundred Proteins Were Differentially Expressed When SH-SY5Y Cells Were Mock-Treated or Treated with Retinoic Acid for 7 Days

We investigated further the abundance profiles of the 8575 proteins quantitated in our end-point experiment. First, we performed hierarchical clustering incorporating Euclidean distance and Ward linkage using the TMT relative abundance values (Figure 2A). We noted very tight clustering among biological replicates, with a clear split between treated and untreated cells. We next performed principal components analysis (PCA) in efforts to determine how much variation could be explained by a given variable. Agreeing with the tight hierarchical clustering, PCA showed a distinct split between treated and untreated cells (Figure 2B). We determined that 98.7% of the variance in the experiment can be explained by the first principal component (PC1), which divides the samples by treatment. These data showed that retinoic acid clearly has a significant impact on proteome remodeling of SH-SY5Y cells. We illustrated these data in two dimensions using a volcano plot. Here, we plotted the Benjamini–Hochberg-corrected p-value versus the log 2 ratio of the average TMT reporter ion signal-to-noise value of the retinoic acid-treated sample (ATRA) to that of the mock-treated control (mock) (Figure 2C). We set stringent significance thresholds consisting of a Benjamini–Hochberg-corrected p-value < 0.01 and an absolute fold change of 2 or greater. These thresholds determined 191 and 399 proteins to be statistically significant in terms of being of lower or higher abundance, respectively, when the cells were treated with retinoic acid.

Figure 2.

Proteome profiling of SH-SY5Y cells with and without a 7-day retinoic acid treatment. (A) Heat map illustrates the relative abundance of the 8,575 proteins quantified in this experiment. Each protein abundance value sums to 100 across the six channels. (B) Principal components analysis (PCA) of the six samples depicts the variance explained by the first two principal components (PC). (C) Volcano plot illustrating the Benjamini–Hochberg (BH)-corrected p-value of retinoic acid (ATRA) vs. mock control and the fold change for these values. Highlighted on the volcano plot are six of the most altered proteins resulting from retinoic acid treatment. These proteins include: (D) CYP26B1, cytochrome P450 26B1; (E) NCAM2, neural cell adhesion molecule 2; (F) ELFN1, extracellular leucine-rich repeat and fibronectin type III.(G) RPH3A, rabphilin-3A; (H) RTL1, retrotransposon-like protein 1; and (I) TIMELESS.

We proceeded to subject these short lists of dysregulated proteins to Gene Ontology (GO) enrichment analysis using the DAVID interface⁵⁰ with the entire list of quantified proteins as background. We required 10 proteins per category and an FDR < 0.05. Proteins having higher relative abundance when cells were treated with retinoic acid were categorized mainly as cell membrane, cell surface, cell migration, or related proteins (Table 2 top). On the contrary, proteins with significantly lower abundance when cells were treated with retinoic acid were categorized mainly as cell division, cell proliferation, or DNA-interaction-related proteins (Table 2 bottom). The names of proteins in each category listed in Table 2 are included in Supplemental Table 5. We noted that these classifications did not discriminate whether the function of the protein was permissive or inhibitory, only that a given protein had some role in that gene ontology classification. Nonetheless, a wide range of biological processes and molecular functions were affected by retinoic acid treatment.

Table 2.

Gene Ontology Classifications for Differentially Abundant Proteins

abundance	category	term	count	%	FDR
up	BP	cell adhesion	26	6.6	4.46 × 10−2
up	BP	extracellular matrix organization	22	5.6	4.46 × 10−6
up	BP	cell migration	15	3.8	4.22 × 10−2
up	BP	interferon-γ-mediated signaling	10	2.5	3.86 × 10−2
up	CC	extracellular exosome	142	36.0	4.37 × 10−22
up	CC	plasma membrane	134	34.1	5.39 × 10−5
up	CC	extracellular space	70	17.8	5.36 × 10−9
up	CC	focal adhesion	50	12.7	5.15 × 10−21
up	CC	Cell surface	47	11.9	9.10 × 10−13
up	CC	Cell–cell junction	21	5.3	2.52 × 10−2
up	CC	extracellular matrix	20	5.1	2.47 × 10−2
up	CC	actin cytoskeleton	19	4.8	1.24 × 10−3
up	CC	membrane raft	16	4.1	4.48 × 10−2
down	BP	Cell division	39	20.9	7.07 × 10−25
down	BP	DNA replication	31	16.7	1.09 × 10−26
down	BP	DNA repair	23	12.4	6.41 × 10−12
down	BP	Cell proliferation	17	9.1	2.18 × 10−3
down	BP	DNA replication initiation	15	8.1	6.14 × 10−17
down	CC	nucleus	118	63.4	4.65 × 10−20
down	CC	Cytoplasm	78	41.9	3.38 × 10−2
down	CC	nuclear chromatin	14	7.5	7.70 × 10−5
down	MF	protein binding	133	71.5	2.30 × 10−8
down	MF	DNA binding	41	22.0	2.97 × 10−4
down	MF	ATP binding	37	19.9	1.16 × 10−3
down	MF	Chromatin binding	23	12.4	1.09 × 10−7

^aThreshold for including a given GO category: >10 proteins per category and a false discovery rate (FDR) ≤ 0.05. BP, biological pathway; CC, cellular component; MF, molecular function.

We also subjected the lists to Kyoto Encyclopedia of Genes and Genomes (KEGG)⁵¹ pathway analysis using the STRING⁵² interface. These results were in line with those determined from the GO enrichment analyses. Specifically, proteins that increased in abundance when cells were treated with retinoic acid were observed in several pathways, including membrane protein-related extracellular matrix (ECM)– receptor interaction and cholesterol metabolism, as well as retinol metabolism (Table 3 top). Meanwhile, proteins with significantly lower abundance when cells were treated with retinoic acid were in pathways reflecting the GO enrichment analysis, specifically, DNA-related pathways and the cell cycle (Table 3 bottom). The names of proteins in each pathway in Table 3 are included in Supplemental Table 6. These data reinforced that a wide variety of cellular pathways were affected by retinoic acid treatment.

Table 3.

KEGG Pathway Classifications for Differentially Abundant Proteins

abundance	category	pathway	count	%	FDR
up	hsa04512	ECM–receptor interaction	11	13.6	1.60 × 10−4
up	hsa04979	cholesterol metabolism	5	10.4	3.63 × 10−2
up	hsa05410	hypertrophic cardiomyopathy	8	10.1	6.10 × 10−3
up	hsa00830	retinol metabolism	6	10.0	2.49 × 10−2
down	hsa03030	DNA replication	16	44.4	9.75 × 10−19
down	hsa03430	mismatch repair	5	21.7	1.20 × 10−4
down	hsa03410	base excision repair	5	15.2	4.80 × 10−4
down	hsa04110	cell cycle	16	13.0	9.05 × 10−12
down	hsa00240	pyrimidine metabolism	11	11.0	2.20 × 10−7
down	hsa03420	nucleotide excision repair	5	10.9	1.80 × 10−3
down	hsa03440	homologous recombination	4	10.0	8.50 × 10−3

^aThreshold for including a given KEGG pathway category: Number of proteins were ≥10% of the background gene count per category and a false discovery rate (FDR) ≤ 0.05.

We highlighted several proteins that showed elevated changes in abundance in the TMT6-plex data set. We noted three examples of upregulated proteins that were involved in the metabolism of retinoic acid or have well-studied functions in neurons. First, CYP26B1 (cytochrome P450 26B1) is involved in the metabolism of retinoic acid (RA), rendering it inactive through oxidation⁵³ (Figure 2D). Next, NCAM2 (neural cell adhesion molecule 2) plays a prominent role in the projection of axons⁵⁴ (Figure 2E). Finally, ELFN1 (extracellular leucine-rich repeat and fibronectin type III domain containing 1) is a postsynaptic protein that modulates the temporal dynamics of interneuron recruitment⁵⁵ (Figure 2F). Moreover, we also highlighted three examples of downregulated proteins that have roles in protein transport and DNA processing. First, RPH3A (rabphilin-3A) plays a crucial role in docking and fusion steps of regulated exocytosis and is implicated in intracellular protein transport⁵⁶ (Figure 2G). Next, RTL1 (retrotransposon-like protein 1) is a retrotransposon that reverts RNA into DNA through the process of reverse transcription using an RNA transposition intermediate⁵⁷ (Figure 2H). Finally, TIMELESS (a human homolog of the identically named Drosophila protein) has a role in controlling DNA replication, DNA repair, maintenance of genome stability, and in the regulation of the circadian clock⁵⁸ (Figure 2I). We will interrogate these proteins again in the quantitative time-course data set.

Differences in the Relative Protein Abundance Profiles of SH-SY5Y Cells Were Traced Over a 7-Day Treatment of Retinoic Acid

After noting striking proteome remodeling in the end-point assay, we examined the protein profile alterations over the course of retinoic acid treatment. We measured the quantitative temporal proteome profile of 9418 proteins across eight time points in duplicate. As with the end-point assay, we performed hierarchical clustering with Euclidean distance and Ward linkage using TMT relative abundance values (Figure 3A). We noted very tight clustering among biological duplicates with adjacent time points grouped closely together, as expected. We also observed a major split between Day 3 and 4 time points, potentially indicating that major proteome remodeling occurred between these 2 days. We again performed principal components analysis (PCA) to explore how much variation can be explained by a certain variable (Figure 3B). PCA showed tight clustering between duplicates, as in the hierarchical clustering heat map. PC1 explained 71.2% of the variance in the experiment, which was associated with length of treatment. More specifically, we can observe that the pattern along the PC1 axis was chronological, that is, the leftmost data points were Day 0 and the days were sequential from left to right until Day 7, which was the rightmost point on the plot. These data supported that retinoic acid clearly had a significant impact on proteome remodeling of SH-SY5Y cells and that we can track the increase or decrease in protein abundance over time.

Figure 3.

Temporal differences in protein expression. (A) Heat map illustrates the relative abundance of the 9418 proteins quantified in this time-course experiment. The abundance values for each protein sum to 100 across the 16 channels. (B) Principal components analysis (PCA) of the 16 samples depicts the variance explained by the first two principal components (PC). Temporal expression patterns of proteins highlighted in Figure 2: (C) CYP26B1, cytochrome P450 26B1; (D) NCAM2, neural cell adhesion molecule 2; (E) ELFN1, extracellular leucine-rich repeat and fibronectin type III, (F) RPH3A, rabphilin-3A; (G) RTL1, retrotransposon-like protein 1; and (H) TIMELESS.

We revisited the proteins highlighted in Figure 2D-I to determine the pattern of protein abundance alteration over the 7-day retinoic acid treatment. We focused on three proteins that exhibited lower abundance with retinoic acid treatment. First, CYP26B1 showed a sharp increase over 2 days of treatment and then leveled off (Figure 3C). Similarly, NCAM2 exhibited a gradual increase after Day 2 that plateaued at Day 4 (Figure 3D). In contrast, ELFN1 showed a gradual increase over the first 6 days of treatment and dropped slightly on the final day of treatment (Figure 3E). Likewise, we again highlighted three examples of downregulated proteins. First, RPH3A displayed a decrease in abundance of about 2-fold over the first 2 days of retinoic acid treatment before leveling off (Figure 3F). RTL1 exhibited a gradual decrease over the first 6 days of treatment (Figure 3G). Finally, the abundance of TIMELESS began to decrease on Day 2 and then decreased more than 50% by Day 4 (Figure 3H). We noted that in this limited sampling, different patterns of protein abundance alterations emerged.

Next, we assessed the statistical significance of our temporally changing proteins. We applied the time-course package in R Bioconductor, which used a multivariate empirical Bayes model to rank temporally altered proteins.⁴³ More precisely, we employed the mb.long function to calculate a Hotelling T² statistic and rank proteins according to their temporal abundance alterations.⁵⁹ The mb.long function is used to compute multivariate empirical Bayes Statistics for longitudinal replicated time-course data. Specifically, the mb.long function computes the T² statistic of differential expression for longitudinal replicated time-course data allowing for the ranking of genes or proteins.⁶⁰ We have included the Hotelling T² statistic for each protein in this data set as a column in Supplemental Table 3. We next compared the GO classifications determined from the end-point assay to that of the time course. We graphically illustrated the significant temporally altered proteins by plotting the log 10 of the Hotelling T² statistic versus the ratio of the Day 7 to the Day 0 time point (Supplemental Figure 1A). We set stringent significance thresholds consisting of an absolute fold change of 2 or greater and a Hotelling T² statistic of 322.836, which encompasses about 30% of the data.²⁷ As such, the number of significantly changing proteins approached the number in the end-point data set. Here, 202 and 345 proteins decreased or increased temporally, respectively, during the 7-day retinoic acid treatment.

We again subjected these short lists of proteins to GO enrichment analysis using the DAVID interface,⁵⁰ with the entire list of quantified proteins as background. As with the previous analysis, we required 10 proteins per category and an FDR < 0.05. As expected, the analysis revealed categories on par with those highlighted in the end-point data set (Table 2). As in the previous data set, proteins of higher abundance in treated cells were categorized as cell migration, cell surface, cell junction, or related proteins (Supplemental Figure 1B). On the contrary, proteins with significantly lower abundance in treated cells were categorized mainly as cell proliferation, cell division, or DNA-related proteins (Supplemental Figure 1C). Although both data sets produced similar results, the time-course data set permitted us to determine when the largest shift in abundance for a given protein change occurred to single-day resolution.

Neuron-Related Proteins Demonstrated a Time-Dependent Increase in Their Relative Abundance Following Retinoic Acid Treatment

We showcased our time-course data set further to examine alterations in proteins with roles in neuronal development. As we mentioned previously, retinoic acid-differentiated SH-SY5Y cells are commonly used in neuroscience research, as they acquire more neuron-like properties after treatment. We used the time-course data to extract proteins with functions in neurofilament bundle assembly, neuron projections, and neuronal cell body formation. The temporal abundance profiles of these 21 proteins are illustrated in Figure 4A-U.

Figure 4.

Temporal protein profiles of neuron-related proteins. (A–U) Proteins associated with neurons, particularly neurite outgrowth, are highlighted. These proteins include: (A) Annexin A3, ANXA3; (B) PDZ and LIM domain protein 5, PDLIM5; (C) NAD(P)H dehydrogenase [quinone] 1, NQO1; (D) Metalloproteinase inhibitor 2, TIMP2; (E) Secretagogin, SCGN; (F) BDNF/NT-3 growth factors receptor, NTRK2; (G) Integrin alpha1, ITGA1; (H) Apolipoprotein E, APOE; (I) α-actinin-4, ACTN4; (J) Receptor-type tyrosine-protein phosphatase-like N, PTPRN; (K) Midkine, MDK; (L) Homer protein, HOMER2; (M) Plasma membrane calcium-transporting ATPase 4, ATP2B4; (N) Ankyrin-3, ANK3; (O) Neurofilament light polypeptide, NEFL; (P) Copper-transporting ATPase 1, ATP7A; (Q) Synaptotagmin-5, SYT5; (R) Proto-oncogene tyrosine-protein kinase receptor, RET; (S) Neurexin-1, NRXN1; (T) SH3 domain-containing kinase-binding protein 1, SH3KBP1; and (U) BAG family molecular chaperone regulator 3, BAG3.

We observed several different patterns of increasing abundance of certain proteins over the 7-day treatment with retinoic acid. Several proteins exhibited a gradual increase in abundance, specifically annexin A3 (Figure 4A), PDZ and LIM domain protein 5 (Figure 4B), NAD(P)H dehydrogenase [quinone] 1 (Figure 4C), α-actinin-4 (Figure 4D), Metalloproteinase inhibitor 2 (Figure 4E), Secretagogin (Figure 4F), BDNF/NT-3 growth factors receptor (Figure 4G), integrin alpha1 (Figure 4H), and apolipoprotein E (Figure 4I). In contrast, the abundance of some proteins increased on the second day of retinoic acid treatment, including receptor-type tyrosine-protein phosphatase-like N (Figure 4J), midkine (Figure 4K), Homer protein (Figure 4L), plasma membrane calcium-transporting ATPase 4 (Figure 4M), ankyrin-3 (Figure 4N), and neurofilament light polypeptide (Figure 4O). In addition, two proteins increased in abundance gradually and plateaued after 4 days, specifically, copper transporting ATPase 1 (Figure 4P) and synaptotagmin-5 (Figure 4Q). Two other proteins increased rapidly in abundance and reached their maximum after approximately 2 days, specifically, proto-oncogene tyrosine-protein kinase receptor (Figure 4R) and neurexin-1 (Figure 4S). Finally, SH3 domain-containing kinase-binding protein 1 (Figure 4T) and BAG family molecular chaperone regulator 3 (Figure 4U) decreased slightly in abundance during the first day of treatment with retinoic acid but peaked at Day 6. Although most of these proteins showed a similar pattern, no consensus increase was apparent, which supports that complex time-dependent mechanisms were regulating the retinoic acid-mediated differentiation of SH-SY5Y cells.

Several transcriptomic profiling studies have also investigated the differentiation of SH-SY5Y cells with retinoic acid. Here, we compare our data to two such reports. First, Pezzini, et al. performed RNA-seq analysis on retinoic acid-treated SH-SY5Y cells differentiation with subsequent growth in neurobasal medium.⁶¹ Although the differentiation conditions differed from our workflow, similar conclusions were drawn as dysregulated proteins were implicated in remodeling of the plasma membrane and the cytoskeleton, also had implications in neuritogenesis, which agreed with our gene ontology data (Table 2) and Figure 4. In fact, 146 of the 399 proteins that we deemed as significantly higher in abundance upon retinoic acid treatment were also significantly upregulated in Pezzini, et al. Likewise, 85 of the 191 proteins that we determined to be significantly lower in abundance upon retinoic acid treatment were significantly downregulated in that publication as well. Second, Korecka, et al. investigated retinoic acid differentiation of SH-SY5Y cells via a genome-wide transcriptional analysis.⁶² Cell growth conditions were slightly different from ours as cells were propagated on Matrigel for that study. Nonetheless, like Pezzini, et al. and our own work, dysregulated proteins included those associated with proliferation, neuronal differentiation, and development. Moreover, 91 of our 399 proteins with significantly higher abundance upon retinoic acid treatment were also significantly upregulated and 24 of our 191 proteins with significantly lower in abundance were significantly downregulated in Korecka, et al. These researchers highlighted the neural cell adhesion molecule, NCAM2, which is highly involved in neuronal compartmentalization, which was 12-fold upregulated by Day 8. We similarly observed an almost 10-fold increase in the abundance of NCAM2 after 7 days of retinoic acid treatment (Figure 3D). Although discrepancies among studies are expected due to differences in treatment lengths, media, and growth conditions, our findings specifically those related to plasma membrane proteome remodeling, neuron projections, and neuronal cell body formation were in agreement.

CONCLUSIONS

We quantitatively profiled over 9400 proteins across the 7-day retinoic acid treatment using TMTpro16 reagents and a FAIMS-equipped Orbitrap Eclipse mass spectrometer with real-time database searching. This data set represents a rich resource for the study of temporal protein abundance, which can be used for further data mining for deeper studies into retinoic acid differentiation in SH-SY5Y cells. We highlighted proteins involved in neuron projections; however, many other biological insights can be extracted from these data. Moreover, the sample preparation and data acquisition strategies used herein can be applied to virtually any biological system with a sequenced genome. Although retinoic acid is the most common method to induce differentiation in SH-SY5Y cells, several other compounds are also known to differentiate this cell line. For instance, SH-SY5Y cells may be driven toward a mature dopaminergic phenotype by retinoic acid treatment with coadministration of phorbol esters, such as 12-O-tetradecanoyl-phorbol-13 acetate (TPA).^7,63 In addition, exposure of SH-SY5Y cells to dibutyryl cyclic AMP (dbcAMP) results in neurite extension.⁶⁴ Staurosporine, a pan-kinase inhibitor, can trigger neurite formation and cell cycle arrest in SH-SY5Y cells.⁶⁵ Similarly, culturing SH-SY5Y cells in neurobasal medium with the addition of B27 has resulted in enhanced differentiation.⁶⁶ Other compounds including treatment with cholesterol,⁶⁷ vitamin D,⁶⁸ and insulin⁶⁹ have been shown to promote neuronal differentiation in SH-SY5Y cells. In summary, the quantitative proteomics techniques outlined herein can be used to temporally profile the alterations in protein abundance for any mechanism of cellular differentiation, including, but not limited to, those mentioned above.