Quantification of mRNA Decay Rates in HeLa and SH-SY5Y Cell Lines Reveals Novel Properties of Membrane Protein Coding Transcripts

Abstract

Posttranscriptional regulation is crucial for siRNA design, as decay rates in cell lines influence perceived siRNA potency. This study profiles transcripts with ‘fast’ and ‘slow’ half-lives in HeLa and SH-SY5Y cells, commonly used in drug discovery. We calculated half-lives for 1,815 HeLa and 5,376 SH-SY5Y transcripts, finding comparable half-lives between cell lines, though HeLa cells generally had longer half-lives. Comparing mRNA and protein half-lives, ‘fast’ decay transcripts encoded proteins with shorter half-lives, while ‘slow’ decay transcripts encoded stable proteins. We linked mRNA decay rates to siRNA activity by comparing HeLa data to a previous siRNA screen, discovering that faster decay transcripts had lower knockdown. Surprisingly, stable transcripts, more amenable to knockdown, were over-represented by membrane protein-coding transcripts. Despite their stability, these transcripts had low-to-moderate expression, regardless of miRNA regulation. We explored cis- and trans- features affecting mRNA stability and expression, suggesting that low RNA binding protein (RBP) binding, combined with specific stabilizing RBP regulation, contributes to the stability of these membrane protein-coding transcripts. This study highlights the importance of understanding transcript features, mRNA decay and its potential impact on siRNA efficacy, particularly for transcripts encoding membrane proteins.

Keywords

mRNA decay membrane proteins protein turnover therapeutic siRNA

Introduction

mRNA decay is a tightly regulated process that controls gene expression in combination with mRNA synthesis. Understanding mRNA kinetics and features contributing to steady-state expression is crucial for delineating mechanisms underlying siRNA knockdown efficacy. Improved knowledge of the mRNA lifecycle aids in designing efficient siRNA therapeutics and matching RNA therapeutic modalities to targets.

Studies in mammalian cells and Caenorhabditis elegans show varying siRNA knockdown levels across transcripts.^1,2 While literature exists on mRNA characteristics impacting siRNA knockdown (e.g., secondary structure, RNA binding protein motifs, preexisting miRNA binding sites), these studies are case-specific, and many mRNA biology measures for targets remain unknown. Determining the impact of each feature on mRNA expression is challenging; hence, studies use target mRNA decay as a key property to compare with siRNA knockdown. Experimental evidence and modeling suggest that transcripts with faster decay rates are harder to knockdown.^1,2 This is crucial when selecting targets for siRNA-mediated knockdown as a therapeutic approach.

About 20% of drug targets targeted by small molecules and biologics are cell surface proteins.³ So far, genes targeted by oligonucleotide therapeutics have been historically undruggable; these genes encode a near-even mix of membrane and intracellular proteins implicated in disease pathways.^4–6 While several cell surface proteins have been characterized extensively at a protein level, transcript level assessments remain underexplored. Kawata et al., determined synthesis and decay rates in HeLa cells and have shown that mRNA turnover rates relate to the biological functions of specific classes of genes.⁷ These observations highlight the need to profile decay rates of transcripts in cell lines used for discovery research to design therapeutic siRNA screening paradigms and define characteristics of transcripts encoding membrane proteins from the perspective of mRNA decay. mRNA decay can be assessed in vitro by transcriptional shutoff or metabolic labeling of nascent mRNA. Transcriptional shutoff methods, like actinomycin D, inhibit translation, leading to pleiotropic effects and potentially biased RNA decay measures.⁸ Metabolic labeling methods, such as SLAM-sequencing (SLAM-seq), use nucleoside analogs (e.g., 5′-bromouridine, 4-thiouridine [s⁴U], 5-ethynyl uridine) to label transcripts during synthesis; decay rates of these labeled transcripts are measured by pulse-chase or approach-to-equilibrium techniques.^8–11

To address these questions, we conducted SLAM-seq in HeLa and SH-SY5Y cells, extensively used in early drug discovery for siRNA library screening. While several HeLa cell decay rate datasets exist,^12,13 no such compendium is available for SH-SY5Y cells, and most datasets focus on rapid turnover transcripts. In this study, we generate an SH-SY5Y mRNA decay rate dataset, compare HeLa and SH-SY5Y mRNA decay rates to protein decay rates and we identify some novel properties of membrane protein coding transcripts.

Methods

Cell culture

HeLa cells were cultured in DMEM media (Gibco #11885084) containing 1% penicillin-streptomycin and 10% Heat Inactivated Fetal Bovine Serum (HI-FBS). SH-SY5Y cells were cultured in DMEM/F-12 media (Corning #10–090-CV) containing 1% sodium pyruvate, 1% sodium bicarbonate, 1% nonessential amino acids, and 10% HI-FBS. Cells were grown to approximately 70% confluency before being passaged.

Cell viability titration

Lexogen’s Cell Viability Titration Module (Lexogen #059.24) was used to determine the optimal s⁴U concentration(s) that limit cell toxicity in HeLa and SH-SY5Y cells. 5k cells were plated per well in a 96-well opaque plate in a volume of 200 µL culturing medium the day before the labeling experiment. A 10-point, 1:2 serial dilution of s⁴U starting at 4 µM was prepared and added to cells 24 h after plating. After 24 h of s⁴U labeling, CellTiter-Glo Luminescent Cell Viability Assay (Promega # G7570) was used to assess cell viability using a BioTek Synergy Neo Multi-Mode Reader. The IC₁₀ for each cell line was determined using GraphPad Prism Version 10.2.3.

SLAM sequencing

SLAM sequencing setup was followed according to manufacturer’s instructions in SLAM-seq Kinetics Kit–Catabolic Kinetics Module (Lexogen #062.24). 125k HeLa and 350k SH-SY5Y cells were plated per well in 6-well plates. Cells were cultured for 24 h before s⁴U-containing media was added to HeLa and SH-SY5Y cells at 200 µM and 350 µM, respectively. Fresh s⁴U-containing media was added to cells every 3 h for 24 h, at which media was exchanged for 100x excess uridine-containing media. Cells were collected in TRIzol (Invitrogen #15596018) at 0-, 1-, 6-, and 24-hour timepoints following the media exchange. RNA was extracted from TRIzol using the Purelink RNA Mini Kit (ThermoFisher #12–183-018A). After total RNA isolation, RNA was treated with iodoacetamide to alkylate the s⁴U-containing transcripts. Sample quality control and quantification was performed by Lexogen followed by QuantSeq FWD library preparation for Illumina sequencing. QuantSeq 3′ mRNA-sequencing (forward-SR100) was performed on an Illumina NextSeq P2, and data were processed using the SLAM-DUNK pipeline (workflow outline in Fig. 1A).¹⁵

FIG. 1.

Global transcript half-lives are consistent between different cell lines. (A) SLAM-sequencing experimental workflow. Cells were pulsed with fresh s⁴U-containing media every 3 h for 24 h, after which media containing excess uridine was added to cells. Samples were collected at 0-, 1-, 6-, and 24-h post-chase and RNA was subsequently isolated and treated with iodoacetamide. QuantSeq 3′ mRNA-sequencing and decay analysis was performed by Lexogen. (B) Spearman correlation for experimentally determined SH-SY5Y and HeLa decay rates (1/t_0.5). Boxplot insert shows log₂-transformed half-lives for HeLa and SH-SY5Y cells. (C) Decay curve for EEF2 in HeLa and SH-SY5Y cells. (D, E) Spearman correlation for experimentally determined HeLa (D) and SH-SY5Y (E) half-life ranks and published ranks from Agarwal & Kelley decay compendium.¹³ (F) Boxplot of whole cell protein degradation rates in HeLa cells¹⁴ for transcripts with ‘fast’ or ‘slow’ mRNA decay rates in HeLa cells. (G) Boxplot of whole cell protein degradation rates in SH-SY5Y cells (OmicScouts) for transcripts with ‘fast’ or ‘slow’ mRNA decay rates in SH-SY5Y cells.

Decay rate analysis

Size factors were calculated for read counts output from the SLAMdunk pipeline using the median absolute deviation method implemented by DESeq2. Counts for T > C conversions were normalized using the size factors for all reads to account for differences in sequencing depth. Two types of models were created based on T > C counts. One set of models were created for UTRs that had ≥ 1 T > C conversion across all time points and replicates. A second set of models was created to save potential transcripts with few T > C counts at the 24-hour time-point. The second set of models required ≥ 1T > C at all time points except the 24-hour time-point. Genes were included in downstream analysis if all three untreated replicates had a CPM (Counts Per Million) ≥2. The half-life for each 3′ UTR was modeled using a Poisson regression model:

\ln (c_{t}) = β_{0} + β_{1} t

(1)

where

c_{t}

is the normalized T > C read counts for the 3′ UTR at time

t

β_{0}

is the intercept, and

β_{1}

is the slope. Half-life models were implemented using the R glm function. Half-lives are calculated for each 3′ UTR model under the assumption that we are observing a first-order reaction using the equation:

t_{0.5} = - \frac{\ln (2)}{β_{1}}

(2)

where

\ln (2)

is the exponential decay constant and

β_{1}

is the reaction rate constant (slope of the Poisson regression model). Transcripts with 0 <= t_0.5 <= 48 h were collapsed into genes by taking the average half-life if a gene had more than one transcript included. Going forward, the term “decay rate” will refer to 1/t_0.5. Our dataset and the data provided from the decay compendium¹³ were combined based on gene name to compare the decay rates between the cell lines and published data. Transcripts were categorized into slow and fast bins based on if their half-life was greater or less than 2.8 and 2 h for HeLa and SH-SY5Y cells, respectively. These values were determined by plotting the cumulative distribution function (CDF) of the half-lives and selecting the point where CDF >0.1 (dashed red line, Fig. 2C).

FIG. 2.

High-decay RNA transcripts are resistant to siRNA repression in HeLa cells. (A) Venn diagram displaying the number of transcripts included in internal and external datasets. (B) Pearson correlation for HeLa decay rate ranking as determined by SLAM-seq and historical data. (C) Cumulative distribution of half-lives as determined by SLAM-seq in HeLa cells. Any transcript with a half-life less than 2.8 h was labeled as fast for subsequent analysis. (D) Log₂-transformed expression plotted against decay rate (1/t_0.5) as determined by SLAM-sequencing. Silencer siRNA screening data from Larsson et al. is integrated. Transcripts in red were historically targeted by siRNAs resulting in over 90% knockdown, while transcripts in blue were targeted by siRNAs with less than 90% knockdown. Transcripts in gray were not featured in the Larsson et al. dataset.¹

Protein degradation rate analyses

SH-SY5Y cell pellets were lysed using lysis buffer consisting of 5% SDS, 50 mM TEAB (pH 7.5). Total protein was determined using a BCA assay (Thermo Fisher Scientific) according to the manufacturer’s instructions. Proteins were further digested into peptides using the R2P1 approach¹⁶ with 1:25 trypsin:protein at 37°C overnight. Peptides were desalted by C18 solid-phase extraction (using the Bravo AssayMap platform with C18 cartridges) and dried in vacuo. TMT11 labeling was performed according to Zecha et al.¹⁷ where there were equal amounts combined per TMT plex and desalted by C18 solid-phase extraction cartridges (Sep-Pak, Waters, 50 mg cartridges). Desalted peptides were chromatographically separated via basic reversed phase chromatography into 24 fractions, dried down, and stored at −20°C until further analysis.¹⁸ The samples were measured by nLC-MS/MS system (NanoLC: UltiMate3000; MS: Orbitrap Eclipse; both Thermo Fisher Scientific) with the following settings: 120 min nLC gradient, data-dependent acquisition, fragmentation method: HCD. Peptides and proteins were identified and quantified using MaxQuant (Version: 2.0.3.0), with the following settings: protease: Trypsin; fixed modifications: Carbamidomethyl (C); variable modifications: Oxidation (M); Acetyl (Protein N-term); Arg10 (R); Lys8 (K); protein sequence database: Human SwissProt Database; PSM/Protein False Discovery Rate (FDR): 1%. Protein degradation rate calculations were conducted based on methods described in Zecha et al.¹⁸

siRNA knockdown in SH-SY5Y cells

Briefly, 10k SH-SY5Y cells were reverse transfected in OptiMEM media (Gibco # 31985062) using Silencer® siRNAs in duplicates for 48 h. Cells were harvested and gene expression was measured using the TaqMan™ Cells-to-Ct™ Express kit (ThermoFisher Scientific #A57985). siRNA assay IDs and TaqMan/IDT gene expression assay IDs provided in Supplementary Table S4.

Downstream membrane protein coding transcript analysis

Transcript annotation

Expression values were categorized into low (<64 HeLa; <32 SH-SY5Y), medium (64–512 HeLa; 32–256 SH-SY5Y), and high (>64 HeLa; >256 SH-SY5Y) buckets based on average CPM (Fig. 3A, B). Known membrane proteins were annotated using SurfaceGenie dataset which assigns a SPC (surface protein consensus) score to proteins based on annotation evidence in up to 4 different membrane and surface protein datasets.¹⁹ Slow and low SH-SY5Y SPC 3′ and 5′ UTRs were compared to non-SPC transcripts using MEME Suite 5.5.7.²⁰

FIG. 3.

Transcript half-lives are influenced by function and miRNA targeting. (A, B) Scatterplot of expression and decay rate (1/t_0.5) for HeLa (A) and SH-SY5Y (B) transcripts. SPC-coding transcripts are colored red, and non-SPC-coding transcripts are colored gray. The percentage of bucket based on expression (low, moderate, high) and decay (fast and slow) out of all transcripts is listed. (C, E) Bar chart of top-expressed miRNAs in HeLa (C) and SH-SY5Y (E) cell lines. Left Y-axis is average reads per million and right Y-axis is cumulative percent of reads. (D, F) CDF plots of HeLa (D) and SH-SY5Y (F) transcript half-lives. Colored dotted line: transcript encodes surface membrane protein and is targeted by miRNAs, black dotted line: transcript does not encode surface membrane protein and is targeted by miRNAs, colored solid line: transcript encodes surface membrane protein and is not targeted by miRNAs, black solid line: transcript does not encode surface membrane protein and is not targeted by miRNAs. SPC, surface protein consensus.

miRNA analysis

Reads Per Million (RPM) normalized miRNA expression profiles for SH-SY5Y cells were downloaded from the miRmine database of miRNA-seq experiments.²¹ Replicates were averaged together, and miRNA were filtered for those with ≥2 reads per million. Read depth normalized HeLa miRNA profiles were downloaded from ArrayExpress accession E-MTAB-9690.²² miRNA that made up the top 90% of expressed miRNA were considered for downstream analysis. Predicted miRNA target genes were curated from miRDB (version 6) and filtered for a target score cutoff ≥90 based on authors suggestions. ECDF plots and D statistics were calculated using R stats and ggplot2 packages. The P values reported are from a one-sided Kolmogorov–Smirnov test.

RNA binding protein site analysis

Enriched in vitro RNA binding k-mers for 78 known RNA binding proteins (RBPs) were downloaded from a publicly available dataset. Nostrand et al. have generated one of the most comprehensive maps of RBPs through five assays: enhanced CLIP, Bind-n-Seq, DNA association by ChIP-seq, functional assessment by knockdown RNA-seq, and subcellular localization by immunofluorescence.²³ The RBP binding sites were converted to a FASTA format of DNA sequences and were then aligned to Ensembl (v112) canonical mRNA transcripts of the genes of interest.^24,25 The alignment was performed with Bowtie v1.3.1²⁶ allowing multiple perfect sense-stranded mapping. The frequency of occurrences and coverage in nucleotides of all mapped RBP motifs are calculated for each transcript. 1,908 HeLa transcripts and 5,577 SH-SY5Y transcripts were included in the analysis. Pearson’s correlation coefficients (R) and P value between the frequency of occurrences, coverage in nucleotides, and mRNA decay rates across all transcripts were estimated for HeLa and SH-SY5Y cell lines, respectively.

Motif analysis

5′ and 3′ UTRs were downloaded from Ensembl v113 for all HeLa and SH-SY5Y transcripts with slow decay rates. These were divided into SPC and non-SPC categories and uploaded to MEME Suite (5.5.7).²⁰ Using the Simple Enrichment Analysis (SEA) tool, slow SPC UTRs were compared to slow non-SPC UTRs as a control. The homo sapiens DNA-encoded RNA motif database was used to test for enrichment using default settings.

3′UTR length analysis

Human 3′UTR lengths were extracted from Supplementary Table S2 of Fansler et al.²⁷ and filtered to include 3′UTRs whose lengths had clear values and were not blacklisted for exclusion. As there are multiple 3′UTRs per gene, the highest-expressed 3′UTR was used for analysis based on the fraction of reads from the gene across all cell types. 3′UTRs that did not account for any fraction of reads were excluded, and for genes whose 3′UTRs were equally expressed across all cell types, the length was averaged. Welch two sample t-tests were used to determine significant differences between groups.

Results

SLAM-Seq determines global transcriptome decay rates in HeLa and SH-SY5Y cells

SLAM-seq utilizes s⁴U to metabolically label newly synthesized RNA transcripts. Metabolic labeling of RNA enables precise measurement of mRNA synthesis and decay rates. To determine the decay rates of transcripts expressed in HeLa and SH-SY5Y cell lines, we utilized SLAM-seq, a well-established method which has been previously used for identifying mRNA kinetics in primary cells such as mouse embryonic stem cells¹⁰ and in immortalized cell lines such as K562.²⁸

To ensure transcriptome-wide incorporation of s⁴U in both cell lines, we pulsed cells at fixed concentrations for 24 h with frequent media changes (Fig. 1A, details in Methods). As s⁴U can be toxic at high concentrations, a cell viability titration experiment was conducted for both HeLa and SH-SY5Y cells to identify the s⁴U concentration that allowed for 90% cell viability. 200 µM and 350 µM s⁴U were determined to be the optimal concentrations for HeLa and SH-SY5Y cells, respectively (Supplementary Fig. S1A, B). Following the s⁴U pulse, we substituted the s⁴U-containing media with media containing excess uridine. Post-chase, cells were harvested at 0-, 1-, 6-, and 24-h to sample transcripts with both short and long half-lives. RNA was extracted for quantification and decay rate estimation post sequencing. Using the untreated samples, we were able to identify 1,815 transcripts in HeLa cells and 5,376 transcripts in SH-SY5Y cells following data cleaning. T > C conversions were measured using the SLAM-DUNK analysis pipeline, and we utilized a Poisson regression model to model decay rates using reads with T > C conversions (representing nascent transcripts) at all timepoints. To capture analysis of fast decay transcripts where nascent transcripts are not detected at 24 h (i.e., no reads with T > C conversions were detected), we modeled decay rates only using the 0-, 1-, and 6-hour timepoints. When we compare the half-lives of all genes across the two cell lines, we find a statistically significant correlation (P < 2.2e-16, Pearson’s R = 0.69) (Fig. 1B). We find for some genes that are highly expressed in both cell lines, such as EEF2 (a translation elongation factor), half-lives can be significantly different (Fig. 1C). Despite having relatively similar CPM values of 2,170 (HeLa) and 1,970 (SH-SY5Y), the half-life ranges from 9.38 to 23.74 h, highlighting that specific cell lines may employ unique posttranscriptional mechanisms to regulate gene expression.

Next, we wanted to compare the calculated half-lives to published datasets. In 2022, a compendium was published that summarized and normalized transcriptome-wide mRNA decay rates from 39 human datasets consisting of 19 unique cell lines.¹³ To this end, we rank ordered all HeLa and SH-SY5Y transcripts based on their half-lives and compared it to the rank ordered transcripts from the published compendium (Fig. 1D, E). Briefly, across our SLAM-seq datasets and the compendium, transcripts were ranked such that stable transcripts with long half-lives have low numerical values. We find that transcriptome-wide estimation of half-lives by SLAM-seq in both HeLa and SH-SY5Y cell lines correlates significantly with transcriptome-wide estimation of stability across datasets compiled by Aggarwal et al. (HeLa: P < 2.2e-16, Spearman’s R = 0.78, SH-SY5Y: P < 2.2e-16, Spearman’s R = 0.73).

In addition to mRNA decay rates, another key aspect of gene expression is protein decay rates. So far, the relationship between these two aspects of gene expression remains unexplored. To address this, we obtained protein decay rate SILAC datasets for HeLa¹⁴ cells where Boisvert et al. evaluated 5 timepoints to determine decay rate of nuclear and cytoplasmic proteins and for SH-SY5Y cells (licensed from OmicScouts). We classified HeLa and SH-SY5Y SLAM-seq datasets into ‘fast’ (<2.83 h HeLa and <2 h SH-SY5Y) and ‘slow’ (>2.83 h HeLa and >2 h SH-SY5Y) transcripts. We then looked at the distribution of protein decay rates for these two broad classes and we find that overall ‘fast’ decaying transcripts encode proteins with faster protein degradation rates as well (Fig. 1F, G), while ‘slow’ decay transcripts encode proteins with longer protein degradation rates (HeLa P value = 0.09; SH-SY5Y P value = 0.003). This observation suggests dual levels of regulation of gene expression potentially based on the intended biological function.

SLAM-Seq confirms findings that fast decay RNA transcripts are resistant to siRNA repression in HeLa cells

Intrinsic properties of mRNA are important to consider for RNA interference (RNAi) activity, especially with the advancement of siRNA therapeutics. Both HeLa and SH-SY5Y cells are routinely used for screening siRNA libraries in early drug discovery. We aimed to investigate the connectivity between mRNA kinetics and observed siRNA efficacy. Previous studies in HeLa cells have demonstrated that mRNAs with fast decay rates are resistant to efficacious siRNA knockdown.¹ However, in this study, RNA expression was determined using microarray technology and decay rates were determined by transcriptional shutoff (actinomycin D treatment) for multiple timepoints between 0 and 2 h.^1,29 Since 2010, several advancements have been made in metabolic labeling of mRNAs and multiple studies have demonstrated that novel metabolic labeling methods are more robust than previous actinomycin D treatment protocols.^10,12 Additionally, these methods prevent the pleiotropic effects that actinomycin D can exert on a biological system.⁸

HeLa half-lives as determined by SLAM-sequencing were compared to previously published decay rates in HeLa cells.^1,29 Between the two datasets, 148 transcripts were shared (Fig. 2A). 1,667 transcripts were unique to the SLAM-seq datasets, while 439 transcripts were to Larsson et al.’s microarray dataset. Differences in transcript profiles could have arisen due to differences in detection methods (3′UTR sequencing vs. microarray), limited transcript probes on the Affymetrix expression array, and timepoint selection bias. While time points between the two studies were different, for the transcripts that were shared between the datasets, turnover rate rankings were correlated (P < 1.6e-13, Spearman’s R = 0.56) (Fig. 2B).

Based on the distribution of half-life values in HeLa cells, 51 transcripts with half-lives less than 2.8 h were qualified as “fast decay transcripts” (Fig. 2C). When plotted against expression and overlaid with previously published siRNA repression data in HeLa cells, the data corroborated findings that transcripts with short half-lives (fast decay) are more resistant to siRNA repression (Fig. 2D). Out of the siRNAs tested, those that had 90% or greater mRNA knockdown, targeted transcripts whose half-lives were greater than 2.8 h (Fig. 2D). We find that the overall correlation between mRNA decay rates and expression is negative (Pearson correlation r_p = −0.32, P < 2.2e-16) and the fast-decay transcripts, as expected based on literature, trend towards medium-low expression. Importantly, even the highest expressed “fast decay” transcript in our dataset (log2(Avg CPM) ∼ 9) appears to be resistant to high repression. To extend our observations to the SH-SY5Y cell line, we selected 2 unstable transcripts (MAT2A and SLC38A2) as well as 4 stable transcripts (KDELR1, ELOVL1, ELOVL5 and FABP5) and conducted a transfection based knockdown experiment for 48 h. We find that at 10 nM, overall repression of all transcripts is lower than that reported for HeLa cells by Larsson et al¹ at an unknown but presumably higher concentration (Supplementary Fig. S2, Table 1). Despite differences between the SH-SY5Y and HeLa dataset, in SH-SY5Y cells MAT2A and SLC38A2 show lower repression (>50%) while ELOVL1, KDELR1 and FABP5 show higher repression for both or at least 1 of the siRNAs tested (>50%). ELOVL5 shows lower repression (<50%) for both tested siRNAs, its half-life is close to the cutoff of slow vs fast and it contains a long 3′UTR (1723 nucleotides), a feature often associated with faster turnover and thus low repression. In summary, mRNA decay has been shown to be a factor in influencing siRNA efficacy, with the probability of greater knockdown occurring with transcripts that tend to be more stable.

Table 1.

Table of Representative Unstable (Gray Rows) and Stable (White Rows) Describing Expression and Decay Rate in HeLa and SH-SY5Y Cells as Well as 3′UTR Length, HeLa Repression Data from Larrson et al.¹ Repression Data Generated in SH-SY5Y Cells

Gene name	t_0.5 HeLa	t_0.5 SH-SY5Y	Decay rate classification	log2AvgCPM HeLa	log2AvgCPM SH-SY5Y	Repression in HeLa at unknown conc. [#siRNAs]	Repression in SH-SY5Y at 10nM [#siRNAs]	SPC	3′UTR length
MAT2A	2.21	1.61	Fast	7.58	7.43	<95% [2/2]	<50% [2/2]	N/A	1509
SLC38A2	2.13	2.43	Fast in HeLa and near the cutoff in SH-SY5Y	8.79	6.86	<95% [2/2]	<50% [2/2]	3	2923
ELOVL1	5.32	7.56	Slow	6.53	4.67	>95% [2/3]	>50% [2/2]	1	515
ELOVL5	4.95	4.74	Slow	7.51	7.17	>95% [2/3]	>50% [0/2]	2	1723
KDELR1	15.84	9.94	Slow	8.87	8.37	>95% [2/2]	>50% [2/2]	1	718
FABP5	13.75	7.39	Slow	6.22	50.18	>95% [2/2]	>50% [1/2]	N/A	208

SPC, surface protein consensus.

Transcript function and targetability by miRNAs impact half-life

Transcript decay and synthesis rates have been previously shown to differ for functional classes of genes.⁷ Here, we examined the half-lives for transcripts that code for membrane proteins against (SPC: Surface Protein Consensus as defined by SurfaceGenie; details in Methods section) all other transcripts (non-SPC). Transcripts in HeLa and SH-SY5Y cells were categorized according to SurfaceGenie annotations to determine which transcripts coded for membrane proteins.¹⁹ Expression levels were plotted against decay rates (1/t_0.5) for HeLa (Fig. 3A) and SH-SY5Y (Fig. 3B) transcripts with membrane protein coding transcripts displayed in fuchsia. Less than 3% of the membrane protein coding transcripts had fast decay rates (<2.83 h HeLa and <2 h SH-SY5Y, Fig. 3A, B). Additionally, 80–90% of the membrane protein coding transcripts in both cell lines had low-moderate expression with slow decay rates.

We hypothesized that membrane protein coding transcripts were stable irrespective of being regulated by miRNAs. To this end, we analyzed transcripts that are known to be targets of endogenous miRNA pathways. miRNA expression profiles for HeLa and SH-SY5Y cells were collected from ArrayExpress and miRmine databases, respectively.^21,22 For both cell lines, two miRNAs make up the majority of miRNA reads (Supplementary Fig. S3A, B). In HeLa cells, miR-21-5p and let-7i-5p account for about 70% of all miRNA expression, while in SH-SY5Y cells, miR-10b-5p and miR-191-5p make up about 50% of expression (Figs. 3C, 4E). To determine if transcripts canonically targeted by known miRNAs have differential half-life values, we analyzed their cumulative distribution in both cell lines. We categorized the cumulative distributions based on two variables: if the transcript coded for a membrane protein (SPC or non-SPC) and if the transcript is a known target of endogenous miRNAs in the cell (miRNA or non-miRNA). For both cell lines, there is a statistically significant difference between the half-lives of membrane protein coding transcripts and all other transcripts regardless of miRNAs regulation (Fig. 3D, F). Specifically, transcripts that code for membrane proteins have longer half-lives than those that do not. As expected, since miRNAs function to induce mRNA decay, transcripts targeted by miRNAs have shorter half-lives irrespective of the type of protein they encode. Taken together, these results prompted us to explore the causes of mRNA regulation of membrane protein coding transcripts further.

FIG. 4.

RNA binding proteins influence the stability of surface protein coding transcripts. (A, B) Density contour plots for HeLa (A) and SH-SY5Y (B) SPC (magenta in HeLa and green in SH-SY5Y) and non-SPC (coral in both HeLa and SH-SY5Y) transcripts based on the frequency of RBP binding sites in transcripts. The contour plots are a representation of a scatter plot (each dot being a transcript) where the shape of the scatter is emphasized and density of the dots is shown with a blue-white gradient. In this plot, the blue hue is highest in the center of the contour where most dots/transcripts are concentrated, and the shape of the contours show the relationship between half-life and RBP frequency. (C, D) Boxplots of log₂-transformed 3′UTR length for HeLa (C) and SH-SY5Y (D) based on decay rate and SPC/non-SPC categorization. (E) Top RBP motifs enriched in slow SPC 3′and 5′ UTRs compared to slow non-SPC transcripts. RBP, RNA binding protein.

Multiple cis- and trans- features could potentially influence stability of surface protein coding transcripts

To explore the reasons for which membrane protein coding transcripts have longer half-lives, we considered three metrics: (1) overall RBP coverage, (2) 3′UTR length, and (3) binding of specific “stabilizing” RBPs to the UTRs. While some RBPs such as HuR play protective roles in preventing transcript degradation,³⁰ others such as Lsm proteins are known to promote RNA turnover.³¹ Additionally, we considered 3′UTR length which is believed to be a destabilizing feature^32,33 as well as specific motifs in the UTRs that recruit RBPs to promote mRNA stability.³⁴ It is plausible that unique features in the UTRs of membrane protein coding transcripts contributed to their enhanced stability and longer half-lives.

To order to delineate the role of RBPs in driving stability of surface protein coding transcripts, we first identified all RBPs expressed in the respective cell lines for which the binding sites have been identified by eCLIP-seq and Bind-N-Seq in the literature.²³ We then calculated the frequency and density (“coverage”) of RBP binding sites on membrane protein coding transcripts (SPC+ in Fig. 4A, B) and all other transcripts (SPC- in Fig. 4A, B). We found that in HeLa cells, the correlation between transcript half-life and average occurrence of transcript RBP binding sites was significantly decreased for membrane protein coding transcripts; in SH-SY5Y cells, the correlation is higher than in HeLa but still negative correlation remained statistically significant in both cell lines (Figure 4A, B and Supplementary Figure S3A, B).

To explore if mRNA length contributes to enhanced stability of membrane protein coding transcripts, we examined 3′UTR lengths of all transcripts (Fig. 4C, D). Briefly, we utilized the recently published 3′UTRome²⁷ to evaluate the relationship across 3′UTR length, decay rates and subcellular location of the encoded protein. As expected, between fast and slow decay transcripts, fast decay transcripts had significantly longer 3′UTRs (P < 0.01 for HeLa and P = 6.494e-08 for SH-SY5Y represented by coral boxplots in Fig. 4C, D). Surprisingly, between all SPC and non-SPC transcripts, membrane protein coding transcripts had longer 3′UTRs, with mean lengths of 1,222bp (SPC) and 877 bp (non-SPC) for HeLa and 1,441bp (SPC) and 1,179bp (non-SPC) for SH-SY5Y, respectively. These differences were statistically significant, with P = 7.798e-13 for HeLa and P = 4.691e-13 for SH-SY5Y. This finding suggests that the stability of membrane protein coding transcripts is likely regulated by proteins binding the mRNAs at specific motifs.

Additionally, we wanted to identify potential RBP binding motifs associated with stable membrane protein coding transcripts. To examine which sequence features could be differentiating membrane protein coding transcripts from all other transcripts, we performed de novo motif searching. 5′UTRs and 3′UTRs for membrane protein coding transcripts with slow half-lives were used for motif analysis using MEME Suite’s Simple Enrichment Analysis function. Input sequences were compared against all other transcripts that also had slow half-lives. Top motifs for HeLa (purple) transcripts are displayed in Figure 4E. In HeLa cells, both 5′UTR and 3′UTR motifs are thought to bind HuR and FMR1 respectively, both proteins have been associated with mRNA stability.^35,36 In case, of SH-SY5Y cells, the motifs identified in both 3′ and 5′UTR are quite similar and associated primarily with a splicing factor RBM5. Further bioinformatic analyses and experimental evaluation will be required to identify factors influencing stability of membrane protein coding transcripts in SH-SY5Y cells.

Discussion

This study utilized SLAM-sequencing to measure mRNA decay rates in HeLa (cervical cancer) and SH-SY5Y (neuroblastoma) cell lines. By metabolically labeling newly synthesized RNA with s⁴U, we tracked the incorporation and subsequent decay of RNA molecules over time. The results indicated a significant correlation between the half-lives of transcripts in both cell lines (P < 2.2e-16, Pearson’s R = 0.69) even though median half-lives in HeLa are higher than that in SH-SY5Y (Fig. 1B). As expected, the decay rate rankings for all transcripts in both cell lines correlated well with the rankings in the published decay compendium¹³ (Fig. 1D, E). While other datasets exist where metabolic labeling of mRNA in HeLa cells was conducted,^7,37 these studies focused on identifying transcripts with shorter half-lives. Our study is the first to identify not just transcripts with short half-lives (<2 h) but also transcripts with longer half-lives (12–24 h) and further compare both mRNA and protein half-lives (Fig. 1). This study will aid in design and interpretation of experiments testing activity of RNA therapeutics in SH-SY5Y cells, which are especially relevant for neurodegenerative and pain disorders.

Immortalized cell lines are often used for screening libraries of RNA therapeutics. An important factor affecting the efficacy of RNA therapeutics is the turnover rate of the target mRNA. There is experimental evidence in Larsson et al. to show that siRNA knockdown is poor for transcripts with high decay rate as measured by actinomycin D treatment. We built upon this dataset by comparing siRNA knockdown published by Larsson et al. to decay rates and steady-state expression calculated in our study. We find that the observations made by Larsson et al. hold true even when knockdown data is compared to decay rates calculated by the current standard of metabolic labeling (Fig. 2D). We further extended these observations to SH-SY5Y cells for select stable and unstable transcripts (Supplementary Fig. S2 and Table 1). Taken together, these observations align with recent studies where gene-specific considerations of RNAi in C. elegans were modeled and miRNA induced decay of reporters expressed at different levels was tested to show that at steady-state, transcripts with higher turnover and/or low expression will be more challenging to knockdown.^2,38 Our study adds to the growing body of evidence around target transcript features being a key consideration for assessing siRNA mediated knockdown.³⁹

Why do membrane protein coding transcripts have only low-moderate expression despite their long half-lives? This study was designed to identify transcripts with both fast (<1 hour) and slow (between 6–24 h) decay rates. Surprisingly, we find that transcripts that encode membrane proteins are overrepresented among the classes of “slow decay/low expression” and “slow decay/moderate expression” transcripts (Fig. 3A, B). First, we find that within this class of transcripts, half-life remains longer with or without miRNA targeting when compared to all other transcripts (Fig. 3C–F). We then looked at known RBP occupancy and found that decay rates in both cell lines are negatively correlated with frequency of RBP sites on the transcript, suggesting, as expected, that transcripts with faster turnover are highly decorated with RBPs (Fig. 4A, B). Again, we observe a decrease in the negative correlation, suggesting less decoration by RBPs, for transcripts encoding membrane proteins compared to those that do not, although this decrease is more prominent in HeLa than in SH-SY5Y cells. To elaborate on mRNA features that could account for enhanced stability for membrane protein coding transcripts, we compared 3′UTR lengths between transcript types. As expected, 3′UTRs for fast-decay transcripts were longer than those from slow-decay transcripts (Fig. 4C, D), corroborating earlier findings that mRNAs with longer 3′UTR sequences decay at a significantly faster rate than shorter 3′UTRs.⁴⁰ However, when we compared 3′UTR lengths between membrane protein coding and non-membrane protein coding transcripts, we did not observe shorter 3′UTRs in the membrane protein coding group (Fig. 4C, D). We expected shorter 3′UTRs due to the improved stability of these transcripts as defined by their slower decay rate. This discrepancy suggests that there are other structural or biological features of these mRNAs that account for their enhanced stability beyond 3′UTR length that need to be further explored. It is likely that transcripts encoding membrane proteins are less decorated with RBPs and the RBPs that do bind promote stability instead of signaling degradation. To explore this hypothesis, we conducted a de novo motif search in the 5′ and 3′ UTRs of transcripts encoding membrane proteins within the “slow decay rate” class as compared to all other transcripts in the same class. We identified motifs binding RBPs that induce stability (HuR and FMR1) in HeLa cells; however, the motifs enriched in the SH-SY5Y transcripts require further investigation. It is likely that the difference in motifs between “slow SPC” and “slow non-SPC” transcripts in SH-SY5Y cell line is not large enough to be statistically meaningful (Fig. 4E).

Current study focused on uncovering posttranscriptional mechanisms in immortalized cell lines, these findings will need to be validated on a case-by-case basis for transcripts of interest in primary cells which may exhibit a RNA binding protein milieu that is different from cell lines thus affecting transcript structure, occupancy, decay rates and expression. Detailed exploration of sequence features, RBP interactions, and UTR regions could provide deeper insights into posttranscriptional regulation of membrane protein coding transcripts.

Footnotes

Author Disclosure Statement

All the authors associated with Eli Lilly and Company are current employees and stockholders of Eli Lilly and Company.

Funding Information

This work was supported by Eli Lilly and Company.

Data Availability Statement

Data generated in this study have been included in the Supplementary Tables S1, S2, S3, and . SH-SY5Y proteomics data were licensed from OmicScouts only for meta-analyses, raw data and quantifications are not included in this publication.

Supplemental Material

References

Larsson

, Sander

, Marks

. mRNA turnover rate limits siRNA and microRNA efficacy. Mol Syst Biol, 2010; 6:433; doi: 10.1038/msb.2010.89

Knudsen-Palmer

, Raman

, Ettefa

, et al. Target-specific requirements for RNA interference can arise through restricted RNA amplification despite the lack of specialized pathways. Elife, 2024; 13:RP97487; doi: 10.7554/eLife.97487.3

Santos

, Ursu

, Gaulton

, et al. A comprehensive map of molecular drug targets. Nat Rev Drug Discov, 2017; 16:19–34; doi: 10.1038/nrd.2016.230

Sun

, Setrerrahmane

, Li

, et al. Nucleic acid drugs: Recent progress and future perspectives. Signal Transduct Target Ther, 2024; 9:316–331; doi: 10.1038/s41392-024-02035-4

Zhu

, Zhu

, Wang

, et al. RNA-based therapeutics: An overview and prospectus. Cell Death Dis, 2022; 13:644; doi: 10.1038/s41419-022-05075-2

Ozcan

, Ozpolat

, Coleman

, et al. Preclinical and clinical development of siRNA-based therapeutics. Adv Drug Deliv Rev, 2015; 87:108–119; doi: 10.1016/j.addr.2015.01.007

Kawata

, Wakida

, Yamada

, et al. Metabolic labeling of RNA using multiple ribonucleoside analogs enables the simultaneous evaluation of RNA synthesis and degradation rates. Genome Res, 2020; 30:1481–1491; doi: 10.1101/gr.264408.120

Lugowski

, Nicholson

, Rissland

. Determining mRNA half-lives on a transcriptome-wide scale. Methods, 2018; 137:90–98; doi: 10.1016/j.ymeth.2017.12.006

Kishor

, Fritz

, Haque

, et al. Activation and inhibition of nonsense-mediated mRNA decay control the abundance of alternative polyadenylation products. Nucleic Acids Res, 2020; 48:7468–7482; doi: 10.1093/nar/gkaa491

10.

Herzog

, Reichholf

, Neumann

, et al. Thiol-linked alkylation of RNA to assess expression dynamics. Nat Methods, 2017; 14:1198–1204; doi: 10.1038/nmeth.4435

11.

Duffy

, Simon

. Enriching s4U-RNA using methane thiosulfonate (MTS) chemistry. Curr Protoc Chem Biol, 2016; 8:234–250; doi: 10.1002/cpch.12

12.

Russo

, Heck

, Wilusz

, et al. Metabolic labeling and recovery of nascent RNA to accurately quantify mRNA stability. Methods, 2017; 120:39–48; doi: 10.1016/j.ymeth.2017.02.003

13.

Agarwal

, Kelley

. The genetic and biochemical determinants of mRNA degradation rates in mammals. Genome Biol, 2022; 23:245; doi: 10.1186/s13059-022-02811-x

14.

Boisvert

, Ahmad

, Gierliński

, et al. A quantitative spatial proteomics analysis of proteome turnover in human cells. Mol Cell Proteomics, 2012; 11:M111.011429; doi: 10.1074/mcp.M111.011429

15.

Neumann

, Herzog

, Muhar

, et al. Quantification of experimentally induced nucleotide conversions in high-throughput sequencing datasets. BMC Bioinformatics, 2019; 20:258; doi: 10.1186/s12859-019-2849-7

16.

Leutert

, Rodríguez‐Mias

, Fukuda

, et al. R2‐P2 rapid‐robotic phosphoproteomics enables multidimensional cell signaling studies. Mol Syst Biol, 2019; 15:e9021; doi: 10.15252/msb.20199021

17.

Zecha

, Meng

, Zolg

, et al. Peptide level turnover measurements enable the study of proteoform dynamics. Mol Cell Proteomics, 2018; 17:974–992; doi: 10.1074/mcp.RA118.000583

18.

Zecha

, Satpathy

, Kanashova

, et al. TMT labeling for the masses: A robust and cost-efficient, in-solution labeling approach. Mol Cell Proteomics, 2019; 18:1468–1478; doi: 10.1074/mcp.TIR119.001385

19.

Waas

, Snarrenberg

, Littrell

, et al. SurfaceGenie: A web-based application for prioritizing cell-type-specific marker candidates. Bioinformatics, 2020; 36:3447–3456; doi: 10.1093/bioinformatics/btaa092

20.

Bailey

, Johnson

, Grant

, et al. The MEME Suite. Nucleic Acids Res, 2015; 43:W39–W49; doi: 10.1093/nar/gkv416

21.

Panwar

, Omenn

, Guan

. miRmine: A database of human miRNA expression profiles. Bioinformatics, 2017; 33:1554–1560; doi: 10.1093/bioinformatics/btx019

22.

Parkinson

, Kapushesky

, Shojatalab

, et al. ArrayExpress—a public database of microarray experiments and gene expression profiles. Nucleic Acids Res, 2007; 35:D747–D750; doi: 10.1093/nar/gkl995

23.

Van Nostrand

, Freese

, Pratt

, et al. A large-scale binding and functional map of human RNA-binding proteins. Nature, 2020; 583:711–719; doi: 10.1038/s41586-020-2077-3

24.

Morales

, Pujar

, Loveland

, et al. A joint NCBI and EMBL-EBI transcript set for clinical genomics and research. Nature, 2022; 604:310–315; doi: 10.1038/s41586-022-04558-8

25.

Harrison

, Amode

, Austine-Orimoloye

, et al. Ensembl 2024. Nucleic Acids Res, 2024; 52:D891–D899; doi: 10.1093/nar/gkad1049

26.

Langmead

, Trapnell

, Pop

, et al. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol, 2009; 10:R25; doi: 10.1186/gb-2009-10-3-r25

27.

Fansler

, Mitschka

, Mayr

. Quantifying 3′UTR length from scRNA-seq data reveals changes independent of gene expression. Nat Commun, 2024; 15:4050; doi: 10.1038/s41467-024-48254-9

28.

Muhar

, Ebert

, Neumann

, et al. SLAM-seq defines direct gene-regulatory functions of the BRD4-MYC axis. Science, 2018; 360:800–805; doi: 10.1126/science.aao2793

29.

Iwamoto

, Stadler

, Chalupníková

, et al. Transcription-dependent nucleolar cap localization and possible nuclear function of DExH RNA helicase RHAU. Exp Cell Res, 2008; 314:1378–1391; doi: 10.1016/j.yexcr.2008.01.006

30.

Gallouzi

, Brennan

, Stenberg

, et al. HuR binding to cytoplasmic mRNA is perturbed by heat shock. Proc Natl Acad Sci U S A, 2000; 97:3073–3078; doi: 10.1073/pnas.97.7.3073

31.

Beggs

. Lsm proteins and RNA processing. Biochem Soc Trans, 2005; 33:433–438; doi: 10.1042/BST0330433

32.

Chen

CYA

, Shyu

. AU-rich elements: Characterization and importance in mRNA degradation. Trends Biochem Sci, 1995; 20:465–470; doi: 10.1016/s0968-0004(00)89102-1

33.

Grzybowska

, Wilczynska

, Siedlecki

. Regulatory functions of 3′UTRs. Biochem Biophys Res Commun, 2001; 288:291–295; doi: 10.1006/bbrc.2001.5738

34.

Jia

, Mao

, Ji

, et al. Decoding mRNA translatability and stability from the 5′ UTR. Nat Struct Mol Biol, 2020; 27:814–821; doi: 10.1038/s41594-020-0465-x

35.

, Sun

, Li

, et al. HuR-mediated nucleocytoplasmic translocation of HOTAIR relieves its inhibition of osteogenic differentiation and promotes bone formation. Bone Res, 2023; 11:53–16; doi: 10.1038/s41413-023-00289-2

36.

Richter

, Zhao

. The molecular biology of FMRP: New insights into fragile X syndrome. Nat Rev Neurosci, 2021; 22:209–222; doi: 10.1038/s41583-021-00432-0

37.

Arango

, Sturgill

, Alhusaini

, et al. Acetylation of cytidine in mRNA promotes translation efficiency. Cell, 2018; 175:1872–1886.e24; doi: 10.1016/j.cell.2018.10.030

38.

Kraus

, Wang

, Zheng

, et al. Absolute quantification of mammalian microRNAs for therapeutic RNA cleavage and detargeting. Rna, 2025; 31:1081–1090; doi: 10.1261/rna.080566.125

39.

Davis

, Hildebrand

, MacMillan

, et al. Systematic analysis of siRNA and mRNA features impacting fully chemically modified siRNA efficacy. Nucleic Acids Res, 2025; 53:gkaf479; doi: 10.1093/nar/gkaf479

40.

Yang

, van Nimwegen

, Zavolan

, et al. Decay rates of human mRNAs: Correlation with functional characteristics and sequence attributes. Genome Res, 2003; 13:1863–1872; doi: 10.1101/gr.1272403

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

4.58 MB

0.40 MB

0.35 MB

0.25 MB

0.31 MB

1.57 MB

1.14 MB

0.01 MB

1.57 MB

1.14 MB

0.01 MB