Human cells adapt to translational errors by modulating protein synthesis rate and protein turnover

ABSTRACT

Deregulation of tRNAs, aminoacyl-tRNA synthetases (aaRS) or tRNA modifying enzymes, increase the level of protein synthesis errors (PSE) and are associated with several diseases, but the cause-effect mechanisms of these pathologies remain elusive. To clarify the role of PSE in human biology, we have engineered a HEK293 cell line to overexpress a wild type (Wt) tRNA^Ser and two tRNA^Ser mutants that misincorporate serine at non-cognate codon sites. Then, we followed long-term adaptation to PSE of such recombinant cells by analysing cell viability, protein synthesis rate and activation of protein quality control mechanisms (PQC). Engineered cells showed higher level of misfolded and aggregated proteins; activated the ubiquitin-proteasome system (UPS) and the unfolded protein response (UPR), indicative of proteotoxic stress. Adaptation to PSE involved increased protein turnover, UPR up-regulation and altered protein synthesis rate. Gene expression analysis showed that engineered cells presented recurrent alterations in the endoplasmic reticulum, cell adhesion and calcium homeostasis. Herein, we unveil new phenotypic consequences of protein synthesis errors in human cells and identify the protein quality control processes that are necessary for long-term adaptation to PSE and proteotoxic stress. Our data provide important insight on how chronic proteotoxic stress may cause disease and highlight potential biological pathways that support the association of PSE with disease.

Impact statement

We unveil new phenotypic consequences of protein synthesis errors in human cells and identify the protein quality control processes that are necessary for long-term adaptation to PSE and proteotoxic stress. The data provide important insight on how chronic proteotoxic stress may cause human disease.

Background

Tight control of protein synthesis is essential for cell homeostasis, but the high rate of ribosome decoding, which is necessary to produce sufficient proteins for normal cell functioning, is not compatible with error-free protein synthesis. Lowering translation rate increases protein synthesis accuracy, but impacts negatively on growth rate and fitness [1,2]. Protein synthesis errors (PSE) are, therefore, inevitable and normally arise from erroneous aminoacylation of tRNAs by aminoacyl tRNA synthetases and by mRNA misreading by tRNAs in the ribosome [3,4]. The rate of eukaryotic PSE measured under normal experimental conditions, i.e., downstream of protein quality control (PQC) processes, is approximately 10⁻³ to 10⁻⁴ [4,5], with most misincorporations resulting in conserved or semi-conserved protein mutations that have small impacts on protein stability. However, sustained elevation of marginal levels of PSE induces proteotoxic stress, and may even cause cell death [6–8].

Not surprisingly, mutations that deregulate the expression or alter the function of protein synthesis fidelity factors have been associated with several human diseases, including neurodegenerative diseases and cancer [9–11], as is the case of mutations in the human glycyl-tRNA synthetase (GlyRS) gene in the Charcot-Marie-Tooth neuropathy [12,13]. Mutations in mitochondrial tRNA^Leu are linked to mitochondrial encephalomyopathy, lactic acidosis, and stroke-like episodes (MELAS) [14,15], while expression deregulation of tRNAs and tRNA modifying enzymes have also been observed in cancer [16–21], where there is strong upregulation (>10 fold) of nuclear and mitochondrial encoded tRNAs[17]. Despite these associations with disease, how tRNA expression deregulation and mutations in tRNA and tRNA interacting factor genes cause disease is poorly understood. One possibility is that they elevate PSE and saturate PQC systems, leading to accumulation of aggregated proteins and ultimately to constitutive proteotoxic stress [22–26]. Indeed, the accumulation of misfolded proteins in the endoplasmic reticulum (ER) saturate ER chaperones (GRP78/BiP), triggering the activation of the UPR through the ATF6, IRE1 or PERK [27], which in turn downregulates protein synthesis. This is believed to decrease ER load and increase both ER folding capacity and protein degradation [28]. For example, a mutation in the editing domain of the mouse alanyl-tRNA synthetase (AlaRS) gene leads to the gradual accumulation of protein aggregates across the lifespan and to increased protein ubiquitination, formation of autophagosomes, induction of molecular chaperones, upregulation of the UPR, causing Purkinje cell death and ataxia [22].

Interestingly, tumors have high cell proliferation, metabolism, and protein synthesis rates, which are likely to affect protein synthesis accuracy [29]. PSE due to amino acid deprivation may also increase oxidative stress in tumor microenvironments [30]. This hypothesis is strongly supported by a recent study showing that colon tumors misincorporate amino acids into proteins at higher rate than normal tissue, raising the hypothesis that PSE may be advantageous to tumor growth [11]. This possibility is further corroborated by recent work carried out in the human pathogen Candida albicans, where PSE induced alterations of the cell wall influence fungal adhesion and accelerate the acquisition of drug resistance [31].

To clarify the biology of PSE, we have engineered a HEK293 cell line to synthesize erroneous proteins above background level and used experimental evolution to follow the long-term consequences of the proteome instability generated by such amino acid misincorporations. These cell lines express mutant serine tRNAs that randomly misincorporate serine (Ser) at alanine (Ala) and leucine (Leu) codon sites, on a proteome-wide scale. A cell line expressing extra copies of the wild-type (Wt) tRNA_AGA^Ser was also produced to obtain insight into the cellular consequences of tRNA imbalances. Cellular responses were studied across different time points (cells passages) to clarify how these recombinant HEK293 cells adapt to PSE.

Our data show that those cells cope relatively well with the engineered PSE and that PQC mechanisms were differentially activated, according to the type of error introduced in the proteome.

Results

Expression of mutant tRNAs increase PSE in HEK293 cells

We used mutant and Wt serine tRNAs to investigate how HEK293 cells cope with protein synthesis errors over time (Fig. 1A,B). Mutant Ser-tRNA_AGC^Ala and Ser-tRNA_AAG^Leu further allowed us to investigate how HEK293 cells responded to the random introduction of Ala to Ser and Leu to Ser protein mutations in their proteome. Ser is polar and hydrophilic and is normally present at protein surfaces, whereas Ala is non-polar, hydrophobic and is found both inside and outside protein structures. Leu is hydrophobic and is generally buried in the core of folded proteins [32]. Those cell lines were denominated tRNA^Ser(A) and tRNA^Ser(L). Two additional cell lines were also produced, one was transfected with the empty plasmid and worked as negative control (Mock), and another overexpressed the Wt tRNA_AGA^Ser (tRNA^Ser(S)). To understand long-term adaptation of HEK293 cells to PSE, P1, P15, and P30, post-selection time points, corresponding to the number of cell passages after transfection and selection in geneticin containing media, were chosen. The expression of the mutant tRNAs was monitored by Sanger sequencing (Supplementary Fig. 1). Some of the cell lines acquired mutations in the recombinant tRNA genes after P30, namely in the acceptor’s arm, and, for this reason, the evolution was terminated at P30. tRNA expression was determined using a primer extension reaction (SNaPshot analysis), which allowed us to specifically detect both endogenous and recombinant Wt tRNA_AGA^Ser, as well as the mutant tRNAs (Fig. 1C,D).

Figure 1.

Recombinant tRNAs used in this study and their quantification. (A) representation of the human tRNA_AGA^Ser and mutant tRNAs used in the study. (B) schematic representation of PSE incorporation by the mutant tRNA_AAG^Ser. The diagram shows that the SerRS does not discriminate between endogenous tRNA^Ser and mutant tRNAs charging them with Ser, leading to Ser misincorporation at non-cognate Leu codon sites. (C) detection of tRNA_AGA^Ser expression in mock and tRNA^Ser(S) cell lines. (D) expression of mutant tRNAs relative to the WT endogenous tRNA_AGA^Ser. From C to D, tRNA values were normalized to an endogenous control, GAPDH. Data represent average±SEM of one biological replicate and at least two technical replicates.

In the Mock cell line, the expression levels of the endogenous tRNA^Ser were constant between P1 and P30. In contrast, in recombinant tRNA^Ser(S) cells, the levels of the Wt tRNA_AGA^Ser increased in P1 and P15 (Fig. 1C), and decreased in P30 (Fig. 1C), when compared to the Mock control. In both tRNA^Ser(A) and tRNA^Ser(L) cells, expression of mutant tRNAs decreased sharply in P15, and remained constant from P15 to P30 (Fig. 1D). To understand if the decrease in recombinant tRNA levels was due to tRNA gene copy number changes, we checked the insertion of the tDNAs in the genome of the HEK293 cells (Supplementary Fig. 2). No copy number changes were seen for the tDNA^Ser in recombinant tRNA^Ser(S) and tRNA^Ser(L) cells, while tRNA^Ser(A) cells presented a slight decrease for the tDNA^Ser(A) (Supplementary Fig. 2).

To confirm that recombinant tRNAs were functional and were misincorporating Ser at Ala and Leu sites, we analysed the proteome of the respective HEK293 cell lines by mass spectrometry at P1 (Supplementary Fig. 3). The level of Ser misincorporations at Ala GCT and GCC sites (decoded by the mutant Ser-tRNA_AGC^Ala) was 1.33-fold higher in the tRNA^Ser(A) cell line relative to the Mock cell line (Supplementary Fig. 3A,C). Similarly, the levels of misincorporations of Ser at Leu CTT (2.0 × 10⁻⁴) and CTC (6.0 × 10⁻⁴) sites (decoded by the mutant Ser-tRNA_AAG^Leu) were 4-fold and 12-fold higher in the tRNA^Ser(L) cell line relative to the Mock cell line (Supplementary Fig. 3B,D). These data validated the translational activity of our mutant tRNAs (Ser-tRNA_AAG^Ala and Ser-tRNA_AGC^Leu).

Tolerance of HEK293 cells to PSE

We next assessed the impact of Ser misincorporation, in HEK293 cell’s proteome and over time, by studying cell doubling time, cell viability, proliferation, protein synthesis rate and accumulation of insoluble proteins in recombinant cells. Overexpression of Wt tRNA_AGA^Ser and induced expression of mutant tRNAs did not affect the doubling time, indicating that the number of generations across the experiment was similar for all cell lines (Fig. 2A). In tRNA^Ser(A) and tRNA^Ser(L) cell lines, there was a statistically significant increase in the percentage of viable cells (5.14% and 6.08%, respectively) relative to the Mock cell line in P1 (Fig. 2B), which decreased slightly from P1 to P15 and recovered from P15 to P30 (Fig. 2B and Supplementary Fig. 4A).

Figure 2.

Phenotypic consequences of protein synthesis errors. (A) doubling time of cells. (B) percentage of viable cells in culture determined by cell counting with trypan blue. (C) relative cell proliferation determined using a BrdU ELISA Kit. Values were normalized to the mock cell line of each passage and represent average±SEM of three independent experiments in triplicate. One-way ANOVA followed by Dunnett’s post-test was used to assess differences between the control (Mock cell line) and our cell lines at each time point (*p< 0.05; **p< 0.01; ***p< 0.001).

tRNA^Ser(S) and tRNA^Ser(A) cell lines showed significant alterations in cell proliferation in P15 and P30, in contrast to tRNA^Ser(L) cells. In P15, only the tRNA^Ser(A) cell line proliferated better than the Mock cell line (1.43-fold) (Fig. 2C), while in P30 both tRNA^Ser(S) and tRNA^Ser(A) cell lines showed higher levels of proliferation than Mock cells (1.08-fold in both). These results demonstrate that misexpression of the Wt tRNA_AGA^Ser and induced expression of the mutant Ser-tRNA_AGC^Ala provide a proliferative advantage during evolution, conversely to the non-conservative Ser-tRNA_AAG^Leu misreader.

We next used the SunSET assay [33] for protein synthesis rate quantification. tRNA^Ser(A) cells displayed higher rates of protein synthesis in P1 (1.23-fold) and P15 (1.48-fold), that decreased in P30, relative to Mock cells (Fig. 3A). The high protein synthesis rate in tRNA^Ser(A) cells likely explains the high proliferation levels observed in P15, which remained high in P30. In contrast, tRNA^Ser(L) cells were not different from Mock cells in P1 and P15, regarding protein synthesis rate, but presented a significantly lower relative protein synthesis rate in P30 (0.61-fold) (Fig. 3A). The later result is compatible with the lack of changes in proliferation in this tRNA mutant cell line.

Figure 3.

Protein synthesis rate and insoluble protein content. (A) relative protein synthesis rate. Upper panel: Representative immunoblot images for each time point and cell line and β-tubulin. Lower panel: Protein synthesis rate determined by SunSET. (B) relative insoluble protein fraction. Upper panel: Representative acrylamide gel of the insoluble fraction, after coomassie staining of each time point and cell line. Lower panel: Relative insoluble protein fraction. Values were normalized to the mock cell line of each passage and represent average+SEM of at least three independent experiments. Mock values for each passage were considered 1 and were not represented in the graphs. One-way ANOVA followed by Dunnett’s post-test was used to assess differences between the control (Mock cell line) and our cell lines at each time point (*p< 0.05; **p< 0.01).

Globally, the fraction of insoluble proteins (fraction where protein aggregated accumulate) increased in P15 in the three cell lines, although it was only statistically significantly different from Mock cells for the tRNA^Ser(L) cell line (Fig. 3B). This phenotype was consistently transient, and the level of insoluble proteins decreased to the Mock cell line levels in P30. This result may be related to the lower expression level of the mutant tRNA in P30, but it may also indicate that cells counteracted protein aggregation through activation of PQC mechanisms or through accumulation of compensatory genome mutations [34].

PQC mechanisms allow HEK293 cells to adapt to PSE

In order to understand how these cells tolerate PSE, we analysed the transcriptional deregulation in each passage (P1, P15 and P30) relative to Mock (Fig. 4). The global gene expression deregulation pattern was consistent with the induction of ER stress and protein folding in P1, and/or P15. In the case of the tRNA^Ser(S) cell line, genes involved in the response to ER stress, ER membrane, protein processing in the reticulum and protein folding were upregulated in P1 and P15. Interestingly, autophagy was downregulated in the tRNA^Ser(S) cell line (in P1), indicating that it is not the first line of defence against PSE. In the tRNA^Ser(A) cell line, the genes involved in the response to ER stress and ERAD pathway were also deregulated, namely in P1. In P15 and P30, cells upregulated genes involved in cell proliferation and Ras signalling, indicating that PQC mechanisms were activated at the beginning of the evolutionary process, and likely explaining the increased proliferation in P15 (Fig. 2B). Regarding tRNA^Ser(L) cell line, PQC mechanisms were activated in P15 and P30, by upregulation of genes involved in ER stress and protein ubiquitination. PSE in this cell line resulted in alterations in telomere organization and nuclear chromosome in P15 and in cell cycle arrest in P30, as well as downregulation of calcium ion homeostasis genes. Extracellular region genes were downregulated in all cell lines, likely due to saturation of the ER and the secretory pathway with misfolded proteins (Fig. 4A).

Figure 4.

GO terms deregulated by PSE in HEK293 cells. Percentage of genes deregulated in each GO term in the A – tRNA^Ser(S), B – tRNA^Ser(A) and C – tRNA^Ser(L), cell lines, in each passage (P1, P15 and P30). Microarrays statistical analysis was carried out using MeV software and functional annotation using DAVID tools.

Since several extracellular membrane genes were down deregulated, including integrins that mediate cell adhesion to collagen type 1, cell-matrix adhesion was functionally investigated. Decreased cell-matrix adhesion was observed in tRNA^Ser(A) and tRNA^Ser(L) cell lines (0.74 and 0.76-fold, respectively), in P15 (Fig. 5A). Adhesion recovered to nearly control levels in P30, indicating that these alterations were transient.

Figure 5.

Alterations in cell adhesion and intracellular calcium levels. (A) relative cell adhesion determined by adhesion to collagen I assay. (B) relative intracellular calcium levels determined using Fluo-8 calcium assay kit. One-way ANOVA followed by Dunnett’s post-test was used to assess differences between the control (Mock cell line) and cell lines at each time point (*p< 0.05).

Calcium intracellular levels also suffered oscillations in cells accumulating PSE, namely in tRNA^Ser(S) tRNA^Ser(L) cells in P1 (0.75 and 0.8-fold, respectively) (Fig. 5B). Since calcium homeostasis is crucial for ER function, these data are consistent with increased ER stress in P1.

PSE activate the UPS and deregulate molecular chaperones

We have observed high levels of ubiquitinated proteins in P15, but not in P30, in all cell lines 1.36, 1.37 and 1.48-fold, for tRNA^Ser(S), tRNA^Ser(A) and tRNA^Ser(L), respectively (Fig. 6A), which was consistent with the trend for elevated levels of insoluble proteins in P15. In particular, the tRNA^Ser(A) cell line had higher levels of ubiquitinated proteins in P1 (1.35-fold), but ubiquitination levels decreased to Mock levels between P15 and P30 (Fig. 6A).

Figure 6.

Activity of PQC systems. (A) relative protein ubiquitination determined by immunoblot. Left panel: Representative immunoblot images for each time point and cell line plus β-tubulin are shown. Right panel: Protein ubiquitination relative to mock in each passage. (B) relative proteasome activity assessed by fluorescent measurement of the labelled substrate Suc-LLVY-AMC. Values were normalized to the mock cell line of each passage and represent average±SEM of at least three independent experiments in triplicate. Mock values for each passage were considered 1 and were not represented in the graphs. One-way ANOVA followed by Dunnett’s post-test was used to assess differences between the control (Mock cell line) and cell lines at each time point (*p< 0.05; **p< 0.01).

The tRNA^Ser(A) and tRNA^Ser(L) cell lines had higher proteasome activity in P1 (1.19 and 1.27-fold, respectively) (Fig. 6B), and such activity remained high in the tRNA^Ser(A) cell line in P15 and P30 (2.17 and 1.66-fold, respectively) (Fig. 6B). In contrast, there was a decrease in proteasome activity in the tRNA^Ser(L) cell line from P15 to P30 (Supplementary Fig. 4C), suggesting that the accumulation of ubiquitinated and insoluble proteins in P15, particularly in the tRNA^Ser(L) cell line, could have inhibited proteasome activity and its degradative capacity.

The expression of the molecular chaperones Hsp70, Hsp27, Hsp60, Hsp90α and BiP was also investigated (Table 1). Hsp70 levels decreased during evolution, namely in P15 in the tRNA^Ser(S) cell line (0.60-fold) and in P30 in the tRNA^Ser(A) and tRNA^Ser(L) cell lines (0.67 and 0.55-fold, respectively) (Fig. 7A and Supplementary Fig. 5A), which is consistent with deregulation of Hsp70 during stress and in aged cells [35].

Table 1.

Summary of the deregulation of molecular chaperones during in vitro evolution of the recombinant HEK293 cells.

Molecular chaperone	Role	tRNASer(S)	tRNASer(A)	tRNASer(L)
Hsp70	Prevents aggregation and promotes folding. Recruits ubiquitin ligases to tag proteins for proteasomal degradation.	↓P15	↓P30	↓P30
Hsp27	Refolding. Favours degradation of some ubiquitinated proteins by the proteasome and is involved in apoptotic signalling pathways.	↑P15	↑P15	↑P15 ↓P30
Hsp60	Protein assembly by forming a hetero-oligomeric protein complex in the mitochondria.	-	-	↓P30
Hsp90α	Refolding of specific proteins involved in signal transduction and ER transmembrane kinases that participate in the UPR.	↑P30	↓P1	↓P1
BiP	Facilitates folding and assembly of nascent polypeptides prevents misfolding and aggregation, controls the signalling for the initiation of the various arms of the UPR.	-	-	↑P30

Figure 7.

Responses of molecular chaperones to protein synthesis errors. (A) Hsp70 expression. (B) Hsp27 expression. (C) Hsp60 expression. (D) Hsp90α expression. (E) BiP expression. The corresponding immunoblots are represented for each molecular chaperone. Values were normalized to the mock cell line of each passage and represent average±SEM of at least three independent experiments in triplicate. Mock values for each passage were considered 1 and were not represented in the graphs. One-way ANOVA followed by Dunnett’s post-test was used to assess differences between the control (Mock cell line) and our cell lines at each time point (*p< 0.05; **p< 0.01; ***p< 0.001).

Hsp27 expression increased in tRNA^Ser(S), tRNA^Ser(A) and tRNA^Ser(L) cells in P15 (1.49, 1.35 and 1.69-fold, respectively) and was downregulated in all cell lines, in particular in the tRNA^Ser(L) cell line, in P30 (0.53-fold) (Fig. 7B). The increase in Hsp27 levels was consistent with the increase in ubiquitinated (Fig. 6A) and insoluble proteins in the tRNA^Ser(L) cell line in P15 (Fig. 3B), suggesting that Hsp27 was recruited to destabilize aggregated proteins. However, the levels of Hsp27 decreased between P15 and P30 in tRNA^Ser(S), tRNA^Ser(A) and tRNA^Ser(L) cells (Supplementary Fig. 5B,C).

Hsp60 expression decreased in P30 in tRNA^Ser(L) cell line (0.69-fold) (Fig. 7C), indicating that cytoplasmic PSE may also have an impact on mitochondrial homeostasis, particularly in this cell line. Regarding Hsp90α, its levels were downregulated in tRNA^Ser(A) and tRNA^Ser(L) cell lines (0.78 and 0.71-fold, respectively) (Fig. 7D), but increased in tRNA^Ser(S) cell line in P30 (1.31-fold) (Fig. 7D). BiP expression levels increased during evolution (Supplementary Fig. 5D) and were higher in the tRNA^Ser(L) cell line relative to Mock in P30 (1.34-fold) (Fig. 7E), suggesting the presence of misfolded proteins in the ER (Table 1).

PSE activate the UPR

To clarify whether PSE activated the UPR, we studied the activation of the transcription factor 6 (ATF6), which is cleaved under stress and functions as a transcriptional factor [28]. The ratio of fragmented ATF6/total ATF6 increased in the tRNA^Ser(S) cell line in P15 and in the tRNA^Ser(A) cell line in P30 (1.69 fold in both cases), relative to Mock cells (Fig. 8A). In the tRNA^Ser(A) cell line, the levels of the fragmented ATF6 increased, in line with the accumulation of ubiquitinated proteins and proteasome activity. This is physiologically relevant since ATF6 is also responsible for protection against ER stress-induced apoptosis and cell survival [36].

Figure 8.

Differential UPR activation by protein synthesis errors. (A) ATF6f/ATF6t expression. (B) eIF2αP/eIF2α expression. The corresponding immunoblots are represented in the figure. Values were normalized to the mock cell line of each passage and represent Average±SEM of at least three independent experiments in triplicate. Mock values for each passage were considered 1 and were not represented in the graphs. One-way ANOVA followed by Dunnett’s post-test was used to assess differences between the control (Mock cell line) and our cell lines at each time point (*p< 0.05). (C) specific deregulation of PQC genes (UPR, UPS, and autophagy) by protein synthesis errors. Gene expression profiles assessed using cDNA microarrays of each cell line in each passage. Upregulated genes (red) and downregulated genes (green).

Since UPR activation normally leads to the phosphorylation of the eukaryotic initiation factor 2α (eIF2α-P) by PERK with concomitant decrease in protein synthesis rate [37], the ratio of eIF2α-P/eIF2α was determined. It was increased in tRNA^Ser(L) cells (1.6 fold) in P30, relative to Mock (Fig. 8B), which is consistent with the decreased protein synthesis rate in this passage (Fig. 3A). The decrease in eIF2α-P observed in the tRNA^Ser(A) was rather surprising and suggested that tRNA^Ser(A) and tRNA^Ser(L) cell lines used different strategies to cope and adapt to PSE.

We also assessed whether transcription of UPR, UPS, and autophagy genes (Fig. 8C) was altered. There was a different pattern of gene deregulation in each cell line, but the PQC genes were mainly upregulated indicating, once again, the importance of these mechanisms for cell adaptation to PSE. The tRNA^Ser(S) cell line upregulated PQC genes in P1 and P15, while almost no deregulation was observed in P30. The tRNA^Ser(A) and tRNA^Ser(L) cell lines upregulated PQC genes mainly in P1 and P15, respectively.

The expression of XBP1 mRNA and the factor that catalyzes its splicing (ERN1) were also studied. In P1, ERN1 was downregulated 2.4-fold in tRNA^Ser(S), 2.0-fold in tRNA^Ser(A) and 1.8-fold in tRNA^Ser(L) cells, while XBP1 (unspliced transcript) was upregulated 1.5-fold in P1 and 1.8-fold in P15 in tRNA^Ser(S) and 1.4-fold in tRNA^Ser(A) cells. ERN1 downregulation in P1 in tRNA^Ser(S) and tRNA^Ser(A) cells should lead to accumulation of XBP1u, which is constitutively expressed and thought to function as a negative feedback regulator of XBP1s. Such putative shut down of transcription of target genes during the recovery phase of ER stress may explain the level of deregulation of PQC genes in tRNA^Ser(S) and tRNA^Ser(A) cells in P30 and P15, respectively.

The microarray data also showed upregulation of the autophagy ATG12 gene in tRNA^Ser(L) cells (1.3-fold), whose complex is required for the formation of the autophagosomes involved in the degradation of protein aggregates [38]. In other words, it is likely that autophagy activation may lower the levels of insoluble proteins in the tRNA^Ser(L) cell line.

Discussion

Recent works suggest that PSE may cause disease by overloading chaperones, the proteasome and autophagy. Downstream effects are likely to involve increased energetic costs of protein degradation, deregulation of cell signalling and metabolism pathways, accumulation of toxic protein aggregates, repression of protein synthesis and genomic instability [7,24,31,39]. We have also observed alterations in intracellular calcium levels and cell-matrix adhesion. Alterations in calcium homeostasis are correlated with ER stress and are common pathological events in protein misfolding diseases [40]. Indeed, ER chaperones require calcium for their protein folding activity and a decrease in ER-calcium may inhibit the folding and maturation of secretory proteins leading to stress, while calcium increase in the cytoplasm may induce mitochondrial-mediated apoptosis [40,41]. The transient decrease in P1 in tRNA^Ser(S) and tRNA^Ser(L) cells showed that PSE have the potential to alter calcium homoestasis. Cell adhesion was also compromised in the cell lines expressing mutant tRNAs in P15. Several genes coding for adhesion proteins, such as integrins and cadherins, and extracellular matrix proteins were downregulated in tRNA^Ser(A) and tRNA^Ser(L) cell lines, probably to attenuate ER stress, leading to decreased cell adhesion to collagen type 1 matrix. When the levels of protein misfolding and aggregates were restored, cell adhesion was no longer compromised (Figs. 3B, 5A and 6A).

Kalapis and Bezerra have shown that misincorporation of Ser at Leu sites leads to upregulation of protein synthesis and protein degradation, as well as increased uptake of glucose in yeast [42]. Mistranslating yeast clones evolved for 250 generations were able to reduce protein aggregates and recovered fitness to almost wild-type levels, but at a high metabolic cost [42]. The strong negative effect of mistranslation observed in yeast growth was not observed in HEK293 cells, but our data are in line with the yeast data, as protein synthesis and degradation rates increased during evolution in the tRNA^Ser(A) cell line (Figs. 3A, 6B).

The decrease in protein aggregation levels observed during evolution of both yeast and HEK293 cells (namely in tRNA^Ser(L) cell line) has implications for understanding the biology of protein misfolding diseases. Protein aggregation studies use cell models expressing aggregation-prone proteins, but do not evaluate long-term adaptation to the aggregates [43–45]. Even in cases where these proteins are expressed constitutively, the norm is to maintain cell passage number as low as possible to avoid genomic instability [46]. Our data suggest that human cell models of Alzheimer’s, Parkinson’s and other protein misfolding diseases should be characterized in long-term adaptation experiments to capture the full spectrum of metabolic and physiological changes induced by protein aggregation. Indeed, aggregates associated with neurological disorders can block proteasome activity and may activate mechanisms that repress protein synthesis [10,47], compromising adaptation to such aggregates [48,49]. Moreover, different human cells cope differently with protein aggregation, suggesting that adaptation to aggregation may follow different routes in different cell types. Recent studies showing that repression of eIF2α kinases alleviates Alzheimer’s symptoms in mice support this hypothesis [50].

A rather surprising result of our study was the phenotypic variability induced by the heterologous expression of the Wt tRNA^Ser (tRNA^Ser(S)). It is known that increased expression of Wt tRNAs, which is common in cancer, alters translation rate, enhances expression of oncogenes and may also increase the level of PSE, leading to accumulation of misfolded proteins [17,51]. We observed, in the tRNA^Ser(S) cells (P15), accumulation of ubiquitinated proteins, UPS and UPR activation through the ATF6 branch (Fig. 9). This has been associated with tolerance to chronic ER stress, via upregulation of several genes that preserve ER function, including protein folding, protein degradation and maintenance of general ER homeostasis (Fig. 8)[52]. Therefore, it is likely that misexpression of Wt tRNAs activates the stress response and lead to translational deregulation of gene expression [53].

Figure 9.

Summary of the PQC alterations identified in the different cell lines. Mutant tRNAs and misexpression of tRNA^Ser led to the accumulation of ubiquitinated proteins, suggesting increased levels of misfolded proteins. PQC mechanisms were recruited in an error type and time-dependent manner to counteract proteotoxic stress. Increased protein turnover and decreased protein synthesis seem to be two important mechanisms induced by PSE that cells used to thrive in culture after several generations.

Misincorporations of Ser at Ala sites resulted in increased protein synthesis rate (P1 to P15) and proteasome activity (P1 to P30), confirming previous results obtained in mistranslating yeast strains (Fig. 9) [42]. Therefore, this Ser-to-Ala misincorporation model may be relevant to address the biology of PSE in cancer, where PQC mechanisms are highly activated [54,55]. Conversely, the more disruptive Ser-to-Leu misincorporation model displayed accumulation of aggregated proteins (P15) (Fig. 9) and adaptation was more dependent on UPR activation, namely phosphorylation of eIF2α, and consequent inhibition of protein synthesis rate. Phosphorylation of eIF2α promotes polysome disassembly, accumulation of untranslated messenger ribonucleoprotein particles (mRPs), stress granules, and is responsible for reprogramming mRNA metabolism and also contributes to cell survival [56,57].

During evolution in vitro, HEK293 cells presented several alterations that allowed them to thrive, mitigate PSE and decrease the level of misfolded and aggregated proteins. Some of these alterations were transient, such as the initial stress response, characterized by deregulation of intracellular calcium levels, activation of proteasome activity and increase in protein synthesis rate (Fig. 9). In P30, some important molecular alterations were still observed, namely the downregulation of mutant tRNAs (error mitigation), increased expression of specific molecular chaperones (such as HSP90 in and BiP) and of UPR, UPS and autophagy genes.

Overall, we unveil new phenotypic consequences of protein synthesis errors in human cells and identified the protein quality control processes that are necessary for long-term adaptation to PSE and proteotoxic stress. The data provide important insight on how chronic proteotoxic stress may cause disease.

Materials and methods

Cell culture

Human embryonic kidney 293 (HEK293) cells were purchased from American Type Culture Collection (ATCC®CRL-1573). Cells were grown in Minimum Essential Medium (Gibco, Cat.41090–028) supplemented with 10% foetal bovine serum (FBS) (Sigma, Cat.F1051), 1% of Pen/Strep (Gibco, Cat.15070–063) and 1% of non-essential amino acids (Gibco, Cat.11140–050) in a humidified atmosphere at 37ºC in the presence of 5% CO2.

Construction of mutant tRNA plasmids

A DNA fragment of 248 bp, containing the gene encoding the human wild-type tRNA_AGA^Ser (Chr6 tRNA#5) and its flanking region, was amplified by PCR and cloned into the modified vector pIRES2-DsRed with new multiple cloning sites. The anticodon of the tRNA_AGA^Ser was mutated to non-cognate anticodons by site-directed mutagenesis.

Generation of mistranslating cell lines

HEK293 cells with approximately 60% of confluency were transfected with 1 µg of plasmid DNA using Lipofectamine2000 (Invitrogen, Cat.11668019), following manufacturer’s instructions. Cells were transfected with an empty vector (Mock), and with the plasmid carrying tRNA_AGA^Ser(S), tRNA_AGC^Ser(A) or tRNA_AAG^Ser(L) genes. Stable cell lines were established by selection with 800 µg/ml of geneticin (Formedium, Cat.G4185) for 1 month. Cells were kept in 100 µg/ml of geneticin after selection and during evolution in culture. Geneticin was removed before experiments.

Evolution of cells in culture

After transfection with the plasmids and selection, cells were kept in culture dishes (60 mm) and subcultured every 3 days using the same dilution (1/6) until passage 30 (P30). In P1, P15 and P30 cells were plated in 100 mm culture dishes, to obtain sufficient cells to perform the experiments and extract DNA, RNA, and protein. The presence of the recombinant tRNAs was confirmed and monitored throughout evolution (Supplementary Fig. 1A,B).

Total RNA extraction

RNA was extracted from 5 × 10⁵ cells using Trizol®Reagent (Thermo Fisher Scientific, Cat.15596026). Purification of RNA was carried out using DNase I, Amplification Grade kit (Invitrogen, Cat.18068015), following manufacturer’s instructions. RNA was then precipitated with a standard Phenol/Chloroform/Isoamyl alcohol (25:24:1) (Acros Organics, Cat.327111000) extraction protocol and conserved at −80°C.

Quantification of tRNA expression and tDNA copy number

Two-hundred nanograms of total RNA were used for cDNA synthesis using NCode™ VILO™ miRNA cDNA Synthesis Kit (Invitrogen, Cat.A11193050), following the manufacturer’s instructions. To determine the copy number of the Wt tDNA^Ser and the mutant tDNA genes, genomic DNA was extracted using Wizard® Genomic DNA Purification Kit (Promega, Cat.TM050). Amplification of the Wt and mutant tRNAs from cDNA (2 µL) or DNA (200 ng) was carried out by PCR (Supplementary Fig. 2A,B). GAPDH was quantified and used as an internal control to normalize tRNA expression levels [58,59]. After the first amplification of cDNA and DNA, quantification of tRNA expression and copy number was performed by Snapshot Sequencing as described in[11].

Gene expression microarrays

Gene expression profiling using microarrays was performed using the Agilent protocol for One-Color Microarray-Based Gene Expression Analysis Low Input Quick Amp Labeling v6.9 (Agilent Technologies). RNA quality determination was performed using 2100 Bioanalyser (Agilent Technologies) and the Agilent RNA 6000 Nano kit (Agilent Technologies, Cat.5067–1512). One hundred nanograms of total RNA were used to synthesize labelled cDNA (with Cyanine 3-CTP), using Agilent T7 Promoter Primer and T7 RNA polymerase Blend (Agilent Technologies, Cat.5190–2305). Six hundred nanograms of labelled cDNA were hybridized in Sure Print G3 Human Gene Expression 8 × 60 k v2 microarrays (Agilent Technologies, Cat.G4851B). Hybridizations were carried out using Agilent gasket slides in a rotating oven for 17 h at 65°C. Slides were then washed following manufacturer’s instructions and scanned in an Agilent G2565AA microarrays scanner.

Probes signal values were extracted using Agilent Feature Extraction Software. Data were normalized using median centering of signal distribution with Biometric Research Branch BRB-Array tools v3.4.o software [60,61]. Microarrays statistical analysis was carried out using MeV software (TM4 Microarray Software Suite) [62,63]. A t-test was performed to identify genes that showed differences in expression between control (Mock) and samples. Significant genes that showed a fold change above 1.5 or below −1.5 were considered for downstream analysis.

The microarray raw data were submitted to the GEO database and has been given the following accession number: GSE93854.

Proteasome inhibition and isolation of the insoluble protein fraction

2 × 10⁵ cells/well were plated in 6 well plates; after 48 h of growth, the proteasome inhibitor MG132 (Sigma-Aldrich, Cat.SML1135) was added to each well at a final concentration of 5 µM and cells were incubated overnight. Cells were detached using Protein Lysis Buffer, sonicated with a probe sonicator in 5 pulses of 5 s, incubated on ice for 30 min and centrifuged at 5000 rpm for 15 min at 4°C. Ten microliters of the supernatant (total protein fraction) was stored to measure protein concentration with BCA assay (Thermo Fisher Scientific). Three hundred micrograms of total protein was centrifuged again at 12000 rpm for 20 min at 4°C to isolate the insoluble fraction of the protein extract. The pellet (insoluble fraction) was then washed with 160 µL of LB and 40 µL of 10%Triton X-100 (Sigma-Aldrich, Cat.X100) and centrifuged at 15000 g for 20 min at 4°C. The pellet was solubilized in 50 µL of LB. The total volume was then resolved in a 10% SDS-PAGE gel. The gel was stained with 0.1% Coomassie Brilliant Blue G solution (Sigma-Aldrich, Cat.B0770) for at least 2 h.

Identification of PSE by mass spectrometry

Complete lanes of 10% polyacrylamide gels were manually cut and sliced into eight sections, destained with 25 mM ammonium bicarbonate/50% acetonitrile and dried under vacuum (SpeedVac®, Thermo Savant, USA). The dried gel pieces were rehydrated with 25 μL of 10 µg/mL trypsin (Promega V5111) in 50 mM ammonium bicarbonate and digested overnight at 37°C. Tryptic peptides were extracted from the gel with 10% formic acid/50% acetonitrile and were then dried in a vacuum concentrator and re-suspended in 10 µL of a 50% acetonitrile/0.1% formic acid solution. Separation of tryptic peptides by nano-HPLC was performed on the module separation Proexeon EASY-nLC 1000 from Thermo equipped with a 50-cm EASY C18 column with particle size 2-µm. Each sample was separated by a gradient of 5–32% ACN in 90 at 250 nl/min. Peptide cations were converted to gas-phase ions by electrospray ionization and analysed on a Thermo Orbitrap Fusion Lumos. Precursor scans were performed from 300 to 1,500 m/z at 120 K resolution (at 445 m/z) using a 1 × 105 AGC target. Precursors selected for tandem MS were isolated at 1 Th with the quadrupole, fragmented by HCD with a normalized collision energy of 30, and analysed using rapid scan in the ion trap. The maximum injection time for MS2 analysis was 50 ms, with an AGC target of 1 × 10⁴. Precursors with a charge state of 2–5 were sampled for MS2. Dynamic exclusion time was set at 60 s, with a 5 ppm tolerance around the selected precursor. The raw files were searched directly against the Homo sapiens reference proteome obtained from human Uniprot database, using PEAKS8 software and peptides with mutations were identified using the SPIDER tool[64]. Searches were performed using a precursor search tolerance of 5 ppm. Search criteria included a static modification of +57.0214 Da on cysteine residues, variable modification of +15.9949 Da on oxidized methionine, and in the case of the human samples, variable modifications +15.994915 on Ala and −26.052036 Da on Leu were also included. Searches were performed with semi-tryptic digestion and allowed a maximum of three missed cleavages on peptides analysed by the sequence database. False discovery rates (FDR) were set to 1% for each analysis. A list of all the peptides detected and mutant peptides was generated and used for further analysis. UniProt and Ensembl databases, as well as PERL scripts developed in-house, were used to match the misincorporations (Ala to Ser and Leu to Ser) detected with the correspondent codon sites. The number of misincorporations (Ala for Ser and Leu for Ser) found was normalized by the total number of peptides detected in each cell line.

Cell fitness assessment

To measure cells doubling time, 3 × 10⁴cells/well were plated in 6-well plates. After 72 h, cells were detached and counted in a Neubauer chamber with Trypan blue 0.4% (Lonza, Cat.17-942E). Population doubling time was calculated using the formula: Doubling time = duration*log(2)/(log(final concentration)-log(initial concentration)) [65]. For viability assays, the number of viable cells in culture was determined with Trypan blue exclusion assay. 3 × 10⁴cells/well were plated in 6-well plates. After 72 h, cells were counted in a Neubauer chamber using Trypan Blue 0.4% (Lonza, Cat.17-942E). For the quantification of cell proliferation, we used a colorimetric immunoassay ELISA, based on the measurement of BrdU incorporation during DNA synthesis (Roche, Cat.11647229001), following manufacturer’s instructions. 1 × 10⁵cells/well were plated in a 96-well and analysis was performed after 48 h.

Protein synthesis rate determination

To determine protein synthesis rate we used the SUnSET method [33] with few modifications. 2 × 10⁵cells/well were plated in 6-well plates and after 48 h, puromycin (Sigma Aldrich, Cat. 07635) was added to each well to a final concentration of 10%. Cells were incubated for 10 min, washed twice with 1%PBS and returned to the incubator for 50 min. After protein extraction with Lysis Buffer and denaturation, 100 µg of protein were resolved in 10% SDS-PAGE and blotted onto nitrocellulose membranes (0.2 µm) (GE Healthcare Life Sciences). Anti-puromycin, clone 12D10 (kindly given by Philippe Pierre) was used (1:5000 dilution) to detect the incorporation of puromycin into proteins. IRDye800 goat anti-mouse secondary antibody (Li-cor Biosciences, Cat.400–33) was used (1:10000 dilution) and detected in an Odyssey Infrared Imaging System (Licor Biosciences). Membranes were also probed with Anti-β-tubulin (Invitrogen, Cat.32–2600) (1:1000 dilution) as loading control.

Quantification of the insoluble protein fraction

2 × 10⁵cells/well were plated in 6-well plates and processed as described above to obtain the insoluble protein fraction after proteasome inhibition. Fifteen microliters of samples were denaturated with loading buffer (6x) at 95°C for 5 min and resolved in 10% SDS-PAGE. Gels were stained with 0.1% Coomassie Brilliant Blue G solution (Sigma-Aldrich, Cat.B0770) for at least 2 h. After destaining with a solution of 10% ethanol and 7.5% acetic acid, gels were scanned using the Odyssey Infrared Imaging System (Licor Biosciences). Lane signals corresponding to each sample were quantified and normalized to the total amount of protein determined with BCA assay.

Quantification of proteasome activity

2 × 10⁵cells/well were plated in 6-well plates. After 48 h cells were washed with 1%PBS and resuspended in 100 µL of Lysis Buffer. Cells were sonicated with a probe sonicator in 5 pulses of 5 s and centrifuged at 13000 rpm for 10 min at 4°C. Protein content in the supernatant was quantified using the Bradford method (Sigma-Aldrich, Cat.B6916). Twenty micrograms of protein were incubated with the substrate suc-LLVY-MCA (50 µM) (Sigma-Aldrich, Cat.S4939) in the presence or in the absence of the proteasome inhibitor MG132 (10 µM) (Sigma-Aldrich, Cat.SML1135). Substrate degradation was monitored every 5 min during 1 h at 37°C in a fluorescence-luminescence detector Synergy™ HT Multi-Mode Microplate Reader (Biotek), set to 380 and 460 nm, excitation and emission wavelengths, respectively. Specific proteasome activity was determined by subtracting the values for each sample without MG132 to the values with MG132. The final activity was calculated as fluorescence emission at 0 min subtracted from fluorescence after 1 h relative to control (Mock).

Cell-matrix adhesion assay

Ninety-six-well plates were coated with 30 uL/well of 40 ug/mL Collagen I solution in 1% PBS (Sigma-Aldrich, cat.no.C7661). Cells were deprived of serum for 8 h before the adhesion assay. 2 × 10⁵ cells/mL were resuspended in MEM with 0.1% BSA and added to the collagen I-coated wells (100 uL). Cells were incubated at 37°C for 30 min. Non-adherent cells were washed four times with MEM. Cells were supplemented with MEM 10% FBS and incubated at 37°C for 4 h for recovery. Ten microlitersof 1 mg/mL MTT solution (Abcam, cat. No. ab146345) was added to each well and incubated for 2 h at 37°C. MTT treated cells were lysed with 100 uL of DMSO and absorbance was measured at 570 nm on a spectrophotometer.

Determination of intracellular calcium levels

80 × 10⁴cells/well were plated in 96-well plates with black wall and clear bottom. After 24 h cells were washed with HHBS and Fluo-8-dye-loading solution was added to the cells following the protocol from the kit: Abcam Fluo-8 No Wash Calcium Assay, ab112129. Cells were incubated 1 h in a cell incubator. Absorbance (Ex/Em:485/528) was determined at a steady state. Carbachol was added after the steady-state measurement to induce calcium release from intracellular sources and determine intracellular calcium levels (50 uL/well, 100 uM). Absorbance was monitored for 4 min, and several readings were obtained using a Synergy™ HT Multi-Mode Microplate Reader (Biotek). Calcium levels were determined by subtracting the values for each sample with Carbachol and in a steady state. Values were normalized for the control cell line (Mock).

Immunoblots

2 × 10⁵ cells were plated in 6-well plates. After 48 h, cells were washed with 1%PBS and then lysed with protein Lysis Buffer. Cells were then sonicated with a probe sonicator in 5 pulses of 5 s. After centrifugation, 16000 g for 30 min, protein in the supernatants was quantified using the BCA assay (Thermo Fisher Scientific, Cat. 23,225). Samples were denaturated with loading buffer (6x) at 95°C for 5 min.

Protein samples were fractionated by 10% SDS-PAGE gels (or 8% for molecular chaperones and ATF6), transferred to nitrocellulose membranes (0.2 µm) and immunobloted. The following primary antibodies were used: anti-ubiquitin (1:1000 dilution) (Sigma-Aldrich, Cat.U0508), anti-Hsp70 (1:1000 dilution) (Stress Marq Biosciences, Cat.SMC-100B), anti-Hsp27 (0.5 µg/ml dilution) (Stress Marq Biosciences, Cat.SMC-161A), anti-Hsp60 (1:1000 dilution) (Stress Marq Biosciences, Cat.SPC-105), anti-Hsp90α (1:1000 dilution) (Stress Marq Biosciences, Cat.SMC-147), anti-BiP (1:1000 dilution) (Stress Marq Biosciences, Cat.SPC-180), anti-ATF6 (1:400 dilution) (Stressgen, Cat.70B1413.1), anti-eIF2α (1:1000 dilution) (Cell Signalling, Cat.9722), anti-phosphorylated eIF2α (1:400 dilution) (Abcam, Cat.ab4837), anti-β-tubulin (Invitrogen, Cat.32–2600). β-tubulin was used in all the immunoblots as loading control.

Statistical analysis

For all assays, our data represent at least three independent experiments and three replicates. Statistical analysis was performed using One-way ANOVA analysis of variance followed by the Dunnett’s or Bonferroni’s post-tests, as indicated in the figures.

Funding Statement

Ipatimup integrates the i3S Research Unit. iBiMED and i3SResearch Units are partially supported by the Portuguese Foundation for Science and Technology (Fundação para a Ciência e a Tecnologia-FCT). This research was founded by 1) FEDER – Fundo Europeu de Desenvolvimento Regional funds through the COMPETE 2020 –- Operational Programme for Competitiveness and Internationalization (POCI), Portugal 2020, and by Portuguese funds through FCT –- Foundation for Science and Technology/Ministério da Ciência, Tecnologia e Inovação in the framework of the projects: ‘Institute for Research and Innovation in Health Sciences’(POCI-01-0145-FEDER-007274), and ‘PEst-C/SAU/LA0003/2013’; 2) NORTE-01-0145-FEDER-000029, supported by Norte Portugal Regional Programme (NORTE 2020), under the PORTUGAL 2020 Partnership Agreement, through the European Regional Development Fund (ERDF); 3) The Aveiro Institute of Biomedicine – iBiMED is supported by FCT grant UID/BIM/04501/2013, the Ilídio Pinho Foundation and the University of Aveiro, POCI-01-0145-FEDER-007628, and through European and Structural funds (CENTRO2020) provided by CCDR through project PAGE: CENTRO-01-0145-FEDER-000003. The funders had no role in study design, data collection, and analysis, decision to publish, or preparation of the manuscript; Fundação para a Ciência e a Tecnologia [POCI-01-0145-FEDER-016428]; Fundação para a Ciência e a Tecnologia [SFRH/BD/91020/2012]; Fundação para a Ciência e a Tecnologia [SFRH/BD/76417/2011]; Fundação para a Ciência e a Tecnologia [POCI-01-0145-FEDER-028834]; Fundação para a Ciência e a Tecnologia [PTDC/BiA-MiB/31238/2017]; Fundação para a Ciência e a Tecnologia [PEst-C/SAU/LA0003/2013]; Fundação para a Ciência e a Tecnologia [PTDC/BIM-MEC/1719/2014]; Fundação para a Ciência e a Tecnologia [PTDC/BEX-BCM/2121/2014]; Fundação para a Ciência e a Tecnologia [UID/BIM/04501/2013]; FEDER [POCI-01-0145-FEDER-007628]; Fundo Europeu de Desenvolvimento Regional [NORTE-01-0145-FEDER-000029]; FEDER [POCI-01-0145-FEDER-007274]; CENTRO2020 [CENTRO-01-0145-FEDER-000003].

Acknowledgments

The authors are most grateful to the Portuguese Foundation for Science and Technology (FCT), POCH, FEDER, COMPETE2020 and CENTRO2020-CCDRC. The authors would also like to acknowledge Dr Philippe Pierre for kindly provide the 12D10 antibody against puromycin and all the arrays used in the microarray experiment.

Authors’ contributions

ASV, MS, and ARS did the experiments. ASV generated the cell lines. ASV and MS detected the tRNAs and protein synthesis errors. ASV did the phenotypic experiments and chaperone analysis by western blot. ASV and MS did the UPR analysis. ASV and ARS did the microarray experiment. ASV and PO did the microarray analysis. ASV and RV did the mass spectrometry experiment. MASS and CO conceived, designed the study and corrected the manuscript. ASV wrote the manuscript. All authors read and approved the final manuscript.

Availability of data and materials

All data generated or analysed during this study are included in this published article [and its supplementary information files].

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental material

Supplemental data for this article can be accessed here.