Global changes in unproductive splicing and NMD efficiency in tumors

Cover Page

Cite item

Full Text

Abstract

The nonsense-mediated mRNA decay (NMD) pathway is a mRNA quality control mechanism which not only degrades deleterious transcripts but also orchestrates a large number of post-transcriptional regulatory programs through unproductive splicing. We have developed a robust metric derived from splicing quantification in the RNA-seq data to measure NMD efficiency at a sample level. We demonstrate that NMD efficiency varies substantially both between and within tissues, with the magnitude of the variation comparable to that observed upon knockdown of the core NMD factor UPF1. By analyzing TCGA cancer cohorts, we further show that, in many tumors, unproductive splicing events undergo coordinated changes towards either collective suppression or collective activation of NMD isoforms, which is indicative of global deregulation of the activity of the NMD pathway. Consistently, we observed a striking divergence of NMD efficiency in cancers from the tissue-specific baseline level, suggesting that tumors partially erase the NMD signature of their tissue of origin. The application of the developed metric to RNA-binding protein knockdowns made it possible to identify several novel potential regulators of NMD efficiency. In sum, this study provides a solid framework for quantifying NMD efficiency, describes its biological and clinical relevance, and opens new avenues for dissecting mechanisms of post-transcriptional gene expression regulation by the NMD pathway.

Full Text

ABBREVIATIONS

NMD – nonsense-mediated decay; USE – unproductive splicing event; RBP – RNA-binding protein; RNA-seq – RNA sequencing.

INTRODUCTION

The nonsense-mediated mRNA decay (NMD) pathway is a mRNA surveillance system that selectively eliminates transcripts containing premature termination codons [1]. Such transcripts may result not only from nonsense or frameshift mutations but also from alternative splicing [2]. On the one hand, elimination of mRNAs with premature termination codons prevents the production of deleterious proteins. On the other hand, NMD acts together with alternative splicing as a widespread post-transcriptional mechanism of gene expression regulation, referred to as unproductive splicing [2, 3]. The activity of the NMD system exerts a substantial impact on the transcriptome composition and has been linked to diverse physiological processes and pathological conditions, such as cellular differentiation, lymphocyte development, stress response, genetic disorders, and cancers [4–9].

Despite the existing knowledge, the role of NMD in tumorigenesis remains controversial [8–10]. While cancer cells may exploit NMD to downregulate tumor suppressors, their adaptation to stress conditions such as hypoxia or therapeutic pressure often requires precise tuning of its efficiency [11, 12]. As a result, NMD can exert both pro- and anti-tumorigenic effects depending on the conditions and the microenvironment [12–16]. In accordance with this, the core NMD factor UPF1 has been reported to be downregulated in certain cancers and upregulated in the others [17–20]. Measuring sample-level NMD efficiency is challenging partly because NMD factors per se undergo negative autoregulation, making their mRNA levels unreliable proxies for pathway activity [21, 22].

Transcript-level determinants of NMD efficiency have been widely studied [11, 23, 24], yet the information on global, condition-specific NMD efficiency often remains contradictory. Early works used reporter constructs and monitored a small number of endogenous NMD targets, revealing apparent inter-tissue variability, but these estimates could suffer from gene-specific regulatory biases [22, 25]. Approaches based on allele-specific expression of transcripts carrying nonsense mutations concluded that NMD efficiency is largely uniform across tissues [23, 24], but a later study of the expression levels of NMD-sensitive isoforms and matched productive isoforms again reported non-random inter-tissue differences [26].

Motivated by these conflicting observations, we sought to develop a robust and sensitive metric for the quantitative assessment of NMD efficiency using RNA sequencing (RNA-seq) data. Here, we introduce an integral NMD efficiency metric based on unproductive splicing events (USE), i.e. local alternative splicing events that generate isoforms containing premature termination codons and are sensitive to NMD inhibition. USEs are far more abundant than heterozygous nonsense mutations and provide substantially greater statistical power. Estimates using this metric have confirmed that inter-tissue differences may reach up to 80% of the effect size observed in UPF1 knockdown. The application to cancer data demonstrated a correlation between the direction of changes in NMD efficiency in tumors and the direction of changes in most individual events. The metric further showed that tumors frequently lose the NMD efficiency levels characteristic of their tissue of origin; in several cancer types, this alteration correlates with unfavorable prognosis. Juxtaposition of USE splicing changes with the corresponding gene-expression changes allowed one to distinguish event-specific splicing regulation from global NMD efficiency shifts. By applying the developed metric to RNA-binding protein (RBP) knockdown experiments, we confirmed known NMD factors and proposed new candidate regulators of the NMD pathway.

EXPERIMENTAL

RNA sequencing data

Transcriptomes of healthy and tumor human tissues from The Cancer Genome Atlas (TCGA) were downloaded from the dbGaP portal as alignments to the GRCh38 human genome assembly. Thirteen tumor cohorts containing at least 15 paired tumor-normal samples were selected (Supplementary Table S1). Only matched tumor-normal samples were used in differential splicing analysis. The transcriptomes of human tissues from the Genotype Tissue Expression (GTEx) V7 project were downloaded in FASTQ format from the dbGaP portal. We selected 10,100 samples with a read length of 75 nt, containing at least 20 million reads per sample (Supplementary Table S2). Transcriptomes of K562 and HepG2 cells subjected to RBP knockdowns or knockouts were downloaded from the ENCODE consortium website [27] as alignments to the GRCh38 human genome assembly. Perturbations performed in two biological replicates in one or both cell lines were selected (Supplementary Table S3). Transcriptomes of HeLa cells subjected to knockdowns of spliceosomal components were obtained from the Array Express repository under accession number E-MTAB-11202 and converted to the FASTQ format [28]. Each knockdown was represented by one biological replicate. Transcriptomes of HeLa cells subjected to knockdowns of UPF1, SMG6, SMG7, and double knockdown of SMG6 and SMG7, as well as rescue experiments, were obtained from the SRA repository under the accession number GSE86148 [29]. Reads from the samples in FASTQ format were mapped to the human genome version GRCh38 using STAR aligner v2.7.8a [30], with GENCODE 43 [31] annotation used as a reference. Gene expression levels were assessed using the FeatureCounts utility [32]. Differential gene expression in TCGA was analyzed using limma-voom from the edgeR package [33]. In RBP and NMD system inactivation experiments, differential expression was analyzed using DESeq2 [34].

USE Catalog

A catalog of annotated unproductive splicing events (USEs) was generated using NMDj utility [35] based on the ENSEMBL genome annotation (version 108). To identify novel USEs, transcriptomes of each TCGA sample were assembled using the StringTie v2.2.1 software with the “conservative” option [36], and the aggregated annotation across all samples was fed to NMDj. Additionally, skipping events of all constitutive exons flanked by constitutive introns were added to the list of novel events. Novel USEs whose characteristic intron sets of NMD transcripts overlapped with the intron sets of annotated USEs were removed in order to obtain non-redundant catalogs [35].

Quantitative assessment of splicing

The number of split reads supporting intron splicing was computed from short read alignments using the IPSA package with the default settings [37]. The splicing rate of each USE was characterized by the Ψ metric calculated using the NMDj package [35], defined as the ratio of the number a of split reads supporting the NMD isoform to the total number a + b of split reads supporting both NMD and the coding isoforms (the value Ψ = 0 indicates the absence of the NMD isoform). Only USEs with a + b > 15 in at least half of the samples in each comparison group were considered. Differential splicing analysis in TCGA and NMD inactivation experiments was estimated using the random-effects model implemented in the statsmodels.stats.meta_analysis module. The arcsine transformation was applied to Ψ values to stabilize variance. The within-sample variance s was estimated using the formula s = 1 / (4n + 2), where n = a + b, and the between-sample variance was estimated using the non-iterative method proposed by DerSimonian and Laird [38].

Estimates of the group means and total variances were used to conduct z-tests for the significance of intergroup differences. For RBP perturbation experiments with few or no biological replicates, a and b values were pooled across all the replicates, and the statistical significance of Ψ differences between groups was assessed using the proportion test. To compare splicing rates between the knockdown and control conditions, ΔΨ = ΨKD – ΨC was used, where ΨKD and ΨC are the Ψ values in the knockdown and control, respectively. The ΔΨ values were averaged across all the inactivation experiments of the same RBP; the P values were calculated using the Stouffer’s z-score method.

Assessment of NMD efficiency

The following metric α was used to select core USEs specifically responding to inactivation of the NMD system, but not to that of individual RBPs. This metric was calculated as the product of the ΔΨ value between the RBP inactivation and the control, normalized by the standard deviation across all RBPs, and the z-score of the corresponding P-value. It was required that the ratio of the α value for NMD inactivation experiments to the largest α value for all the RBPs from the panel of perturbation experiments be not less than L (see below). For each data set (TCGA, GTEx, and ENCODE), core USEs were selected from the set of events for which the proportion of samples with a + b > 15 was at least N (see below), and the proportion of samples with 0 < Ψ < 1 was at least K (see below). For each core event, the Ψ values were transformed to z-scores across all the samples. The NMD efficiency metric of a sample was defined as the negative mean of z-transformed Ψ values across all the core events.

The parameters N, K, and L, as well as core USEs, were optimized separately for each dataset due to biological (tumor, tissue, perturbation) and technical (sequencing depth) heterogeneity. For TCGA: N > 0.5, K > 0.1 in at least 12 out of 13 cohorts, L > 0.3, 120 events were selected. For GTEx: N > 0.9, K > 0.9 across all tissues, 111 events were selected. For ENCODE: N > 0.7, K > 0.7 across all experiments, L > 0.2, 127 events were selected. Experiments in which less than 70% of core USEs met these criteria were excluded from further analysis.

Survival analysis

Patient survival data were obtained from the UCSC Xena platform [39]. The association between NMD efficiency and the overall survival was assessed using the Cox proportional hazards model. Within each tumor cohort, patients were stratified into two equal groups, with high and low NMD efficiency. Sex and age were included as covariates; age was divided into three equal-sized groups within each cohort.

NMD efficiency and regulated unproductive splicing

The following odds ratios were calculated to assess the abundance of specifically regulated USEs. For a given threshold value of the absolute ΔΨ value in tumors, the ratio of the number of USEs demonstrating opposite directions of splicing and expression changes and the number of USEs with co-directional changes among events with |ΔΨ| above and below the threshold value were compared. Only the genes with |log2FC| > 0.3 and P < 0.05 were considered. The same estimates were calculated separately for positive and negative ΔΨ values to assess the dependence on the direction of changes in NMD efficiency.

Statistical analysis

All statistical tests were performed in Python 3.9.13. Bonferroni–Holm corrections for multiple testing were performed for differential expression and splicing analyses in TCGA and for RBP or NMD factor inactivation experiments separately for each tumor cohort or experiment, when assessing the significance of differences in NMD efficiency between tumor and normal tissue, when searching for associations between NMD efficiency and patient survival, and when assessing the enrichment of functional categories of RBPs using the GSEA method in a list of RBPs ranked by the magnitude of changes in NMD efficiency in the corresponding knockdown. Survival analysis was conducted using the lifelines module. Boxplot whiskers in all the figures correspond to the extreme points deviating from the median by no more than 1.5 interquartile range units. Error bars in Figs. 1 and 4, as well as Supplementary Fig. S3, correspond to 95% confidence intervals. In this paper, r and P denote the Pearson correlation coefficient and the adjusted P value, respectively; FC denotes the fold change in the gene expression levels. All nonparametric tests were performed using normal approximation with a continuity correction.

RESULTS

Unproductive splicing changes exhibit a preferred cancer-specific direction

Our initial goal was to characterize global unproductive splicing changes in tumors. Using the NMDj tool [35], we assembled a catalog of unproductive splicing events (USE) that were either present in the ENSEMBL genome annotation (annotated) or inferred from the RNA-seq data for tumor transcriptomes (novel). We identified 1,986 annotated and 4,078 novel USEs responding significantly (P < 0.05) to NMD inhibition in the SMG6-SMG7 double knockdown experiments [29]. As expected, the relative abundance of the NMD isoform, as measured by the change in the percent-spliced-in (ΔΨ, NMD inhibition vs. control) metric, increased for the vast majority of events (more than 75%) (Fig. 1A).

 

Fig. 1. Changes in unproductive splicing upon NMD inactivation and in cancer cohorts from TCGA. (A) Changes in the splicing rates (ΔΨ) of USE in double knockdown of SMG6 and SMG7 in the HeLa cell line. (B) Changes in the splicing rates (ΔΨ) of USEs that are differentially spliced in at least one TCGA cohort. (C) The proportions of USEs with elevated NMD isoforms in tumor cohorts (ΔΨ > 0). Numbers to the right of the bars indicate the number of USEs. The dotted lines show the average proportions of USEs with elevated NMD isoforms across all cohorts. Error bars represent 95% confidence intervals. The blue, red, and gray colors represent tumor groups with the same direction of splicing alterations

 

USEs that had changed in the expected direction (i.e., increased the proportion of NMD isoform under NMD inactivation) were selected for further analysis. We found 1,187 annotated and 2,168 novel events that were differentially spliced in at least one tumor cohort (P < 0.05), with remarkably diverse patterns of both positive and negative changes (Fig. 1B). Next, we quantified the proportion of annotated and novel USEs with positive ΔΨ and tested statistically whether they differed from the respective average ΔΨ value across all cancer cohorts (Fig. 1C). Interestingly, most cancer types exhibited a preferred direction of unproductive splicing change: in PRAD, BRCA, KICH, and THCA, the proportion of NMD isoforms tended to decrease, whereas in BLCA, COAD, LIHC, STAD, HNSC, and KIRC it more often increased. This trend was observed for both the annotated and novel events; however, novel events showed a higher proportion of positive changes, which is consistent with their absence in genome annotations because of the low expression levels under normal conditions.

USE-based metric robustly estimates the NMD efficiency

The consistency in the direction of changes in unproductive splicing across tumors suggests that these alterations may be driven by global changes in the NMD pathway activity. To estimate NMD efficiency, we selected a core set of n = 120 USEs that (i) were expressed in at least 13 of the 14 analyzed TCGA cohorts, (ii) responded to NMD inhibition in the expected direction, i.e., by upregulation of NMD-sensitive transcripts, and (iii) were not sensitive to perturbations in the expression levels of RBP (in order to reduce the confounding effect of event-specific regulation, see Methods). For each sample, the NMD efficiency score was defined as the negative mean of z-transformed Ψ values of the selected core events.

We validated this metric in several ways. First, the NMD efficiency scores were robust with respect to choosing a core set of events provided that it was large enough (n > 100) (Supplementary Fig. S1, see “Experimental”). Second, all the USEs in the core set showed highly coordinated splicing changes across TCGA samples (Supplementary Fig. S2). Third, the NMD efficiency score dropped in the experiments involving single and double knockdown of three key NMD factors (UPF1, SMG6, and SMG7), and recovered in the respective rescue experiments (Fig. 2A). Furthermore, changes in the NMD efficiency score in tumors vs. matched normal tissues positively correlated with changes in UPF1 expression levels (Fig. 2B). As expected, the proportion of USEs with collective positive ΔΨ changes in TCGA samples dropped from 80% to 20% with increasing NMD efficiency score (Fig. 2C). Finally, applying our metric to healthy tissues from the GTEx project revealed a strong correlation with an existing allele-specific estimate of NMD efficiency [23], in which the latter explained 62% of the variance in the former if brain samples were excluded (Fig. 2D). Notably, brain tissues deviated from the general linear trend in such a way that allele-specific estimates for different brain subregions were highly variable, while the NMD efficiency score introduced here was consistently high.

 

Fig. 2. The estimation of NMD efficiency. (A) NMD efficiency upon knockdown and rescue of the expression of NMD factors in the HeLa cell line. (B) Correlation between changes in NMD efficiency and changes in UPF1 expression levels in tumors. (C) Correlation between changes in NMD efficiency and proportion of USEs with changes in positive splicing in tumors. Dot colors correspond to the colors of the tumor groups with the same direction of splicing changes in Fig. 1C. (D) Correlation between the NMD efficiency score in GTEx tissues from this study and the score based on nonsense mutations [23]

 

NMD efficiency varies substantially both between and within human tissues

While the existing NMD efficiency metrics allow one to perform relative comparisons across samples, they do not provide a sense of its absolute magnitude [23, 24, 26]. To address this issue, we introduced a reference scale by comparing the observed variation in NMD efficiency to the effect size for the knockdown of UPF1, the core factor of NMD, for which the expression level dropped by a factor of 3.5 [29]. With this scaling, the NMD efficiency score under unperturbed conditions was assumed to be 0, while it was assumed to be –1 under UPF1 depletion.

As applied to the GTEx dataset, this transformed metric confirmed that NMD efficiency varies across tissues in a nonrandom manner, which is in line with recent findings [26] (Fig. 3). It is worth mentioning that 46% of the variance was attributed to the tissue factor (one-way ANOVA, P ≃ 0), and the difference between tissues with the highest (heart) and lowest (cerebellum) median NMD efficiency score was comparable to 80% of the effect of UPF1 knockdown. The inter-tissue variability was also remarkably high, with interquartile ranges reaching up to 40% of the effect of UPF1 knockdown in the heart, uterus, liver, and kidney. The least variable NMD efficiency levels were observed in brain subregions, but the difference between the median NMD efficiency levels in the brain cortex and the cerebellum was approximately half of the effect of the UPF1 knockdown, confirming that these two areas differ drastically in terms of NMD activity [26].

 

Fig. 3. The distribution of NMD efficiencies in GTEx tissues relative to the magnitude of the effect of UPF1 knockdown. NMD efficiencies in the control and UPF1 knockdown HeLa cells were set to 0 and -1, respectively

 

NMD efficiency is associated with patient survival and regulated unproductive splicing

In order to examine the variability in NMD efficiency across tumors, we analyzed TCGA cohorts and found significant changes between NMD activity in cancer and the respective normal tissue in BRCA, PRAD, KICH, and KIRC (Fig. 4A). Remarkably, the direction of changes in NMD efficiency in cancer depended on the baseline efficiency of the corresponding normal tissue: it tends to increase in tissues with a low baseline NMD efficiency, while tending to decrease in tissues with a high baseline NMD efficiency. This led us to a nontrivial conclusion that tumors show a consistent trend toward erasing the tissue-specific NMD efficiency signature.

 

Fig. 4. Association of NMD efficiency with patient survival and regulated unproductive splicing. (A) The least squares regression of the mean NMD efficiency in tumor versus that in normal tissue. Error bars represent standard errors of the mean. The dashed line corresponds to y = x. Tumors with a statistically significant difference in NMD efficiency are highlighted (P < 0.05, Wilcoxon test). (B) The Kaplan–Meier curves for patient groups with high and low NMD efficiency in tumors, for bladder carcinoma (BLCA) and clear cell renal cell carcinoma (KIRC). (C) Odds ratio of the oppositely directed changes in splicing and gene expression as a function of the threshold for the |ΔΨ| value and its sign. Asterisks denote a significant (P < 0.05) difference in odds ratios for positive and negative splicing changes

 

We next assessed the clinical relevance of alterations in NMD efficiency by relating it to patient survival. Significant associations with the overall survival rate were observed in BLCA and KIRC (Fig. 4B). Notably, the direction of these associations indicated the baseline NMD efficiency levels of normal tissue. In BLCA, which originally has low NMD efficiency, elevated NMD efficiency in tumors correlated with a poorer prognosis. Conversely, in KIRC, which exhibits high baseline NMD efficiency, reduction of its activity was associated with unfavorable outcomes (Fig. 4B). Together, these observations indicate that the greater the deviation of tumor NMD efficiency from the tissue-specific baseline, the worse the clinical prognosis; this trend persists for both positive and negative deviations.

As shown earlier, a substantial proportion of cancer-associated changes in unproductive splicing can be attributed to global shifts in NMD efficiency but some events may be regulated in a specific way. Event-specific regulation is expected to produce opposite changes in splicing and in the expression level of the host gene [40]. Using this negative association, we found that the proportion of specifically regulated USEs increased as we applied more stringent thresholds on the magnitude and statistical significance of splicing alterations (Supplementary Fig. S3). Moreover, these regulated USEs were dominant among the events that shift against the global direction imposed by the changes in NMD efficiency. For example, in BRCA, the cancer type with the strongest increase in NMD efficiency, the regulated USEs were significantly enriched among positive splicing changes, whereas an opposite pattern was observed in KIRC, which exhibited the strongest decline in NMD efficiency (Fig. 4C).

The NMD efficiency metric recovers known and suggests novel regulators

The core components of the NMD pathway including the UPF proteins (UPF1, UPF2, UPF3), the SMG kinases and adaptor proteins (SMG1, SMG5, SMG6, SMG7), and the exon junction complex (EIF4A3, RBM8A, MAGOH) have been characterized in considerable detail [41–43]. Yet, several recent studies suggest that the regulatory landscape of NMD extends beyond these canonical components [44, 45]. To explore this broader scope, we applied the NMD efficiency metric to a compendium of 587 RBP knockdown and knockout experiments and put forward a question as to which perturbations globally alter the NMD [28, 46].

Most perturbations reduced NMD efficiency, with exon junction complex components strongly being enriched among those with the largest effects (Fig. 5A). Decreased NMD efficiency was also frequently observed upon depletion of spliceosomal factors, which is consistent with the observations implicating spliceosomal components in the NMD control [45]. The pervasive tendency toward lower NMD efficiency upon depletion of almost any RBP is possibly indicative of the high interconnectivity of the RBP regulatory network, in which perturbing a single RBP can induce a cascade effect causing global deregulation of RNA processing and accumulation of NMD-sensitive transcripts that are normally suppressed by both NMD and splicing regulatory programs [47].

 

Fig. 5. NMD efficiency in RBP inactivation experiments. (A) The gene set enrichment analysis (GSEA) plot for RBP ranked by the magnitude of changes in NMD efficiency in expression inactivation experiments. (B) Correlation between the changes in NMD efficiency and the similarity of transcriptional profiles upon RBP depletion and NMD inactivation. The latter was assessed using the Pearson correlation coefficient applied to the log2 fold change of the expression values (log2FC) of differentially expressed genes in the RBP inactivation experiment and the NMD inactivation experiment (average over four NMD inactivation experiments). (C) Heat map of log2FC for RBP genes upon knockdowns of the AGO3 and NMD factors. (D) A part of the heat map from (B) corresponding to changes in the expression of the AGO1, AGO3, and NMD factors. Statistically insignificant changes are not shown

 

In order to be able to further characterize potential NMD regulators, we compared the gene expression changes that were induced by NMD inactivation with those observed after depletion of each RBP. Specifically, we related two quantities: (1) the changes in the NMD efficiency score under RBP depletion and (2) the similarity of transcriptional profiles upon RBP depletion and NMD inactivation (the average of four NMD inactivation experiments: UPF1, SMG6, SMG7 and SMG6 + SMG7). The latter was measured by the Pearson correlation coefficient of log2 fold change values of the expression of differentially expressed genes. RBPs whose inactivation reduced NMD efficiency tended to drive expression changes resembling those caused by NMD loss per se (Fig.5B). SUPV3L1, HNRNPC, and EIF3A stood out among the RBPs not previously linked to NMD as the closest to bona fide NMD factors both by the NMD efficiency metric and by their transcriptomic signatures.

SUPV3L1 is a helicase best known for its role in mitochondrial RNA turnover, although small amounts of this protein also localize in the nucleus [48]. Our findings suggest that SUPV3L1 may also contribute to degradation of nuclear RNAs, since the core USEs events mostly take place in nuclear genes, which substantially respond to SUPV3L1 knockdown. HNRNPC binds poly-U tracts in mRNAs and is involved in alternative splicing, polyadenylation, and translation [49–52]. HNRNPC is known to suppress the incorporation of numerous cryptic exons [53, 54]. Therefore, its impact on unproductive splicing may arise from event-specific regulation rather than direct involvement in NMD. EIF3A is the RNA-binding subunit of the eIF3 translation initiation complex, which was shown to interact physically with UPF1 and inhibit subsequent rounds of translation [55]. Knockdown of EIF3E, another eIF3 subunit, was previously shown to suppress NMD [56, 57]. Taken together, these observations point to reciprocal regulation between translation initiation and NMD.

We also identified putative negative regulators of NMD, including AGO3, whose depletion produced the strongest increase in NMD efficiency across all the tested RBPs. AGO3 is one of the four Argonaute genes involved in miRNA-guided translational repression [58]. Reporter assays mimicking RISC binding in 3’UTRs of NMD-sensitive transcripts previously showed that its paralog, AGO2, can inhibit NMD by suppressing the translation of NMD-sensitive transcripts, although it remains unclear to what extent it affects endogenous targets [59]. Our data indicate that AGO3 suppresses the degradation of a broad set of endogenous transcripts, leading to gene expression changes opposite to those caused by NMD inhibition (Fig. 5C). However, other components of the miRNA pathway (AGO1, DICER1, DROSHA, XPO5, and DGCR8) did not show similar effects, suggesting that the role played by AGO3 is not simply a consequence of canonical miRNA pathway activity. Notably, AGO3 knockdown---but not AGO1 knockdown---increased the expression of SMG7, which may partly account for the observed increase in NMD efficiency (Fig. 5D). These findings point to AGO3 as a candidate negative regulator of NMD and suggest that it may possess functions that are distinct from those of its paralogs.

DISCUSSION AND CONCLUSIONS

Accurate assessment of the efficiency of the NMD pathway is essential for understanding its contribution to cancer biology. The expression levels of NMD factors are unsuitable as proxies for pathway activity, since core NMD components are subject to negative autoregulation [21]. Existing approaches for estimating NMD efficiency include reporter assays (e.g., globin reporters) [25], analysis of allele-specific expression in genes with heterozygous nonsense mutations [23, 24], and comparison of the expression levels of NMD-sensitive and productive transcripts [26]. Each method has its limitations: reporter assays are low-throughput, allele-specific approaches rely on rare mutations that may be insufficiently represented in a sample [23, 26], and transcript-level quantification of NMD targets can be inaccurate for minor isoforms and sensitive to annotation errors [60].

In this study, we have developed an NMD efficiency metric based on the splicing rates in a core set of USEs. We used it to characterize NMD regulation across human tissues, cancers, and large-scale RBP perturbation datasets. Leveraging USEs as endogenous reporters of NMD provides a powerful alternative to existing transcript- or mutation-based metrics. Using this framework, we confirmed that NMD efficiency varies across human tissues in a nonrandom manner, quantified the magnitude of these differences relative to NMD inhibition experiments, revealed systematic shifts in NMD activity in tumors, and identified putative novel regulators of the NMD pathway.

An unexpected observation emerging from our analyses and previous reports is the substantial heterogeneity of NMD efficiency within individual tissues [26]. The main factor contributing to this variability is that tissues in GTEx and TCGA are mixtures of multiple cell types varying in their morphology, histology, and characteristic expression signatures [61]. Because NMD efficiency can vary between the individual cell types making up a tissue [62], the NMD efficiency metric is indicative not only of the tissue-specific NMD activity, but also its histological composition. Overall, these results emphasize the importance of considering tissue heterogeneity when interpreting any NMD efficiency estimates from RNA-seq.

Although tissue rankings by NMD efficiency produced by our metric were consistent with those from refs. [23, 26], our metric assigned higher NMD efficiency estimates for most brain regions (except Cerebellum, Anterior cingulate cortex (BA24), Hypothalamus and Spinal cord cervical C-1). Transcriptional programs of brain cells, particularly neurons, are known to be tightly regulated by unproductive splicing [63–66], which may confound NMD efficiency estimates, if a substantial portion of core observations (USEs, mutations or transcript pairs) are specifically regulated. Indeed, we observed two clusters among core USEs, the larger one consisting of 91 USEs with lower Ψ values in brain samples; and the smaller, with 20 USEs having discordantly high Ψ values in brain, yet being concordant with the majority of core USEs in samples from other tissues (Supplementary Fig. S4). Since our metric puts more strict requirements on the number of core USEs per sample than those reported in refs. [23] and [26], with median 108 core USEs, compared to not more than 50 transcript pairs [26] and three stop gain variants [23] per sample in GTEx, it may provide a more realistic and conservative view of NMD efficiency in the brain.

In multiple cancer types, we observed a striking divergence of NMD efficiency from the tissue-specific baseline. Tissues with intrinsically low NMD efficiency (e.g., breast) tend to exhibit an increased NMD efficiency in tumors, whereas those with high baseline activity (e.g., kidney) often show reduced NMD efficiency. This “flattening” of the NMD profiles suggests that tumors partially erase the NMD signature of their tissue of origin. A possible explanation is that cancer cells acquire stem-like properties [67]. Stem and progenitor cells display NMD efficiencies distinct from those of mature, differentiated cells under many conditions [4, 68, 69], and tumor dedifferentiation may shift NMD toward these stem-associated states. Alternatively, tumor evolution imposes selective pressures such as chronic stress, altered translation demands, or immune system evasion, which may favor specific levels of NMD independent of the original tissue program. Therefore, deviations from the tissue-specific NMD level were associated with patient survival, suggesting that maintenance of physiological NMD activity may be unfavorable for tumor progression.

Together with the observed pattern of co-directional changes in unproductive splicing, our findings attest to the substantial deregulation of the NMD pathway in tumors, which undoubtedly contributes to cancer pathology.

The authors declare no competing interests.

This work was supported by the Russian Science Foundation grant No. 22-14-00330-P.

Supplementary Materials are available at https://doi.org/10.32607/actanaturae.27892

×

About the authors

L. G. Zavileyskiy

Center for Molecular and Cellular Biology

Author for correspondence.
Email: zavileyskiylev@gmail.com
Russian Federation, Moscow, 121205

A. A. Mironov

Center for Molecular and Cellular Biology; M.V. Lomonosov Moscow State University

Email: zavileyskiylev@gmail.com

Faculty of Bioengineering and Bioinformatics

Russian Federation, Moscow, 121205; Moscow, 119991

D. D Pervouchine

Center for Molecular and Cellular Biology

Email: zavileyskiylev@gmail.com
Russian Federation, Moscow, 121205

References

  1. He F, Jacobson A. Nonsense-mediated mRNA decay: degradation of defective transcripts is only part of the story. Annu Rev Genet. 2015;49:339-366. doi: 10.1146/annurev-genet-112414-054639
  2. Lewis BP, Green RE, Brenner SE. Evidence for the widespread coupling of alternative splicing and nonsense-mediated mRNA decay in humans. Proc Natl Acad Sci U S A. 2003;100(1):189-192. doi: 10.1073/pnas.0136770100
  3. Lareau LF, Inada M, Green RE, Wengrod JC, Brenner SE. Unproductive splicing of SR genes associated with highly conserved and ultraconserved DNA elements. Nature. 2007;446(7138):926-929. doi: 10.1038/nature05676
  4. Weischenfeldt J, Damgaard I, Bryder D, et al. NMD is essential for hematopoietic stem and progenitor cells and for eliminating by-products of programmed DNA rearrangements. Genes Dev. 2008;22(10):1381-1396. doi: 10.1101/gad.468808
  5. Gardner LB. Hypoxic inhibition of nonsense-mediated RNA decay regulates gene expression and the integrated stress response. Mol Cell Biol. 2008;28(11):3729-3741. doi: 10.1128/MCB.02284-07
  6. Goetz AE, Wilkinson M. Stress and the nonsense-mediated RNA decay pathway. Cell Mol Life Sci. 2017;74(19):3509-3531. doi: 10.1007/s00018-017-2537-6
  7. Khajavi M, Inoue K, Lupski JR. Nonsense-mediated mRNA decay modulates clinical outcome of genetic disease. Eur J Hum Genet. 2006;14(10):1074-1081. doi: 10.1038/sj.ejhg.5201649
  8. Popp MW, Maquat LE. Nonsense-mediated mRNA decay and cancer. Curr Opin Genet Dev. 2018;48:44-50. doi: 10.1016/j.gde.2017.10.007
  9. Nagar P, Islam MR, Rahman MA. Nonsense-mediated mRNA decay as a mediator of tumorigenesis. Genes (Basel). 2023;14(2):357. doi: 10.3390/genes14020357
  10. Nogueira G, Fernandes R, Garcı́a-Moreno JF, Romão L. Nonsense-mediated RNA decay and its bipolar function in cancer. Mol Cancer. 2021;20(1):72. doi: 10.1186/s12943-021-01364-0
  11. Lindeboom RG, Supek F, Lehner B. The rules and impact of nonsense-mediated mRNA decay in human cancers. Nat Genet. 2016;48(10):1112-1118. doi: 10.1038/ng.3664
  12. Wang D, Zavadil J, Martin L, et al. Inhibition of nonsense-mediated RNA decay by the tumor microenvironment promotes tumorigenesis. Mol Cell Biol. 2011;31(17):3670-3680. doi: 10.1128/MCB.05704-11
  13. Anczuków O, Ware MD, Buisson M, et al. Does the nonsense-mediated mRNA decay mechanism prevent the synthesis of truncated BRCA1, CHK2, and p53 proteins? Hum Mutat. 2008;29(1):65-73. doi: 10.1002/humu.20590
  14. Fan S, Yuan R, Ma YX, Meng Q, Goldberg ID, Rosen EM. Mutant BRCA1 genes antagonize phenotype of wild-type BRCA1. Oncogene. 2001;20(57):8215-8235. doi: 10.1038/sj.onc.1205033
  15. Bokhari A, Jonchere V, Lagrange A, et al. Targeting nonsense-mediated mRNA decay in colorectal cancers with microsatellite instability. Oncogenesis. 2018;7(9):70. doi: 10.1038/s41389-018-0079-x
  16. Liu C, Karam R, Zhou Y, et al. The UPF1 RNA surveillance gene is commonly mutated in pancreatic adenosquamous carcinoma. Nat Med. 2014;20(6):596-598. doi: 10.1038/nm.3548
  17. Staszewski J, Lazarewicz N, Konczak J, Migdal I, Maciaszczyk-Dziubinska E. UPF1-from mRNA degradation to human disorders. Cells. 2023;12(3):419. doi: 10.3390/cells12030419
  18. Li L, Geng Y, Feng R, et al. The human RNA surveillance factor UPF1 modulates gastric cancer progression by targeting long non-coding RNA MALAT1. Cell Physiol Biochem. 2017;42(6):2194-2206. doi: 10.1159/000479994
  19. Chang L, Li C, Guo T, et al. The human RNA surveillance factor UPF1 regulates tumorigenesis by targeting Smad7 in hepatocellular carcinoma. J Exp Clin Cancer Res. 2016;35:8. doi: 10.1186/s13046-016-0286-2
  20. Chen B, Wang H, Li D, Lin X, Ma Z, Zeng Y. Up-frameshift protein 1 promotes tumor progression by regulating apoptosis and epithelial-mesenchymal transition of colorectal cancer. Technol Cancer Res Treat. 2021;20:15330338211064438. doi: 10.1177/15330338211064438
  21. Yepiskoposyan H, Aeschimann F, Nilsson D, Okoniewski M, Mühlemann O. Autoregulation of the nonsense-mediated mRNA decay pathway in human cells. RNA. 2011;17(12):2108-2118. doi: 10.1261/rna.030247.111
  22. Zetoune AB, Fontanière S, Magnin D, et al. Comparison of nonsense-mediated mRNA decay efficiency in various murine tissues. BMC Genet. 2008;9:83. doi: 10.1186/1471-2156-9-83
  23. Teran NA, Nachun DC, Eulalio T, et al. Nonsense-mediated decay is highly stable across individuals and tissues. Am J Hum Genet. 2021;108(8):1401-1408. doi: 10.1016/j.ajhg.2021.06.008
  24. Kim YG, Kang H, Lee B, et al. A spectrum of nonsense-mediated mRNA decay efficiency along the degree of mutational constraint. Commun Biol. 2024;7(1):1461. doi: 10.1038/s42003-024-07136-y
  25. Sato H, Singer RH. Cellular variability of nonsense-mediated mRNA decay. Nat Commun. 2021;12(1):7203. doi: 10.1038/s41467-021-27423-0
  26. Palou-Márquez G, Supek F. Variable efficiency of nonsense-mediated mRNA decay across human tissues, tumors and individuals. Genome Biol. 2025;26(1):316. doi: 10.1186/s13059-025-03727-y
  27. Van Nostrand EL, Freese P, Pratt GA, et al. A large-scale binding and functional map of human RNA-binding proteins. Nature. 2020;583(7818):711-719. doi: 10.1038/s41586-020-2077-3
  28. Rogalska ME, Mancini E, Bonnal S, et al. Transcriptome-wide splicing network reveals specialized regulatory functions of the core spliceosome. Science. 2024;386(6721):551-560. doi: 10.1126/science.adn8105
  29. Colombo M, Karousis ED, Bourquin J, Bruggmann R, Mühlemann O. Transcriptome-wide identification of NMD-targeted human mRNAs reveals extensive redundancy between SMG6- and SMG7-mediated degradation pathways. RNA. 2017;23(2):189-201. doi: 10.1261/rna.059055.116
  30. Dobin A, Davis CA, Schlesinger F, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15-21. doi: 10.1093/bioinformatics/bts635
  31. Frankish A, Diekhans M, Jungreis I, et al. GENCODE 2021. Nucleic Acids Res. 2021;49(D1):D916-D923. doi: 10.1093/nar/gkaa1087
  32. Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics. 2014;30(7):923-930. doi: 10.1093/bioinformatics/btt656
  33. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010;26(1):139-140. doi: 10.1093/bioinformatics/btp616
  34. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550. doi: 10.1186/s13059-014-0550-8
  35. Zavileyskiy LG, Chernyavskaya EA, Vlasenok MA, Pervouchine DD. Classification and quantification of unproductive splicing events. Acta Naturae. 2025;17(2):75-85. doi: 10.32607/actanaturae.27572
  36. Pertea M, Pertea GM, Antonescu CM, Chang TC, Mendell JT, Salzberg SL. StringTie enables improved reconstruction of a transcriptome from RNA-seq reads. Nat Biotechnol. 2015;33(3):290-295. doi: 10.1038/nbt.3122
  37. Pervouchine DD, Knowles DG, Guigó R. Intron-centric estimation of alternative splicing from RNA-seq data. Bioinformatics. 2013;29(2):273-274. doi: 10.1093/bioinformatics/bts678
  38. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7(3):177-188. doi: 10.1016/0197-2456(86)90046-2
  39. Goldman MJ, Craft B, Hastie M, et al. Visualizing and interpreting cancer genomics data via the Xena platform. Nat Biotechnol. 2020;38(6):675-678. doi: 10.1038/s41587-020-0546-8
  40. Mironov A, Petrova M, Margasyuk S, et al. Tissue-specific regulation of gene expression via unproductive splicing. Nucleic Acids Res. 2023;51(7):3055-3066. doi: 10.1093/nar/gkad161
  41. Lykke-Andersen S, Jensen TH. Nonsense-mediated mRNA decay: an intricate machinery that shapes transcriptomes. Nat Rev Mol Cell Biol. 2015;16(11):665-677. doi: 10.1038/nrm4063
  42. Kim YK, Maquat LE. UPFront and center in RNA decay: UPF1 in nonsense-mediated mRNA decay and beyond. RNA. 2019;25(4):407-422. doi: 10.1261/rna.070136.118
  43. Karousis ED, Nasif S, Mühlemann O. Nonsense-mediated mRNA decay: novel mechanistic insights and biological impact. Wiley Interdiscip Rev RNA. 2016;7(5):661-682. doi: 10.1002/wrna.1357
  44. Cheruiyot A, Li S, Nonavinkere Srivatsan S, et al. Nonsense-mediated RNA decay is a unique vulnerability of cancer cells harboring SF3B1 or U2AF1 mutations. Cancer Res. 2021;81(17):4499-4513. doi: 10.1158/0008-5472.CAN-20-4016
  45. Embree CM, Paul D, Stephanou A, Singh G. Direct and indirect effects of spliceosome disruption compromise gene regulation by nonsense-mediated mRNA decay. RNA Biol. 2025;22(1):1-26. doi: 10.1080/15476286.2025.2552517
  46. Kagda MS, Lam B, Litton C, et al. Data navigation on the ENCODE portal. Nat Commun. 2025;16(1):9592. doi: 10.1038/s41467-025-64343-9
  47. Quattrone A, Dassi E. The architecture of the human RNA-binding protein regulatory network. iScience. 2019;21:706-719. doi: 10.1016/j.isci.2019.10.058
  48. Szewczyk M, Fedoryszak-Kuśka N, Tkaczuk K, Dobrucki J, Waligórska A, Stepień PP. Human SUV3 helicase regulates growth rate of the HeLa cells and can localize in the nucleoli. Acta Biochim Pol. 2017;64(1):177-181. doi: 10.18388/abp.2016_1419
  49. Rong S, Dai B, Yang C, et al. HNRNPC modulates PKM alternative splicing via m6A methylation, upregulating PKM2 expression to promote aerobic glycolysis in papillary thyroid carcinoma and drive malignant progression. J Transl Med. 2024;22(1):914. doi: 10.1186/s12967-024-05668-9
  50. Kim JH, Paek KY, Choi K, et al. Heterogeneous nuclear ribonucleoprotein C modulates translation of c-myc mRNA in a cell cycle phase-dependent manner. Mol Cell Biol. 2003;23(2):708-720. doi: 10.1128/MCB.23.2.708-720.2003
  51. Shetty S. Regulation of urokinase receptor mRNA stability by hnRNP C in lung epithelial cells. Mol Cell Biochem. 2005;272(1-2):107-118. doi: 10.1007/s11010-005-7644-2
  52. Fischl H, Neve J, Wang Z, et al. hnRNPC regulates cancer-specific alternative cleavage and polyadenylation profiles. Nucleic Acids Res. 2019;47(14):7580-7591. doi: 10.1093/nar/gkz461
  53. Zarnack K, König J, Tajnik M, et al. Direct competition between hnRNP C and U2AF65 protects the transcriptome from the exonization of Alu elements. Cell. 2013;152(3):453-466. doi: 10.1016/j.cell.2012.12.023
  54. Attig J, Ruiz de Los Mozos I, Haberman N, et al. Splicing repression allows the gradual emergence of new Alu-exons in primate evolution. Elife. 2016;5:e19545. doi: 10.7554/eLife.19545
  55. Isken O, Kim YK, Hosoda N, Mayeur GL, Hershey JW, Maquat LE. Upf1 phosphorylation triggers translational repression during nonsense-mediated mRNA decay. Cell. 2008;133(2):314-327. doi: 10.1016/j.cell.2008.02.030
  56. Morris C, Wittmann J, Jäck HM, Jalinot P. Human INT6/eIF3e is required for nonsense-mediated mRNA decay. EMBO Rep. 2007;8(6):596-602. doi: 10.1038/sj.embor.7400955
  57. Peixeiro I, Inácio Â, Barbosa C, Silva AL, Liebhaber SA, Romão L. Interaction of PABPC1 with the translation initiation complex is critical to the NMD resistance of AUG-proximal nonsense mutations. Nucleic Acids Res. 2012;40(3):1160-1173. doi: 10.1093/nar/gkr820
  58. Hutvagner G, Simard MJ. Argonaute proteins: key players in RNA silencing. Nat Rev Mol Cell Biol. 2008;9(1):22-32. doi: 10.1038/nrm2321
  59. Choe J, Cho H, Lee HC, Kim YK. microRNA/Argonaute 2 regulates nonsense-mediated messenger RNA decay. EMBO Rep. 2010;11(5):380-386. doi: 10.1038/embor.2010.44
  60. Sarantopoulou D, Brooks TG, Nayak S, Mrčela A, Lahens NF, Grant GR. Comparative evaluation of full-length isoform quantification from RNA-Seq. BMC Bioinformatics. 2021;22(1):266. doi: 10.1186/s12859-021-04198-1
  61. Breschi A, Muñoz-Aguirre M, Wucher V, et al. A limited set of transcriptional programs define major cell types. Genome Res. 2020;30(7):1047-1059. doi: 10.1101/gr.263186.120
  62. Linde L, Boelz S, Neu-Yilik G, Kulozik AE, Kerem B. The efficiency of nonsense-mediated mRNA decay is an inherent character and varies among different cells. Eur J Hum Genet. 2007;15(11):1156-1162. doi: 10.1038/sj.ejhg.5201889
  63. Zhuravskaya A, Yap K, Hamid F, Makeyev EV. Alternative splicing coupled to nonsense-mediated decay coordinates downregulation of non-neuronal genes in developing mouse neurons. Genome Biol. 2024;25(1):162. doi: 10.1186/s13059-024-03305-8
  64. Jaffrey SR, Wilkinson MF. Nonsense-mediated RNA decay in the brain: emerging modulator of neural development and disease. Nat Rev Neurosci. 2018;19(12):715-728. doi: 10.1038/s41583-018-0079-z
  65. Vuong JK, Ergin V, Chen L, Zheng S. Multilayered regulations of alternative splicing, NMD, and protein stability control temporal induction and tissue-specific expression of TRIM46 during axon formation. Nat Commun. 2022;13(1):2081. doi: 10.1038/s41467-022-29786-4
  66. Lee PJ, Yang S, Sun Y, Guo JU. Regulation of nonsense-mediated mRNA decay in neural development and disease. J Mol Cell Biol. 2021;13(4):269-281. doi: 10.1093/jmcb/mjab022
  67. Yadav AK, Desai NS. Cancer stem cells: acquisition, characteristics, therapeutic implications, targeting strategies and future prospects. Stem Cell Rev Rep. 2019;15(3):331-355. doi: 10.1007/s12015-019-09887-2
  68. Li T, Shi Y, Wang P, et al. Smg6/Est1 licenses embryonic stem cell differentiation via nonsense-mediated mRNA decay. EMBO J. 2015;34(12):1630-1647. doi: 10.15252/embj.201489947
  69. Lou CH, Chousal J, Goetz A, et al. Nonsense-mediated RNA decay influences human embryonic stem cell fate. Stem Cell Reports. 2016;6(6):844-857. doi: 10.1016/j.stemcr.2016.05.008

Supplementary files

Supplementary Files
Action
1. JATS XML
Download
2. Fig. 1. Changes in unproductive splicing upon NMD inactivation and in cancer cohorts from TCGA. (A) Changes in the splicing rates (ΔΨ) of USE in double knockdown of SMG6 and SMG7 in the HeLa cell line. (B) Changes in the splicing rates (ΔΨ) of USEs that are differentially spliced in at least one TCGA cohort. (C) The proportions of USEs with elevated NMD isoforms in tumor cohorts (ΔΨ > 0). Numbers to the right of the bars indicate the number of USEs. The dotted lines show the average proportions of USEs with elevated NMD isoforms across all cohorts. Error bars represent 95% confidence intervals. The blue, red, and gray colors represent tumor groups with the same direction of splicing alterations

Download (448KB)
Indexing metadata
3. Fig. 2. The estimation of NMD efficiency. (A) NMD efficiency upon knockdown and rescue of the expression of NMD factors in the HeLa cell line. (B) Correlation between changes in NMD efficiency and changes in UPF1 expression levels in tumors. (C) Correlation between changes in NMD efficiency and proportion of USEs with changes in positive splicing in tumors. Dot colors correspond to the colors of the tumor groups with the same direction of splicing changes in Fig. 1C. (D) Correlation between the NMD efficiency score in GTEx tissues from this study and the score based on nonsense mutations [23]

Download (333KB)
Indexing metadata
4. Fig. 3. The distribution of NMD efficiencies in GTEx tissues relative to the magnitude of the effect of UPF1 knockdown. NMD efficiencies in the control and UPF1 knockdown HeLa cells were set to 0 and -1, respectively

Download (458KB)
Indexing metadata
5. Fig. 4. Association of NMD efficiency with patient survival and regulated unproductive splicing. (A) The least squares regression of the mean NMD efficiency in tumor versus that in normal tissue. Error bars represent standard errors of the mean. The dashed line corresponds to y = x. Tumors with a statistically significant difference in NMD efficiency are highlighted (P < 0.05, Wilcoxon test). (B) The Kaplan–Meier curves for patient groups with high and low NMD efficiency in tumors, for bladder carcinoma (BLCA) and clear cell renal cell carcinoma (KIRC). (C) Odds ratio of the oppositely directed changes in splicing and gene expression as a function of the threshold for the |ΔΨ| value and its sign. Asterisks denote a significant (P < 0.05) difference in odds ratios for positive and negative splicing changes

Download (439KB)
Indexing metadata
6. Fig. 5. NMD efficiency in RBP inactivation experiments. (A) The gene set enrichment analysis (GSEA) plot for RBP ranked by the magnitude of changes in NMD efficiency in expression inactivation experiments. (B) Correlation between the changes in NMD efficiency and the similarity of transcriptional profiles upon RBP depletion and NMD inactivation. The latter was assessed using the Pearson correlation coefficient applied to the log2 fold change of the expression values (log2FC) of differentially expressed genes in the RBP inactivation experiment and the NMD inactivation experiment (average over four NMD inactivation experiments). (C) Heat map of log2FC for RBP genes upon knockdowns of the AGO3 and NMD factors. (D) A part of the heat map from (B) corresponding to changes in the expression of the AGO1, AGO3, and NMD factors. Statistically insignificant changes are not shown

Download (640KB)
Indexing metadata
7. Supplementary 1
Download (464KB)
Indexing metadata
8. Supplementary 1
Download (630KB)
Indexing metadata

Copyright (c) 2026 Zavileyskiy L.G., Mironov A.A., Pervouchine D.D.

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.