The NMD Pathway Regulates GABARAPL1 mRNA during the EMT

EMT is a reversible cellular process that is linked to gene expression reprogramming, which allows for epithelial cells to undergo a phenotypic switch to acquire mesenchymal properties. EMT is associated with cancer progression and cancer therapeutic resistance and it is known that, during the EMT, many stress response pathways, such as autophagy and NMD, are dysregulated. Therefore, our goal was to study the regulation of ATG8 family members (GABARAP, GABARAPL1, LC3B) by the NMD and to identify molecular links between these two cellular processes that are involved in tumor development and metastasis formation. IHC experiments, which were conducted in a cohort of patients presenting lung adenocarcinomas, showed high GABARAPL1 and low UPF1 levels in EMT+ tumors. We observed increased levels of GABARAPL1 correlated with decreased levels of NMD factors in A549 cells in vitro. We then confirmed that GABARAPL1 mRNA was indeed targeted by the NMD in a 3′UTR-dependent manner and we identified four overlapping binding sites for UPF1 and eIF4A3 that are potentially involved in the recognition of this transcript by the NMD pathway. Our study suggests that 3′UTR-dependent NMD might be an important mechanism that is involved in the induction of autophagy and could represent a promising target in the development of new anti-cancer therapies.


Introduction
The epithelial-to-mesenchymal transition (EMT) is a physiological process in which epithelial cells lose their cell polarity and acquire mesenchymal properties; it occurs during embryogenesis or tissue remodeling (EMT I and II). During cancer progression, the EMT (type III) [1] plays a key role in metastasis formation by promoting cellular invasion, dissemination and tissue colonization. Specific morphological changes observed in cells undergoing the EMT are controlled at different levels of gene expression, including epigenetics, transcriptional and post-transcriptional levels, or translation [2]. It was reported that the EMT has an impact on autophagy, and conversely, the induction of harboring a PTC, it has also been shown that NMD can act as a post-transcriptional regulatory mechanism targeting physiological transcripts.
In this study, we demonstrated that the GABARAPL1 mRNA was regulated at the posttranscriptional level by the NMD during the EMT. According to our results, the mechanism by which NMD targeted GABARAPL1 mRNA occurred through the association of both UPF1 and eIF4A3 on its 3 UTR. These data indicated that during the EMT, NMD inhibition would lead to an increase of GABARAPL1 levels, suggesting a role for the NMD pathway in the regulation of autophagy. For the cellular treatments, cycloheximide (01810, Sigma, St. Louis, MI, USA), G418 (11811023, ThermoFisher, Santa Cruz, CA, USA) and NMDi14 (SML1538, Sigma-Aldrich) were used.

Antibodies, Vectors, Chemicals and siRNAs
The pGFP-GABARAP vector cloning was previously reported in [18]. The pGFP-GABARAP + 3 UTR-GABARAPL1 was synthesized by amplifying the 3 UTR of GABARAPL1 added with SacII and BamHI restriction sites via PCR using the following primers: forward-5 -TAAGCACCGCGGGTGGTTGGAAGCCCAGCAG and reverse-ATTTTATTAACGAATGGAATTTTTAGCCTAGGATTCGT. The amplified sequence was then cloned downstream GABARAP CDS in a pGFP-GABARAP plasmid using the SacII and BamHI sites.
The siRNAs used in the transfection were purchased from Eurogentec (Seraing, Belgium). The sequences are listed in Table 1. Table 1. siRNA target sequences used for the transfections.
Transfections of the plasmids were performed using the Lipofectamine 2000 reagent (11668030, ThermoFisher) following the supplier's recommendations. Briefly, cells were seeded in 6-well plates and transfected for 24 h with 1 µg of plasmid and 2 µL of Lipofectamine 2000 reagent and harvested 24 h after the end of the transfection. Transfections of the siRNA were conducted using the same protocol with 20 pmol of siRNA and 3 µL of Lipofectamine 200 reagent.

RNA Extraction and RT-qPCR
The total RNAs were extracted using the Tri Reagent (Molecular Research Center, TR-118, Cincinnati, OH, USA) following the supplier's recommendations. Briefly, the cell pellet (300,000 cells) was suspended in 750 µL of Tri Reagent and 0.2 mL of chloroform was added before centrifugation at 12,000× g for 15 min at 4 • C. The total RNAs were then precipitated using 0.5 mL of isopropanol and centrifuged at 10,000× g for 20 min at 4 • C. The RNA pellet was washed with 70% ethanol before solubilization in water (30 µL) and then incubated for 10 min at 65 • C. Total purified RNAs were then quantified before conservation at −80 • C.
For RT-qPCR analysis, 1 µg of total RNAs were reverse-transcribed using 60 units of M-MLV Reverse Transcriptase (Sigma, M1302), 1.25 µM of oligo (dT)23 (Eurogentec) and 1.25 µM of random primers (Promega, C1181, Madison, WI, USA). Quantitative PCR was run on the Step One Real-Time PCR System (Applied Biosystems, Waltham, MA, USA) using the Syber Green PCR Master Mix (Applied Biosystems, 4309155) and the following parameters: 10 min at 95 • C followed by 40 cycles of 15 s at 95 • C and 1 min at 60 • C. The target gene levels (see the primers in Table 2) were normalized to the ratio of 2 housekeeping gene expressions: H3B2 and 18S rRNA. The primer sequences used for qPCR analysis are described in Table 2. Each sample was analyzed in duplicate and then differences in the expression of each gene were quantified using the ∆∆CT method.

RNA-FISH
The cells were fixed in 4% paraformaldehyde for 30 min at room temperature and RNA-FISH assays were performed using the RNAscope ® Fluorescent Multiplex Assay (Advanced Cell Diagnostics Bio, Milan, Italy) following the supplier's recommendations. Briefly, cells were dehydrated with increasing concentrations of EtOH (50, 70 and 100% for 2 min) and rehydrated with decreasing concentrations of EtOH (70 and 50% for 2 min). Cells were then incubated with Protease III for 10 min at room temperature and subjected to GABARAP and GABARAPL1 probe hybridization and counterstaining with DAPI. Slides were mounted in Vectashield (Polysciences Inc., Warrington, PA, USA) and observed using Z-section imaging with a confocal microscope (Zeiss LSM 800 AiryScan, Oberkochen, Germany).

Immunohistochemistry and Immunostaining Assessment
FFPE specimens of NSCLCs were collected (in collaboration with the Tissue Biobank of the University of Liege, Liege, Belgium) and stained with the antibodies listed in Table S2. The protocol was approved by the Ethics Committee of the University Hospital of Liege and the diagnosis of each case was confirmed by experienced histopathologists. IHC analysis was performed using a standard protocol detailed previously [19]. Samples were classified into two groups, namely, EMT-positive vs. EMT-negative, depending on their Vimentin immunoreactivity. As previously described [19,20], the IHCs were scored by multiplying two values: intensity (0-3) and extent (0-3), leading to a global score ranging between 0 and 9. Using this latter arbitrary scale, all immunolabelled tissues were evaluated independently by two experienced histopathologists.
Following extensive washing in a Lysis buffer, immunoprecipitated mRNAs and proteins were analyzed using RT-qPCR and Western blotting, respectively. To do so, bound complexes were eluted using the Elution buffer (100 mM Tris-HCl, pH 6.8, 20% glycerol, 4% SDS, 12% β-MeSH). Immunoprecipitated mRNAs were then purified using the Tri Reagent extraction protocol described before and proteins were denatured for 5 min at 95 • C prior to the Western blot analysis.

The EMT was Associated with A Loss of UPF1 Expression as well as GABARAPL1 Overexpression
Recent data showed that the key NMD factor UPF1 is downregulated during the epithelial to mesenchymal transition (EMT) [21]. First, we confirmed these data in a cohort (n = 42) of non-small cell lung carcinomas (NSCLCs) in which the tissue samples were sorted according to their Vimentin status: EMT+ tumors that exhibit high percentages of malignant cells expressing this biomarker and EMT− tumors with a complete absence of expression for this protein by the cancer cells [22]. As shown in Figure 1 (bottom panel), EMT+ tumors presented lower levels of UPF1 compared to their EMT− counterparts. Furthermore, according to recent data obtained in our laboratory, we also showed the overexpression of the ATG8 family member GABARAPL1 in the case of the EMT (Figure 1, upper panel) (Jacquet et al., manuscript accepted biology-1328521). Even if a global increase in GABARAPL1 associated with a downregulation of UPF1 was observed in EMT+ tumors, the high variability of protein levels that was observed between tissue specimens (patient heterogeneity) did not allow for a significant inverse correlation.
Biomedicines 2021, 9, x FOR PEER REVIEW 6 of 18 Following extensive washing in a Lysis buffer, immunoprecipitated mRNAs and proteins were analyzed using RT-qPCR and Western blotting, respectively. To do so, bound complexes were eluted using the Elution buffer (100 mM Tris-HCl, pH 6.8, 20% glycerol, 4% SDS, 12% β-MeSH). Immunoprecipitated mRNAs were then purified using the Tri Reagent extraction protocol described before and proteins were denatured for 5 min at 95 °C prior to the Western blot analysis.

The EMT was Associated with A Loss of UPF1 Expression as Well as GABARAPL1 Overexpression
Recent data showed that the key NMD factor UPF1 is downregulated during the epithelial to mesenchymal transition (EMT) [21]. First, we confirmed these data in a cohort (n = 42) of non-small cell lung carcinomas (NSCLCs) in which the tissue samples were sorted according to their Vimentin status: EMT+ tumors that exhibit high percentages of malignant cells expressing this biomarker and EMT− tumors with a complete absence of expression for this protein by the cancer cells [22]. As shown in Figure 1 (bottom panel), EMT+ tumors presented lower levels of UPF1 compared to their EMT− counterparts. Furthermore, according to recent data obtained in our laboratory, we also showed the overexpression of the ATG8 family member GABARAPL1 in the case of the EMT (Figure 1, upper panel) (Jacquet et al., manuscript accepted biology-1328521). Even if a global increase in GABARAPL1 associated with a downregulation of UPF1 was observed in EMT+ tumors, the high variability of protein levels that was observed between tissue specimens (patient heterogeneity) did not allow for a significant inverse correlation.  The regulation of the expression of GABARAPL1, as well as the other members of the ATG8 family, remains unclear. One of the main reasons is that most of the studies investigated their regulation at the translational or post-translational level, and very few studied their post-transcriptional regulation. We then wondered whether ATG8 members could be targeted by the NMD to induce their degradation. The A549 human lung cancer cell line was used to determine the expression of UPF1 and GABARAPL1 at both the mRNA and protein levels during the EMT since this cell line was previously described to undergo the EMT in response to TGFβ/TNFα treatment [23]. The analysis of UPF1 and GABARAPL1 expressions during the induction of the EMT by a TGFβ/TNFα treatment showed the downregulation of UPF1 transcripts, as well as the ones of other NMD factors, after 15 h of treatment for UPF1 and after 6 h for UPF2, UPF3A, 3B and SMG1 (Figure 2a). This decrease was then followed by a re-expression when treatment was prolonged with no significant decrease after 48 h of treatment. These results were also confirmed at the protein level for UPF1 (Figure 2c,d), with a drastic decrease in UPF1 protein level after 6 h of treatment. Interestingly, we observed an inverse profile for GABARAPL1 transcripts, with a high expression of GABARAPL1 mRNA after 6 h of treatment. When the treatment was extended, the GABARAPL1 mRNA level remained higher than in the control condition, but with higher variability, leading to a non-significant increase (Figure 2b). These results suggested that the inhibition of NMD could promote a rapid increase in GABARAPL1 mRNA levels following TGFβ/TNFα induction. The regulation of the expression of GABARAPL1, as well as the other members of the ATG8 family, remains unclear. One of the main reasons is that most of the studies investigated their regulation at the translational or post-translational level, and very few studied their post-transcriptional regulation. We then wondered whether ATG8 members could be targeted by the NMD to induce their degradation. The A549 human lung cancer cell line was used to determine the expression of UPF1 and GABARAPL1 at both the mRNA and protein levels during the EMT since this cell line was previously described to undergo the EMT in response to TGFβ/TNFα treatment [23]. The analysis of UPF1 and GABARAPL1 expressions during the induction of the EMT by a TGFβ/TNFα treatment showed the downregulation of UPF1 transcripts, as well as the ones of other NMD factors, after 15 h of treatment for UPF1 and after 6 h for UPF2, UPF3A, 3B and SMG1 (Figure 2a). This decrease was then followed by a re-expression when treatment was prolonged with no significant decrease after 48 h of treatment. These results were also confirmed at the protein level for UPF1 (Figure 2c,d), with a drastic decrease in UPF1 protein level after 6 h of treatment. Interestingly, we observed an inverse profile for GABARAPL1 transcripts, with a high expression of GABARAPL1 mRNA after 6 h of treatment. When the treatment was extended, the GABARAPL1 mRNA level remained higher than in the control condition, but with higher variability, leading to a non-significant increase (Figure 2b). These results suggested that the inhibition of NMD could promote a rapid increase in GABARAPL1 mRNA levels following TGFβ/TNFα induction.

GABARAPL1 Transcript Increased upon NMD Inhibition
To assess whether the ATG8 family member transcripts subject to NMD, we performed chemical inhibitions of the NMD using cycloheximide (CHX, a protein synthesis inhibitor), G418 (promoting PTC readthrough) or NMDi14 (disrupting SMG7-UPF1 interactions), which are drugs that were shown to inhibit the NMD [24][25][26]. First, amongst the drugs that were used to inhibit the NMD, only the CHX induced a significant effect on the described NMD targets used as positive controls, with an increase in SC35 and ATF4 mRNA levels in the A549 cell line. Indeed, following 5 h of treatment, the ATF4 mRNA levels presented a 7-fold induction and the SC35 levels presented a 15-fold induction (Figure 3a). Similar results were obtained in a second cancer cell line, namely, the SiHa cells ( Figure S1). Among the ATG8 family transcripts, only GABARAPL1 levels and, to a lesser extent, LC3B levels were increased upon CHX treatment. Indeed, the GABARAPL1 mRNA levels were increased about seven-and fourfold, after the CHX treatment in A549 and SiHa cells, respectively. The increase in LC3B was less important with the threefold induction in A549 and about twofold in SiHa cells (Figures 3a and S1). No effect was observed for the other ATG8 family transcripts or endogenous NMD-insensitive transcripts, such as SF3B5 and ATG5. The effect of the CHX treatment on the induction of GABARAPL1 and LC3B transcripts appeared to be independent of an indirect transcriptional effect that is linked to ATF4, a known NMD target, since ATG5, which is a known transcriptional target of ATF4, like the ATG8 family members, was not increased upon CHX treatment. The treatment using NMDi14 for 24 h did not induce any significant effect on the induction of either SC35 or ATF4 transcripts, which are two described NMD targets, demonstrating an inefficient NMD inhibition. Moreover, the treatment of the cells with NMDi14 caused important cell death and a proliferation inhibition, both in SiHa and A549 cells, as well as in U2OS cells (data not shown). The treatment using G418, which can enhance a stop codon readthrough by the ribosome and is therefore used to inhibit the recruitment of NMD factors on transcripts harboring PTCs, had only a minor effect on the ATF4, SC35 and GABARAPL1 mRNA levels in SiHa cells and showed no effect on those same targets in A549 cells. Furthermore, the co-treatment of cells with both G418 and NMDi14 did not show any further increase compared to G418 alone.
To examine whether NMD inhibition could affect the subcellular localization, as well as the expression of GABARAPL1 mRNAs, we performed RNA-FISH using specific fluorescent probes targeting the GABARAPL1 mRNA on A549 cells that were treated, or not, with CHX. In both conditions, the GABARAPL1 mRNAs were observed as punctate dots in the cytoplasm (Figure 3b). We observed a significant increase in GABARAPL1 mRNA dots in CHX-treated cells (about 14 dots per cell) compared to untreated cells (about 4 dots per cell) ( Figure 3c). Therefore, the RNA-FISH experiments confirmed an increase in GABARAPL1 transcripts upon CHX treatment.  To inhibit the NMD with a higher specificity than with CHX, we used siRNA-mediated inhibition of the NMD by targeting the transcripts of two core NMD factors, UPF1 and SMG1, alone or in combination. The inhibition of the NMD by targeting these two factors was previously described to lead to NMD inhibition [27] and seemed to be efficient since the levels of the positive controls ATF4 and SC35 were increased following UPF1 and SMG1 knockdown in both A549 and SiHa cells, with greater effects observed when using both siRNAs (Figures 4a and S2). According to the results obtained with the CHX treatment, the GABARAPL1 mRNA levels were increased when NMD was inhibited by siUPF1 and siSMG1, with, once again, a greater effect obtained when used in combination in both A549 and SiHa cells (Figures 4a and S2). This increase in GABARAPL1 mRNA level upon siRNA-mediated NMD inhibition was also observed in the glioblastoma cells U87 ( Figure S3). However, no significant increase in LC3B mRNAs was observed following siRNA transfection, suggesting that the effect observed for this gene upon CHX treatment was not due to a direct NMD inhibition and was probably due to an off-target effect of the CHX. The increase in GABARAPL1 was then confirmed at the protein level using Western blotting (Figure 4b), with a significant increase of GABARAPL1 protein levels after siUPF1 and siSMG1 co-transfection (Figure 4c). To inhibit the NMD with a higher specificity than with CHX, we used siRNA-mediated inhibition of the NMD by targeting the transcripts of two core NMD factors, UPF1 and SMG1, alone or in combination. The inhibition of the NMD by targeting these two factors was previously described to lead to NMD inhibition [27] and seemed to be efficient since the levels of the positive controls ATF4 and SC35 were increased following UPF1 and SMG1 knockdown in both A549 and SiHa cells, with greater effects observed when using both siRNAs (Figures 4a and S2). According to the results obtained with the CHX treatment, the GABARAPL1 mRNA levels were increased when NMD was inhibited by siUPF1 and siSMG1, with, once again, a greater effect obtained when used in combination in both A549 and SiHa cells (Figures 4a and S2). This increase in GABARAPL1 mRNA level upon siRNA-mediated NMD inhibition was also observed in the glioblastoma cells U87 ( Figure  S3). However, no significant increase in LC3B mRNAs was observed following siRNA transfection, suggesting that the effect observed for this gene upon CHX treatment was not due to a direct NMD inhibition and was probably due to an off-target effect of the CHX. The increase in GABARAPL1 was then confirmed at the protein level using Western blotting (Figure 4b), with a significant increase of GABARAPL1 protein levels after siUPF1 and siSMG1 co-transfection (Figure 4c).

UPF1 Was Associated with GABARAPL1 mRNAs
To confirm our above results and demonstrate a direct targeting of GABARAPL1 transcripts by the NMD, we analyzed the association of the UPF1 protein with different ATG8 transcripts. To do so, RNA-IPs were performed using either an anti-UPF1 antibody (RIP-UPF1) or a control IgG (RIP-CTL). The specificity and efficiency of the UPF1 recovery were assessed using Western blotting ( Figure 5a) and UPF1-bound mRNAs were analyzed using RT-qPCR (Figure 5b). A 10% input was used to normalize the amount of proteins used in the experiment. As expected, described endogenous NMD targets (ATF4 and SC35), used as positive controls, were indeed significantly enriched in the RIP-UPF1 compared to the RIP-CTL, while the mRNAs that were not targeted by the NMD (H3B2 and SF3B5) and used as negative controls were undetected in both the RIP-CTL and the RIP-UPF1. Regarding the ATG8 transcripts, the GABARAPL1 and LC3B mRNAs were detected in the RIP-UPF1 with a fourfold enrichment, but no association of UPF1 with GABARAP mRNAs was detected. Even though LC3B transcripts were detected in the RIP-UPF1, we previously showed in Figure 4 that this transcript was not increased in cells in which the NMD was inhibited, suggesting that LC3B mRNAs were not targeted by the NMD. Taken together, our RIP data indicated that UPF1 did not associate with all ATG8 transcripts and demonstrated that GABARAPL1 mRNAs were specifically recognized and targeted by the NMD.

UPF1 Was Associated with GABARAPL1 mRNAs
To confirm our above results and demonstrate a direct targeting of GABARAPL1 transcripts by the NMD, we analyzed the association of the UPF1 protein with different ATG8 transcripts. To do so, RNA-IPs were performed using either an anti-UPF1 antibody (RIP-UPF1) or a control IgG (RIP-CTL). The specificity and efficiency of the UPF1 recovery were assessed using Western blotting (Figure 5a) and UPF1-bound mRNAs were analyzed using RT-qPCR (Figure 5b). A 10% input was used to normalize the amount of proteins used in the experiment. As expected, described endogenous NMD targets (ATF4 and SC35), used as positive controls, were indeed significantly enriched in the RIP-UPF1 compared to the RIP-CTL, while the mRNAs that were not targeted by the NMD (H3B2 and SF3B5) and used as negative controls were undetected in both the RIP-CTL and the RIP-UPF1. Regarding the ATG8 transcripts, the GABARAPL1 and LC3B mRNAs were detected in the RIP-UPF1 with a fourfold enrichment, but no association of UPF1 with GABARAP mRNAs was detected. Even though LC3B transcripts were detected in the RIP-UPF1, we previously showed in Figure 4 that this transcript was not increased in cells in which the NMD was inhibited, suggesting that LC3B mRNAs were not targeted by the NMD. Taken together, our RIP data indicated that UPF1 did not associate with all ATG8 transcripts and demonstrated that GABARAPL1 mRNAs were specifically recognized and targeted by the NMD.

The 3′UTR of GABARAPL1 mRNAs Contributed to Its Degradation by The NMD
The GABARAP family members presented a strong identity part of their coding regions but differed in the length of their 3′UTR, with a long 3′UTR for GABARAPL1 (1276 bp) and LC3B (1671 bp) compared to GABARAP (433 bp) (Figure 6a). When we looked at the sequences of the ATG8 mRNAs, it was surprising to see that GABARAPL1 could be targeted by the NMD, while GABARAP could not since those transcripts were very similar, as shown on the scheme in Figure 6a. Indeed, their ORFs shared 79% identity at the mRNA level and 90% identity at the protein level. The main difference between these two transcripts lay in their 3′UTRs, which was longer for GABARAPL1 than the one of GABA-RAP (1276 nt vs. 433 nt, respectively) and they shared no identity. Since it was admitted that the NMD could target physiological transcripts harboring long 3′UTRs (>1000 nucleotides), we proposed the hypothesis that the GABARAPL1 mRNA was subject to degradation by the NMD via its long 3′UTR. To test this hypothesis, the sequence of the 3′UTR of GABARAPL1 was cloned downstream of the ORF of GABARAP in a pGFP-C1 plasmid, and the expressions of the fusion transcripts were analyzed at both the mRNA and protein levels. As shown in Figure 6c, the addition of the 3′UTR of GABARAPL1 seemed to decrease the expression of the fusion protein GFP-GABARAP since we observed fewer fluorescent cells expressing the fusion protein. This effect was confirmed using Western blotting, with a decrease in the expression of the fusion protein GFP-GABARAP (Figure 6d,e). To confirm that this decrease of the GFP-GABARAP protein levels was due to a decrease in the mRNA levels, the expression of the GFP-GP + 3′UTR GL1 transcript was assessed using RT-qPCR (Figure 6b). The results showed that, indeed, the addition of the 3′UTR of GABARAPL1 downstream the ORF of GABARAP decreased the levels of the fusion transcript, thus confirming our hypothesis that GABARAPL1 is targeted by the NMD via its

The 3 UTR of GABARAPL1 mRNAs Contributed to Its Degradation by The NMD
The GABARAP family members presented a strong identity part of their coding regions but differed in the length of their 3 UTR, with a long 3 UTR for GABARAPL1 (1276 bp) and LC3B (1671 bp) compared to GABARAP (433 bp) (Figure 6a). When we looked at the sequences of the ATG8 mRNAs, it was surprising to see that GABARAPL1 could be targeted by the NMD, while GABARAP could not since those transcripts were very similar, as shown on the scheme in Figure 6a. Indeed, their ORFs shared 79% identity at the mRNA level and 90% identity at the protein level. The main difference between these two transcripts lay in their 3 UTRs, which was longer for GABARAPL1 than the one of GABARAP (1276 nt vs. 433 nt, respectively) and they shared no identity. Since it was admitted that the NMD could target physiological transcripts harboring long 3 UTRs (>1000 nucleotides), we proposed the hypothesis that the GABARAPL1 mRNA was subject to degradation by the NMD via its long 3 UTR. To test this hypothesis, the sequence of the 3 UTR of GABARAPL1 was cloned downstream of the ORF of GABARAP in a pGFP-C1 plasmid, and the expressions of the fusion transcripts were analyzed at both the mRNA and protein levels. As shown in Figure 6c, the addition of the 3 UTR of GABARAPL1 seemed to decrease the expression of the fusion protein GFP-GABARAP since we observed fewer fluorescent cells expressing the fusion protein. This effect was confirmed using Western blotting, with a decrease in the expression of the fusion protein GFP-GABARAP (Figure 6d,e). To confirm that this decrease of the GFP-GABARAP protein levels was due to a decrease in the mRNA levels, the expression of the GFP-GP + 3 UTR GL1 transcript was assessed using RT-qPCR (Figure 6b). The results showed that, indeed, the addition of the 3 UTR of GABARAPL1 downstream the ORF of GABARAP decreased the levels of the fusion transcript, thus confirming our hypothesis that GABARAPL1 is targeted by the NMD via its long 3 UTR since it can alone destabilize the GABARAP transcript, which was not initially targeted by the NMD.
Biomedicines 2021, 9, x FOR PEER REVIEW 12 of 18 long 3′UTR since it can alone destabilize the GABARAP transcript, which was not initially targeted by the NMD.

eIF4A3 Regulated GABARAPL1 mRNA Degradation via Its 3′UTR.
To decipher how the GABARAPL1 mRNA could be targeted to NMD degradation via its long 3′UTR, proteins bound to the 3′UTRs of GABARAP, GABARAPL1 and LC3B were analyzed using the AURA 2.5.2 database (Figure 7) [28]. We observed that UPF1 was found to bind the 3′UTR of GABARAPL1, as well as the one of LC3B, and no UPF1 binding site was found on the GABARAP transcript, confirming our RIP-UPF1 results. Interestingly, we observed four eIF4A3 binding sites on the 3′UTR of the GABARAPL1 transcript that overlapped with the UPF1 binding sites, but none on the LC3B 3′UTR. Therefore, we hypothesized that eIF4A3, together with UPF1, could regulate the degradation of transcripts via their long 3′UTR. This model could explain why UPF1 could bind the 3′UTR of LC3B without inducing its degradation since it was not bound by eIF4A3.

eIF4A3 Regulated GABARAPL1 mRNA Degradation via Its 3 UTR
To decipher how the GABARAPL1 mRNA could be targeted to NMD degradation via its long 3 UTR, proteins bound to the 3 UTRs of GABARAP, GABARAPL1 and LC3B were analyzed using the AURA 2.5.2 database (Figure 7) [28]. We observed that UPF1 was found to bind the 3 UTR of GABARAPL1, as well as the one of LC3B, and no UPF1 binding site was found on the GABARAP transcript, confirming our RIP-UPF1 results. Interestingly, we observed four eIF4A3 binding sites on the 3 UTR of the GABARAPL1 transcript that overlapped with the UPF1 binding sites, but none on the LC3B 3 UTR. Therefore, we hypothesized that eIF4A3, together with UPF1, could regulate the degradation of transcripts via their long 3 UTR. This model could explain why UPF1 could bind the 3 UTR of LC3B without inducing its degradation since it was not bound by eIF4A3. To confirm the impact of eIF4A3 on the levels of GABARAPL1 transcripts, eIF4A3 was inhibited using specific siRNAs and the levels of ATG8 transcripts were analyzed using RT-qPCR in A549 cells ( Figure 8). As expected, increased mRNA levels of GABARAPL1, ATF4 and SC35 were detected in the eIF4A3-depleted cells (Figure 8a), thus confirming its role in GABARAPL1 mRNA regulation. The same effect was observed in a second cell line, SiHa cells, with a significant increase in GABARAPL1 mRNA levels following the sieIF4A3 treatment ( Figure S4). Furthermore, eIF4A3 inhibition did not lead to a significant increase in LC3B mRNA, suggesting that eIF4A3 has a key role in the degradation of transcripts binding UPF1 in their 3′UTR. The effect of eIF4A3 inhibition on GABARAPL1 expression was then confirmed at the protein level using Western blotting (Figure 8c), and quantitative analysis of protein levels using Western blotting confirmed the elevated levels of GABARAPL1 proteins in eIF4A3-depleted cells (Figure 8d). Altogether, these results suggest that the mechanism by which the NMD targeted GABARAPL1 mRNA occurred through the association of both UPF1 and eIF4A3 on its 3′UTR. To confirm the impact of eIF4A3 on the levels of GABARAPL1 transcripts, eIF4A3 was inhibited using specific siRNAs and the levels of ATG8 transcripts were analyzed using RT-qPCR in A549 cells ( Figure 8). As expected, increased mRNA levels of GABARAPL1, ATF4 and SC35 were detected in the eIF4A3-depleted cells (Figure 8a), thus confirming its role in GABARAPL1 mRNA regulation. The same effect was observed in a second cell line, SiHa cells, with a significant increase in GABARAPL1 mRNA levels following the sieIF4A3 treatment ( Figure S4). Furthermore, eIF4A3 inhibition did not lead to a significant increase in LC3B mRNA, suggesting that eIF4A3 has a key role in the degradation of transcripts binding UPF1 in their 3 UTR. The effect of eIF4A3 inhibition on GABARAPL1 expression was then confirmed at the protein level using Western blotting (Figure 8c), and quantitative analysis of protein levels using Western blotting confirmed the elevated levels of GABARAPL1 proteins in eIF4A3-depleted cells (Figure 8d). Altogether, these results suggest that the mechanism by which the NMD targeted GABARAPL1 mRNA occurred through the association of both UPF1 and eIF4A3 on its 3 UTR.

Discussion
In our study, we decided to investigate the regulation of autophagy genes at the posttranscriptional level since most of the precedent studies focused on transcription or posttranslational levels. To do so, we concentrated our efforts on the key pathway involved in post-transcriptional regulation, which is the NMD, and we decided to use lung cancer as a pathology model since it was previously demonstrated that the NMD was decreased in this cancer and that GABARAPL1, an autophagy gene, was increased. Our results indeed confirmed that the levels of the key NMD factor UPF1 were downregulated in patients presenting with lung cancers with high levels of VIMENTIN, a key marker of the EMT. The EMT was described as being involved in tumor progression, and TGFβ signaling was shown to play an important role in this process. In vitro, the downregulation of UPF1 during the EMT in a lung adenocarcinoma cell line was previously described by Cao et al. [21], where the authors showed that the inhibition of NMD led to the induction of the EMT and that the overexpression of NMD factors led to the inhibition of the EMT. The authors also suggested that this regulation of the EMT by the NMD occurred through the targeting of factors involved in the TGFβ signaling pathway, suggesting an important role of the NMD pathway in the induction of the EMT. Interestingly, we observed that the decreased levels of UPF1 during the EMT were associated with an increase in

Discussion
In our study, we decided to investigate the regulation of autophagy genes at the post-transcriptional level since most of the precedent studies focused on transcription or post-translational levels. To do so, we concentrated our efforts on the key pathway involved in post-transcriptional regulation, which is the NMD, and we decided to use lung cancer as a pathology model since it was previously demonstrated that the NMD was decreased in this cancer and that GABARAPL1, an autophagy gene, was increased. Our results indeed confirmed that the levels of the key NMD factor UPF1 were downregulated in patients presenting with lung cancers with high levels of VIMENTIN, a key marker of the EMT. The EMT was described as being involved in tumor progression, and TGFβ signaling was shown to play an important role in this process. In vitro, the downregulation of UPF1 during the EMT in a lung adenocarcinoma cell line was previously described by Cao et al. [21], where the authors showed that the inhibition of NMD led to the induction of the EMT and that the overexpression of NMD factors led to the inhibition of the EMT. The authors also suggested that this regulation of the EMT by the NMD occurred through the targeting of factors involved in the TGFβ signaling pathway, suggesting an important role of the NMD pathway in the induction of the EMT. Interestingly, we observed that the decreased levels of UPF1 during the EMT were associated with an increase in GABARAPL1 expression. The ATG8 family member GABARAPL1 was shown to be required for the degradation of damaged organelles or protein aggregates via selective autophagy [29], and recent results obtained in our laboratory showed a role of GABARAPL1 during the EMT (Jacquet et al. manuscript under review in Biology). Even though the regulation of GABARAPL1 during the EMT occurs, in part, through a transcriptional level, the mechanisms behind its regulation are not fully understood. Our results demonstrated that NMD factors were expressed at low levels during the EMT and that this expression was inversely correlated with the levels of GABARAPL1 in lung cancer models, suggesting that GABARAPL1 could also be regulated during the EMT at the post-transcriptional level by the NMD pathway as an interplay between mRNA degradation pathway and autophagy. Moreover, crosstalk between autophagy and selective mRNA degradation was previously described in yeast by Makino and collaborators [30], demonstrating that autophagy could selectively target and degrade mRNAs.
To confirm that GABARAPL1 was directly targeted by the NMD, this pathway was inhibited by both chemical inhibitors or siRNAs. These inhibitions led to the increase in GABARAPL1, both at the mRNA and protein levels, confirming that the NMD could regulate autophagy transcripts, as suggested by Wengrod et al. [31]. We also confirmed the overexpression of ATF4 mRNA upon NMD inhibition. However, our results differed from the ones of Wengrod and collaborators since we did not observe any increase in ATG5 mRNA levels. Wengrod and colleagues also showed that the inhibition of the NMD led to the induction of autophagy in an indirect manner through the targeting of ATF4, which is a transcription factor targeted by the NMD [32]. Since this transcription factor is known to activate the transcription of many autophagy-related genes, including ATG5 and several ATG8 family members, these data could explain the differences between our data and the ones from Wengrod and collaborators. This was confirmed by the fact that we did not observe any increase in other ATG8 family members, even though several of them were previously shown to be regulated at the transcriptional level by ATF4 [4,5,33]. Therefore, our data indicated that the GABARAPL1 upregulation was not occurring through an indirect ATF4 regulation but through direct targeting of its transcripts by the NMD pathway. However, transcriptional regulation of ATG8 members' expressions by ATF4 following NMD inhibition is not excluded since we observed a slight but non-significant increase of ATG5 mRNA levels following NMD inhibition by siRNAs, but we might have analyzed the transcripts' expressions too early following NMD inhibition to observe this transcriptional regulatory effect.
Then, our RIP-UPF1 results showed that UPF1 did bind the GABARAPL1 mRNA, as well as the one of LC3B. The mechanism of GABARAPL1 mRNA recognition by the NMD pathway remained quite unclear since this transcript contained no known premature termination codon. However, it was previously proposed in a study analyzing the mechanism by which long 3 UTR mRNAs elicit or escape the NMD that the 3 UTR of GABARAPL1 could promote NMD degradation, leading to the hypothesis of targeting through its 3 UTR [34]. We then cloned the GABARAPL1 3 UTR downstream of the coding sequence of GABARAP and confirmed that GABARAPL1 was targeted via its 3 UTR, suggesting a degradation by an EJC-independent NMD, which was previously suggested for transcripts harboring a long 3 UTR (>1000 nt) [16,35,36]. However, the size of the 3 UTR, as well as its binding by UPF1, did not seem to be sufficient to induce NMD degradation since numerous transcripts harboring long 3 UTR were never described as NMD targets, and that LC3B, which also presents a long 3 UTR, did not seem to be degraded by the NMD, even if this transcript was found to be enriched upon UPF1 RNA-IP.
Using the Atlas of UTR Regulatory Activity (AURA) data, we found that the EJC factor eIF4A3 bound the 3 UTR of GABARAPL1 but not the ones of LC3B and GABARAP. Previous studies indicated that eIF4A3 occupancy was variable along mRNAs with both canonical and noncanonical EJC binding sites [11,37]. Interestingly, eIF4A3-binding sites were located in proximity to UPF1-binding sites on the 3 UTR of GABARAPL1. Therefore, we hypothesized that eIF4A3, together with UPF1, could regulate the degradation of transcripts via their long 3 UTRs. This model could explain why UPF1 could bind the 3 UTR of LC3B without inducing its degradation since it was not bound by eIF4A3.
To further decipher the features that could lead transcripts harboring long 3 UTR being degraded by the NMD, we then analyzed the proteins bound to GABARAPL1 mRNA, as well as other NMD targets, in proximity to UPF1. This analysis confirmed that UPF1 could bind both GABARAPL1 and LC3B 3 UTRs, as well as GADD45B and TBL2 3 UTRs, which are two transcripts that were previously described to be targeted by the NMD through their 3 UTRs [38]. The most important outcome of this analysis was the systematic presence of eIF4A3 in proximity to UPF1 on GABARAPL1, as well as GADD45B and TBL2, but there was an absence of LC3B 3 UTR. This result led us to think that the discrimination of NMD targets between all transcripts that were bound by UPF1 in their 3 UTR could depend on the presence of eIF4A3. Moreover, the same protein environment was observed in proximity to UPF1 for these four transcripts, with the presence of proteins such as RBM47, LIN28, or AGO. For the latter, it was suggested that UPF1 could play a role in the argonaute pathway and the degradation of transcripts by miRNAs, which is a mechanism that then may be involved in the degradation of transcripts that bind UPF1 in their 3 UTRs [39]. However, even if LC3B did not seem to be degraded by the NMD since it did not respond to NMD inhibition from siRNA targeting, it would be interesting to further analyze the role of UPF1 in the regulation of this transcript since it responded to CHX treatment and was enriched upon UPF1 RNA-IP. However, the impact of UPF1 binding on the LC3B mRNA is still unclear, but we hypothesize that it may have a role in the subcellular localization of LC3B since UPF1 was previously described to present non-NMD roles, such as transport and localization of many cellular transcripts, but further experiments will be needed to confirm this hypothesis.
Altogether, our results led us to propose a model in which the 3 UTR-dependent NMD may be an important mechanism for the induction of autophagy during the EMT via the regulation of the levels of the GABARAPL1 transcript (Scheme 1). However, we are aware that further studies are required to determine whether the modulation of the NMD pathway during the EMT could contribute to the induction of selective autophagy mediated by GABARAPL1 and its consequences on tumor progression. we hypothesized that eIF4A3, together with UPF1, could regulate the degradation of transcripts via their long 3′UTRs. This model could explain why UPF1 could bind the 3′UTR of LC3B without inducing its degradation since it was not bound by eIF4A3.
To further decipher the features that could lead transcripts harboring long 3′UTR being degraded by the NMD, we then analyzed the proteins bound to GABARAPL1 mRNA, as well as other NMD targets, in proximity to UPF1. This analysis confirmed that UPF1 could bind both GABARAPL1 and LC3B 3′UTRs, as well as GADD45B and TBL2 3′UTRs, which are two transcripts that were previously described to be targeted by the NMD through their 3′UTRs [38]. The most important outcome of this analysis was the systematic presence of eIF4A3 in proximity to UPF1 on GABARAPL1, as well as GADD45B and TBL2, but there was an absence of LC3B 3′UTR. This result led us to think that the discrimination of NMD targets between all transcripts that were bound by UPF1 in their 3′UTR could depend on the presence of eIF4A3. Moreover, the same protein environment was observed in proximity to UPF1 for these four transcripts, with the presence of proteins such as RBM47, LIN28, or AGO. For the latter, it was suggested that UPF1 could play a role in the argonaute pathway and the degradation of transcripts by miRNAs, which is a mechanism that then may be involved in the degradation of transcripts that bind UPF1 in their 3′UTRs [39]. However, even if LC3B did not seem to be degraded by the NMD since it did not respond to NMD inhibition from siRNA targeting, it would be interesting to further analyze the role of UPF1 in the regulation of this transcript since it responded to CHX treatment and was enriched upon UPF1 RNA-IP. However, the impact of UPF1 binding on the LC3B mRNA is still unclear, but we hypothesize that it may have a role in the subcellular localization of LC3B since UPF1 was previously described to present non-NMD roles, such as transport and localization of many cellular transcripts, but further experiments will be needed to confirm this hypothesis.
Altogether, our results led us to propose a model in which the 3′UTR-dependent NMD may be an important mechanism for the induction of autophagy during the EMT via the regulation of the levels of the GABARAPL1 transcript (Scheme 1). However, we are aware that further studies are required to determine whether the modulation of the NMD pathway during the EMT could contribute to the induction of selective autophagy mediated by GABARAPL1 and its consequences on tumor progression. Scheme 1. Role of the NMD pathway in the regulation of GABARAPL1 mRNA during the EMT. Scheme 1. Role of the NMD pathway in the regulation of GABARAPL1 mRNA during the EMT.