Targeting influenza A virus by splicing inhibitor herboxidiene reveals the importance of subtype-specific signatures around splice sites

M splicing of human IAVs varies in a subtype-specific manner

We first examined the M splicing of human H1N1 and H3N2 viruses in human cells. Overall, six viruses—three H1N1 strains: A/WSN/33 (WSN), A/Puerto Rico/8/34 (PR8), and A/TW/3773/2015 (pH1N1); and three H3N2 strains: A/Udorn/1972 (Udorn), A/TW/3446/2002 (3446), and A/TW/2032/2017 (2032)—were compared. Total RNA was recovered 5- or 8-h post-infection and subjected to RT-PCR analyses using specific primers to detect the various spliced products of the M transcripts. Specifically, primer 1F1–11, in which extra GGGGG nucleotides were introduced at the 5′ end [52], was designed to distinguish all spliced isoforms; primer 2F26–45 amplified both M2 and M4, and primer 3F819–839, targeting the common exon of all transcripts, served as transcription control (Fig. 1a). The splicing patterns of H1N1- and H3N2-infected cells differed regardless of the collection time. A preference for M2 splicing was consistently observed in all H1N1 viruses, whereas fewer M2 transcripts were detected in H3N2-infected cells (Fig. 1b). This M2 enrichment in H1N1-infected cells was further validated by quantitative PCR (qPCR) using isoform-specific primers spanning each splice-site junction (Fig. 1b, c) [43]. We also examined the M segment splicing of human H1N1 and H3N2 viruses in chicken DF-1 cells. Consistently, only a few M2 mRNAs were produced, irrespective of the virus subtype (Fig. 1d). Considering the importance of the intrinsic RNA polymerase in regulating M splicing [33, 35], we used the RNP reconstitution system to determine whether the splicing of M minigene reporters of the WSN (H1N1), pH1N1, and 2032 (H3N2) strains could be affected by swapping the polymerase complex with that of a different subtype. Overall, the splicing pattern of each reporter was similar to the infection outcome when the M reporter was co-transfected with the cognate RNP (Fig. 1b, e). Overexpression of the noncognate RNPs had no effect on strain-specific splicing (Fig. 1e); therefore, the different splicing efficiencies of H1N1 and H3N2 M segments may depend on the intrinsic cis-elements rather than the viral RNP.

Fig. 1figure 1

Preferential M2 splicing in H1N1 over H3N2 influenza viruses. a Diagram of M segment splicing. The primers used are depicted (colored arrows). b, c Splicing pattern of M in HEK293 cells infected by six IAV strains (MOI = 1): A/WSN/33 (WSN), A/Puerto Rico/8/34 (PR8), A/TW/3773/2015 (pH1N1), A/Udorn/1972 (Udorn), A/TW/3446/2002 (3446), and A/TW/2032/2017 (2032). Total RNA was collected 5 and 8 h post-infection (hpi) and the splicing products were detected using RT-PCR (b) and qPCR (c). Fold change compared with WSN is expressed as the ratio of M2 to all M mRNA. d HEK293 and DF-1 cells were infected with WSN, pH1N1, and 2032 (MOI = 1), and total RNA was collected at 5 hpi. e HEK293 cells were co-transfected with expressing plasmids encoding RNP (PB2, PB1, PA, and NP) and M reporter plasmids from different strains as indicated. After 48 h, total RNA was collected and analyzed. Densitometric analysis of each mRNA was quantified using ImageJ

Identification of single nucleotide variations (SNVs) in the splice donor and acceptor sites of human IAV M2 segment

Considering the large difference in M2 levels observed when the M splicing efficiency in the human H1N1 and H3N2 infected cells was compared (Fig. 1b), we analyzed the sequences spanning the M2 splice donor (SD; the last 12 exonic plus first nine intronic nucleotides, − 12 to + 9) and splice acceptor (SA; the last 20 intronic plus first three exonic nucleotides, − 20 to + 3) sites. Pairwise alignment of the M cDNA of the six experimental strains showed that a single nucleotide substitution (C55T) close to the M2 5′ SS was present in all H3N2 strains, whereas another substitution (G740A) immediately downstream of the 3′ SS was identified in two of the H3N2 strains (Fig. 2a). Then, we performed an in silico analysis of the consensus sequence of the SD and SA sites using WebLogo [48]. A total of 31,110 M cDNA sequences downloaded from the online Influenza Research Database were analyzed, including 13,182 and 18,928 human H1N1 and H3N2 sequences, respectively. In contrast to the evolutionarily conserved SD of mRNA3 (Fig. 3), the two substitutions identified in our experimental strains were found to be SNVs of the M sequences, in which the favored combinations in H1N1 and H3N2 were 55C–740G and 55T–740A, respectively. The 55C/T SNV strongly affected the splice site strength, with the splice site score being reduced from 9.51 to 7.69 by the T to C conversion (Fig. 2b). Considering the identical 5′ SS of mRNA3 (with a splice site score of 7.25) carried by H1N1 and H3N2, the different M2 splicing efficiencies might be a consequence of the altered splice site strength caused by 55C/T. Next, we analyzed the distribution of these SNVs in human, avian, and swine IAVs and found that they were only present in human IAVs. Both avian and swine IAVs were predominately associated with 55C and 740G and less than 18% of nonhuman influenza viruses had 55T–740A SNVs, suggesting that these SNVs may be human-specific (Fig. 2c). Characterizing the SNV-association of each subtype revealed that most human H1N1 (97.1%) and H3N2 (99.5%) genomes carried 55C and 55T, respectively, whereas 96.5% and 99.8% of human H1N1 and H3N2 genomes contained 740G and 740A, respectively (Fig. 2d). Examination of all publicly available sequences revealed that the H3N2 strains isolated earlier than 1970 also harbored 55C–740G; thus, the 55C to T and 740G to A switches approximately occurred in 1970 and 1980, respectively, and have been preserved since (Fig. 2e). As the first H3N2 virus appeared in the 1968 outbreak as a reassortment of the human H2N2 and avian H3 viruses (Fig. 2e), the SNVs of human H3N2 viruses may be acquired during evolution after reassortment.

Fig. 2figure 2

Single nucleotide variants contribute to different splice signals of M transcripts in human influenza A viruses (IAVs). a Alignment of the M segment of six IAV strains depicting different splice signals (underlined) in H1N1 and H3N2. The 55C to T and 740G to A variants are highlighted in black boxes. b WEBLOGO plots of the splice donor (SD; 5′ splice site; 5′ SS) and splice acceptor (SA; 3′ SS) of the M segment. The height of each base represents its frequency at a given position within the M sequences. The arrow indicates the position of the 5′ SS and 3′ SS in the SD and SA sites, respectively. Splice site strength was calculated by MaxEntScan (5′: 9 bp, − 3 to + 6; 3′: 23 bp, − 20 to + 3). c Frequency of sequences containing 55T/C and 740A/G in human, avian, and swine IAVs. The y axis of each panel represents the number of isolates containing the annotated nucleotide. d Percentage of the viral subtypes containing the annotated nucleotide. e Yearly distribution plot of 55T and 740A containing sequences of the human H3N2 subtype. The percentage is based on the M sequences of the H3N2 virus

Fig. 3figure 3

WEBLOGO plots of the mRNA3 splice donor site of the M segment. The height of each base represents the frequency at a given position within the M sequences. The arrow indicates the position of the 5′ and 3′ splice site in the splice donor (SD) and splice acceptor (SA) sites, respectively. The splice site strength was calculated using MaxEntScan (5′ splice site: 9 bp, − 3 to + 6)

Adaptive evolution of the human H3N2 virus M segment from 1968 to 2019

To verify our hypothesis that SNVs may be a selection product to fine-tune the viral pathogenicity of the pandemic H3N2 virus to the following seasonal virus, we performed evolutionary analyses. We randomly selected M sequences of human H1N1, H2N2, and H3N2 viruses isolated from 1933 to 2020 and examined their phylogenetic relationships. Sequences of H2N2 were included because they are believed to have originated the H3N2 virus [53]. Two clusters were identified in the phylogenetic tree, which contained the M sequences that were dominant in H1N1 and H3N2 (Fig. 4a). The H3N2-dominant cluster showed an evolutionary path from H1N1 (blue circles) to H2N2 (yellow circles), followed by H3N2 (purple circles), as denoted by the ladder-like tree structure (Fig. 4a). Among the sequences collected before 1970 that harbored the 55C–740G SNVs, including ancient H1N1, all H2N2, and elder H3N2 sequences, the earliest C to T conversion identified (strain A/Taiwan/2/1970) was located within the H3N2 cluster. Moreover, the transition of 740A to G occurred in 1978 (strain A/Albany/14/1978), and almost all sequences (99.96%, Table 2) collected thereafter carried the 55T–740A SNVs (Fig. 4a). Hence, the H3N2-specific 55T–740A trait was not acquired from H2N2 reassortment (Fig. 4a). Notably, preservation of the 55T–740A trait did not change M1 coding, that is, 55T/C and 740A/G SNVs are null and silent mutations, respectively. In contrast, the 55T-740A trait dramatically altered the yield of the M2 isoform, which may be the selected product of evolutionary pressure. Therefore, we further examined the evolutionary relationship of M2 proteins among different human IAVs. Using an identical sequence pool for analyzing M sequences (Fig. 4a), the M2 coding sequences were extracted for evolutionary analysis. A highly similar tree structure of the M2 sequence was revealed, with most H3N2 M2 being grouped within an independent cluster (Fig. 4b). The concurrent evolution of the collinear M1 and spliced M2 isoform (Fig. 5) suggested that the functional impact of the 55T-740A SNVs may rely on the amount and/or function of M2. In addition to intra-species evolution, we examined inter-species evolution to determine whether H3N2 could acquire SNVs due to cross-species reassortment because a small portion of avian and swine IAVs also carry the 55T-type SNV (Fig. 2c; Table 2). Considering that most of the 740A-containing sequences were accompanied by the 55T SNV in human H3N2 (Table 2), we only traced the 55T variant in human, swine, and avian H1N1 and H3N2 (Fig. 4c). The 55T-type M segments could be separated into two clusters within the phylogenetic tree; one cluster contained all kinds of sequences, except for the human H3N2 sequences that were distributed in another cluster. The co-clustering of H1N1 sequences from different species indicated frequent reassortment events, whereas cross-species reassortment seldom occurred in human H3N2 (Fig. 4c). These results indicated that human H3N2 undertakes a unique evolutionary route depending on genetic drift rather than genetic shift. Overall, these findings suggest that the acquisition of 55T–740A mutations was spontaneous and preserved during the adaptive evolution of human H3N2. Next, we investigated the physiological roles of these SNVs in viral replication.

Fig. 4figure 4

Phylogenetic analysis of influenza A virus (IAV) M segments. A phylogenetic tree was constructed based on (a) M segments of human IAV, (b) M2 coding sequences of human IAV, and (c) M sequences of avian and swine IAV, and human H1N1 and H3N2. a The enlarged box shows the details regarding H3N2 emergence (purple circles) among sequences collected before 1970. The single nucleotide variations (SNVs) carried by each strain are denoted in the right-hand column. Orange and yellow arrow heads indicate the first incidence of the 55C to T and 740G to A switches, respectively. c The 55T-type isolates were selected from (a) (M/55T in H3N2 and H1N1 isolates). All sequences were aligned using MAFFT and analyzed using BEAST and TreeAnnotator [50]

Table 2 Analysis of the composition of single nucleotide variations (SNVs) 55–740 (%)Fig. 5figure 5

Concurrent evolution of the collinear M1 and spliced M2 isoform. Comparison between phylogenetic trees of the M segment and M2 coding sequence (CDS). Colors represent the subtypes and lines connect the same strains. The strains are correlated to Fig. 4a, b

A total 52,385 human H3N2, 13,178 human H1N1, 305 avian H3N2, 515 avian H1N1, 1916 swine H3N2 and 2162 swine H1N1 sequences were analyzed within sites 55–740. In the 55–740 SNV table, four different compositions were calculated. In the 740A SNV table, only sequences which had 740A SNV were calculated.

Human M segment 55C/T variant regulates the expression of M2 isoform

We first validated whether SNVs could affect the alternative splicing of M transcripts through RNP reconstitution assays using M reporter plasmids of H1N1 and H3N2 carrying the 55C/T and 740G/A variants. In the H1N1-based assays, introduction of 55T–740A (that is, the H3N2 trait) led to reduced M2 expression, whereas the mRNA3 and M4 isoforms were increased. We attributed the aberrant splicing pattern to the single C55T mutation because the G740A mutation alone only moderately reduced the M2 mRNA yield (Fig. 6a, b). When the H3N2 reporter was mutated to carry the H1N1 trait (that is, 55C–740G), we observed a more than tenfold increase in M2 splicing efficiency in a T55C-dependent manner (Fig. 6a, b). These results demonstrate that 55C/T variation can modulate M2 5′ SS utilization and that 55C favors M2 inclusion. In addition to the cell-based reporter assay, we determined the effects of the SNVs on M splicing in real infection by generating mutant viruses harboring the 55C/T variant through reverse genetics (RG). Regardless of the genotype, RG viruses bearing 55T were consistently characterized by a lower M2 level (Fig. 6c). M2 expression in the C55T mutant H1N1 virus was significantly reduced by approximately 50%. qPCR analysis revealed that collinear M1, mRNA3, and WSN-specific M4 were increased in response to the decreased M2 expression (Fig. 6d). Hence, aberrant M2 splicing mediated by the 55C/T variant occurs through alternative 5′ SS utilization rather than a general defect in splicing processing. Given that M splicing efficiency has been proposed to regulate host restriction [22,23,24], we wondered whether the different splicing efficiency observed between H1N1 and H3N2 M affects viral replication, and whether the SNVs acquired by H3N2 could be beneficial for its adaptation in humans.

Fig. 6figure 6

The 55 nucleotide is critical for the regulation of M2 splicing. Splicing of M reporters bearing each single nucleotide variant (SNV) was assessed by RT-PCR (a) and qPCR (b) using specific primers (as shown in Fig. 1e). WT: wild-type M segment, 55T or 55C: M segment with 55 mutations, 740A or 740G: M segment with a 740 mutation, 55 + 740: M segment with double mutations. c M splicing pattern HEK293 cells infected by RG viruses (MOI = 1) for 5 h. WT: wild-type virus, 55T or 55C: an RG virus containing 55 mutations. d Relative expression of M1, M2, mRNA3, and M4 determined using qPCR. Multiple comparisons between groups were performed using one-way analysis of variance. F values = 622.5 (WSN), 200.8 (pH1N1), and 411.9 (H3N2). Degrees of freedom (DFn, DFd) = (3,4). The data represent means ± standard deviations (error bars) of three independent biological replicates. NS, not significant; *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001

C55T mutation attenuates the infectiveness of the WSN strain and reduces M2 protein levels

As RG viruses bearing different SNVs were associated with altered M2 expression, we specifically investigated the impact of variable M splicing on viral replication. H1N1 (WSN) RG viruses prepared from HEK293 cells were used to infect A549 cells at a MOI of 1 or 0.001 (Fig. 7a, d). When viruses harboring the C55T mutation were examined, approximately one log10 lag in viral growth was observed regardless of the MOI, and the above described SNV-induced splicing pattern was conserved in A549 cells (Fig. 7c, f). As expected, the levels of H1N1 M2, but not H1N1 M1, were consistently reduced during the infection due to impaired splicing (Fig. 7b, e), suggesting that reduced M2 levels may attenuate the infectiveness of H1N1 viruses. Impaired replication was also confirmed by the lower protein levels of PA. However, this M2 dependence was not observed in H3N2 (2032) RG viruses. Although the T55C mutation in H3N2 reciprocally enhanced the M2 splicing efficiency, as well as the M2 protein level during infection (Fig. 7b, c, e, f), the H3N2 RG viruses grew to a similar extent regardless of which SNV was introduced (Fig. 7a, d). Thus, in contrast to the H1N1 viruses, disturbance of the M2 levels apparently has a minor effect on H3N2 replication. However, the importance of M2 on H3N2 cannot be assessed by this discrepancy.

Fig. 7figure 7

Reduced M2 protein due to 55T mutation leads to attenuated H1N1 virus in vitro. A549 cells were infected at a MOI of 1 or 0.001 with 55T- or 55C-type RG viruses. a, d The virus titer was analyzed by quantifying the number of plaques formed at the indicated time points (hpi, hours post-infection). b, e Protein expression was determined by western blotting using specific antibodies. Total RNA was collected after 5 (c) or 36 (f) h of infection. 1F and 1R primers were used to detect different M transcripts, and 3F and 1R primers were used to determine the expression of all M transcripts. Statistical analysis was performed using paired t-tests. g WEBLOGO plots of M2 amino acid sequences based on 242 human H1N1 and 222 human H3N2 isolates. The height of each amino acid represents the corresponding frequency. Protein domains of the M2 protein are depicted. The red line shows the domain region related to the above indicated functions. a H1N1 group, t = 5.001, df = 3 and H3N2 group, t = 0.4146, df = 3. d H1N1 group, t = 2.501, df = 4 and H3N2 group, t = 1.361, df = 4. (t = t-value, df = degrees of freedom). The data represent the mean ± standard deviation (error bars) of three independent biological replicates. NS, not significant; *P < 0.05; and **P < 0.01

Inhibition of human IAV replication by the splicing inhibitor herboxidiene

To confirm the importance of M2 splicing in the replication of both H1N1 and H3N2, we treated infected cells with herboxidiene, a potential antitumor drug that inhibits splicing. Both the M1 mRNA splicing and M2 protein expression of IAV viruses were downregulated after herboxidiene treatment, regardless of the subtype (Fig. 8a, b). Moreover, viral protein PB2 was downregulated in H3N2 and even disappeared in H1N1 infected cells, suggesting the inhibited replication of both viruses. The virus titer of both viruses decreased dramatically (Fig. 8c), and it declined from \(^\) to \(^\) pfu/mL in H1N1, whereas a 2 log decrease was observed in H3N2. Although the replication of H1N1 and H3N2 was affected to a different extent after herboxidiene treatment (Fig. 8b, c), M2 proteins were critical for the growth of both H1N1 and H3N2 viruses. These results indicated that splicing inhibitors may be regarded as potential therapeutic agents against influenza virus infection.

Fig. 8figure 8

Herboxidiene efficiently inhibits the replication of influenza A virus (IAV). A549 cells were infected at a MOI of 0.001 with H1N1 (A/WSN/33) or H3N2 (A/2032/2017) virus after 36 h of infection. a Protein expression was determined by western blotting using specific antibodies. b Total RNA was collected and identified using RT-PCR, 1F and 1R primers were used to detect different M transcripts, and 3F and 1R primers were used to determine the expression of all M transcripts. c Plaque formation was visualized through crystal violet staining

Compatibility of WSN replication with the optimal protein level of H3N2 M2

There might be several underlying mechanisms causing the discrepancy in M2-dependence. First, the functionality of H1N1 and H3N2 M2 may differ, and a low level of H3N2 M2 may function as well as a high level of H1N1 M2. Second, aberrantly spliced products accompanying the weakened M2 5′ SS may compensate for M2 function. Third, other co-variants yet to be identified in H3N2 may contribute to the tolerance of reduced M2. However, 55C/T SNV is a synonymous substitution for the coding of non-M2 isoforms, whereas variant amino acid residues in M2 of H1N1 and H3N2 were identified as important domains for NLRP3 inflammasome activation and the subversion of autophagy machinery [29] (Fig. 7g). Therefore, we pursued the hypothesis that the protein activity of H1N1 and H3N2 M2 might differ, leading to the difference in M2-dependence.

To confirm this, we examined the compatibility between M2 proteins of H1N1 and H3N2. We generated chimeric H1N1 RG viruses by incorporating either the wild-type (WT) 55T (H1N1 + H3wt) or mutant 55C (H1N1 + H3mut) H3N2 M segment instead of the H1N1 M segment. Infection with these chimeric RG viruses was performed in A549 cells at a MOI of 0.001, and these cells were compared with those infected with WT H1N1 RG viruses. Although a much lower M2 protein level was detected in cells infected by H1N1 + H3wt than in cells infected by H1N1, similar virus replication rates were observed (Fig. 9a, c, d), indicating that the lower levels of preserved H3N2 M segment were compatible with H1N1 replication, at least in the WSN strain. In contrast, replication of the H1N1 + H3mut virus, which was characterized by the elevated levels of M2, was significantly impaired (Fig. 9a, c, d), suggesting that the T55C mutation in H3N2 M was not beneficial for H1N1 replication. Notably, the plaque size of the chimeric H1N1 + H3mut was considerably smaller than that of the H1N1 wild-type and H1N1 + H3wt chimeric viruses (Fig. 9b). To validate the splicing pattern, total RNA of the infected cells was collected at various time points and analyzed using RT-PCR. M2 transcripts were the dominant isoform in cells infected with wild-type H1N1 viruses throughout the infection (Fig. 9c). In contrast, infection of the chimeric H1N1 + H3wt virus was characterized by decreased M2 splicing (Fig. 9c). The T55C mutation in the H3N2 M segment considerably switched the splicing from mRNA3-dominant to M2-dominant (Fig. 9a, lane H1N1 + H3mut). Therefore, the regulation of M splicing by SNVs in authentic H3N2 viruses can be recapitulated in heterogeneous chimeric H1N1 RG viruses (Figs. 7c, f, 9c). The preserved M segment splicing pattern of H3N2 in the presence of H1N1 viral proteins not only emphasizes the importance of the 55C/T SNV, but also excludes the possibility that altered M splicing can be mediated by other viral proteins, such as NS1. Consistent with the changes in the mRNA level, fewer M2 proteins were detected in H1N1 + H3wt virus-infected cells, whereas M2 was elevated along with a concomitant decrease in the M1 in cells infected with the H1N1 + H3mut virus (Fig. 9d). We observed that low expression of H3N2 M2 efficiently activated LC3 cleavage as compared to that of H1N1 M2 (Fig. 9d). The efficient LC3 cleavage by M2 might partly explain the deleterious effect of excess H3N2 M2 in the non-cognate H1N1 virus.

Fig. 9figure 9

Compatibility of H1N1 virus replication with the optimal level of H3N2 M2. a A549 cells were infected with the H1N1 WSN strain (WT) or with chimeric H1N1 RG viruses incorporating either WT 55T (H1N1 + H3wt) or mutant 55C (H1N1 + H3mut) H3N2 M segments, at a MOI of 0.001. The virus titer was determined at 12, 24, 36, 48, and 60 h post-infection (hpi) through plaque assay. The data represent means ± standard deviations (error bars) of three independent biological replicates. NS, not significant; *P < 0.05. Statistical analysis was performed using paired t-tests. WSN verses WSN + H3mut groups, t = 2.435, df = 4 and WSN verses WSN + H3wt groups, t = 2.002, df = 4. (t = t-value, df = degrees of freedom). b Plaque formation was visualized through crystal violet staining, and plaque size was determined using ImageJ. c Total RNA was collected at the indicated time points. 1F and 1R primers were used to detect different M transcripts, and 3F and 1R primers were used to determine the expression of all M transcripts. d Total protein extracts were analyzed by western blotting using specific primers. e C57BL/6 mice were intranasally infected with WT, mutant H1N1 (55T), H1N1 + H3wt, and H1N1 + H3mut, or administrated control vehicle (mock). Survival rate and body weight were assessed daily. All data were normalized to the initial weight of each mouse. Data are expressed as mean ± standard error of the mean (n = 5 mice per group) and three independent biological replicates

Optimized M2 expression in a subtype-specific manner determines IAV pathogenicity

To explore the physiological role of the various levels of M2 in IAV pathogenicity, we challenged mice with the H1N1 WSN strain (WT), mutant H1N1 (55T), and chimeric viruses including H1N1 + H3wt and H1N1 + H3mut. Less body weight loss was observed in mice challenged with the H1N1-55T mutant virus. Moreover, H1N1-55T-infected mice showed a significantly better survival rate than WT H1N1-infected mice (60% vs. 0%) (Fig. 9e). Consistent with the attenuated replication of the mutant H1N1 virus in A549 cells (Fig. 7a, d), the pathogenicity of the H1N1-55T virus was also attenuated. All mice challenged with the chimeric H1N1 + H3wt virus died with a similar disease progression to those challenged with the WT H1N1 virus (Fig. 9e), whereas the survival rate of the H1N1 + H3mut infected group dramatically improved from 0 to 40%. These results clearly demonstrated that the functionality of H3N2 M2 was different from that of H1N1 M2, thus contributing to different pathogenicity. In addition, the H3N2-specific 55T variant may have evolved to maintain the optimal level of M2 with the appropriate activity required for virus replication. In addition, human H1N1 viruses, and probably IAVs of other species, may require more M2 proteins for virus replication. Thus, optimization of M2 proteins with respect to their functionality may play a pivotal role in the replication of IAVs.

留言 (0)

沒有登入
gif