HNRNPA1-mediated 3′ UTR length changes of HN1 contributes to cancer- and senescence-associated phenotypes

Cellular senescence has been regarded as a mechanism of tumor suppression. Studying the regulation of gene expression at various levels in cell senescence will shed light on cancer therapy. Alternative polyadenylation (APA) regulates gene expression by altering 3′ untranslated regions (3′ UTR) and plays important roles in diverse biological processes. However, whether APA of a specific gene functions in both cancer and senescence remains unclear. Here, we discovered that 3′ UTR of HN1 (or JPT1) showed shortening in cancers and lengthening in senescence, correlated well with its high expression in cancer cells and low expression in senescent cells, respectively. HN1 transcripts with longer 3′ UTR were less stable and produced less protein. Down-regulation of HN1 induced senescence-associated phenotypes in both normal and cancer cells. Patients with higher HN1 expression had lower survival rates in various carcinomas. Interestingly, down-regulating the splicing factor HNRNPA1 induced 3′ UTR lengthening of HN1 and senescence-associated phenotypes, which could be partially reversed by overexpressing HN1. Together, we revealed for the first time that HNRNPA1-mediated APA of HN1 contributed to cancer- and senescence-related phenotypes. Given senescence is a cancer prevention mechanism, our discovery indicates the HNRNPA1-HN1 axis as a potential target for cancer treatment.


INTRODUCTION
Alternative polyadenylation (APA) of mRNA is a prevalent post-transcriptional gene regulation mechanism in eukaryotes [1]. APA can be divided into UTR-APA (untranslated region APA) and CR-APA (coding region APA), based on the location of polyadenylation (pA) sites [2]. UTR-APA produces transcripts with 3′ UTRs of different length, affecting mRNA stability, translational efficiency, and the cellular localization of mRNA and protein [3], which largely depend on the interactions between alternative 3′ UTR (sequences between the proximal and distal pA sites) and other molecules such as microRNAs (miRNAs), RNA-binding proteins (RBPs) and/or long noncoding RNAs (lncRNAs) [4,5]. Many RNA binding proteins, including those involved in mRNA 3′ end formation and alternative splicing, can modulate the alternative pA site usage, indicating polyadenylation and splicing could be interrelated [6]. For example, SRSF3 (or SRp20), one member of the serine/argininerich (SR) splicing factor family, can regulate the AGING alternative pA site recognition of calcitonin coding gene CALCA [7]. HNRNPC, a member of heterogeneous nuclear ribonucleoproteins (HNRNP) family that involved in alternative splicing, has a genome-wide effect on APA regulation of mRNA with poly(U) motifs [8]. Intriguingly, many SR and HNRNP splicing factors are abnormally expressed in a variety of cancer types [9]. However, whether they could function in regulating APA and cancer-associated processes remains to be explored.
APA-mediated global shortening of 3′ UTRs is prevalent in cancer [10]. 3′ UTR shortening of some gene, such as IMP-1, can even contribute to cancer development. Cancer cells can be regarded as a state escaping from cellular senescence [11], and cellular senescence can serve as an anti-cancer mechanism in both young and elders [12]. Therefore, inducing cancer cells to enter into senescent state is a promising cancer immunotherapy strategy [13]. Cancer and senescence have opposite characteristics in many aspects, such as proliferation capacity and expression of typical molecular markers [14,15]. Interestingly, our recent publication demonstrated that cell senescence underwent 3′ UTR lengthening, and longer 3′ UTR of a specific gene, Rras2, contributed to senescence-associated phenotypes [16]. These above evidences provide the clues that APA-mediated 3′ UTR length regulation may have the possibility to switch cell fate between cancer and senescence. However, whether a single gene could regulate such cell fate transition remains to be known.
To explore whether APA of a specific gene could contribute to cancer or senescence and the possible upstream regulators, we performed an integrative analysis on multiple data sets of cancer and senescence. HN1 (Hematopoietic-and neurologic-expressed sequence 1) was finally screened out due to its preferring 3′ UTR shortening in multiple tumor tissues and favoring longer 3′ UTRs in diverse senescent cells [16,17]. HN1 transcript with longer 3′ UTR was less stable and resulted in decreased protein abundance compared with transcript with shorter 3′ UTR. Downregulation of HN1 caused senescence-associated phenotypes in both regular and cancer cells. Further investigation revealed HNRNPA1 (Heterogeneous Nuclear Ribonucleoprotein A1) as a novel regulator of APA in HN1, increased HNRNPA1 expression is responsible for 3′ UTR shortening, while decreased HNRNPA1 is responsible for 3′ UTR lengthening. This study uncovered for the first time that HNRNPA1-mediated 3′ UTR length changes of HN1 could contribute to cancer-and senescence-associated phenotypes, providing a new perspective to understand the molecular events underlying cancer and senescence, and indicating a potential target for cancer treatment as well.

APA-mediated 3′ UTR shortening of HN1 in carcinomas and lengthening in senescence
By comparing genes preferring shorter 3′ UTR in seven cancer types based on RNA-seq (RNA sequencing) datasets from TCGA (The Cancer Genome Atlas) and genes preferring longer 3′ UTR in senescent mouse embryonic fibroblasts (MEFs) and rat vascular smooth muscle cells (rVSMC) based on PA-seq (polyadenylation sequencing) [16,17], we screened out 36 genes exhibiting opposite changes in 3′ UTR length between cancer and senescence. From these, HN1 (also known as JPT1) was selected for studying the functional link between APA and these two opposite biological processes, since it has been reported to promote tumorigenesis [18,19]. The APA of HN1 was cross-validated by two methods capable of capturing precise pA sites via globally sequencing the 3′ end of mRNAs (PolyA-Seq and 3′ READS (3′ region extraction and deep sequencing) [20,21]. What's more, two pA sites clearly appeared in HN1 by track search on multiple human cells and tissues through UCSC Genome Browser (Supplementary Figure 1). Finally, these two pA sites were experimentally validated in three human cell lines including human embryonic kidney cells (HEK293T), human umbilical vein endothelial cells (HUVEC), and human lung adenocarcinoma epithelial cells (A549) using rapid amplification of cDNA 3′ end (3′ RACE) assay and Sanger sequencing ( Figure 1A). These results demonstrated UTR-APA really existed in HN1.
Since UTR-APA produces isoforms with different 3′ UTR length, we next examined whether HN1's 3′ UTR was dynamically changed between cancer and senescence. By evaluating 3′ UTR length changes with the ΔPDUI (Percentage of Distal polyA site Usage Index) [17], we found HN1 prefered the proximal pA site in six out of seven tested cancers, such as LUSC (Lung squamous cell carcinoma), LUAD (Lung adenocarcinoma), UCEC (Uterine Corpus Endometrial Carcinoma), BLCA (Bladder Urothelial Carcinoma), BRCA (Breast invasive carcinoma), and KIRC (Kidney renal clear cell carcinoma), indicating HN1 underwent a general 3′ UTR shortening in cancer ( Figure 1B). In contrast, 3′ UTR lengthening of HN1 was discovered in senescent HUVECs, rVSMCs and MEFs by our PA-seq method ( Supplementary Figures 2-4) and was validated in senescent HUVEC cells by reverse transcription and quantitative polymerase chain reaction (qRT-PCR) using primers for common or alternative 3′ UTR ( Figure 1C). In addition, qRT-PCR result of 16 hepatocellular carcinoma (HCC) cases from Chinese population also showed that the shorter 3′ UTR of HN1 was preferred in most tumor AGING samples compared to matched normal tissues ( Figure 1D). The above findings collectively demonstrated that HN1 underwent 3′ UTR shortening in various human cancers and 3′ UTR lengthening during cellular senescence.

HN1 shows opposite expression trends between cancer and senescence
To examine the potential consequences of 3′ UTR length changes in cancer and senescence, we first quantified the steady-state mRNA levels, as APAmediated 3′ UTR length changes could affect mRNA decay [1,22]. By analyzing RNA-seq data from TCGA database [23], we observed higher expression of HN1 in 20 cancer types compared to normal tissues ( Figure  2A). HN1 was also highly expressed in most Chinese HCC patients compared to matched controls ( Figure  2B). Importantly, individuals with high HN1 expression tend to have lower survival rates in various cancer types, such as KIRP (Kidney renal papillary cell carcinoma), LIHC (Liver hepatocellular carcinoma), ACC (Adrenocortical carcinoma), HNSC (Head and Neck squamous cell carcinoma), KIRC, PAAD (Pancreatic adenocarcinoma), SKCM (Skin Cutaneous Melanoma), LUAD, and LUSC (Supplementary Figure  5), indicating that HN1 may function in tumor progression. Since cellular senescence can be considered as a cancer prevention mechanism [13,24], changes in HN1 expression were then analyzed in multiple senescence models. Interestingly, HN1 showed decreased expression in multiple senescent cells, including BJ fibroblasts, human foreskin fibroblasts (HFF), three human embryonic lung fibroblasts (WI-38, MRC-5, and IMR90), and HUVEC cells ( Figure 2C and 2D). These results indicated that HN1 tended to express oppositely between cancer and senescence. as JPT1) isoforms with different 3′ UTR length in HEK293T, A549, and HUVEC cells. Forward (F) and Reverse (R) primer pairs used in the 3′ RACE were illustrated (top panel), followed by the gel image showing the PCR fragments representing two isoforms with different 3′ UTR length (middle panel). 3′ RACE amplicon corresponding to either distal or proximal pA site was Sanger sequenced and mapped to human genome (hg38), as was illustrated by UCSC Genome Browser (bottom panel). (B) APA-mediated 3′ UTR shortening of HN1 in six cancer types comparing to matched normal tissues based on public data [17]. Y axis stands for the ΔPDUI value (change in Percentage of Distal polyA site Usage Index) quantified by DaPars method. A minus ΔPDUI value represents 3′ UTR shortening. (C) qRT-PCR assay to evaluate the usage of distal pA site (L) compared to the total isoform expression (T) of HN1 among young (passage 6) and senescent (passage 15) HUVECs (**, ttest, p < 0.01). Left panel showed diagram of the primer pair design, as highlighted in blue rectangle. Right panel showed qRT-PCR result. (D) The relative expression of long isoform normalized to total HN1 expression (L/T) was measured by qRT-PCR assay among 16 paired samples of hepatocellular carcinoma (HCC). The numbers in the X axis represent the labeling ID of given patients. Left part to the dashed black line represented patients with 3′ UTR shortening of HN1.

AGING
Compared to shorter 3′ UTRs in transcripts, longer 3′ UTRs are likely to introduce more recognition sites for miRNAs and/or RBPs, which could affect mRNA stability and/or translational efficiency [1,25]. We then investigated the downstream effects of the two transcripts of HN1 with shorter 3′ UTR (HN1-S) and longer 3′ UTR (HN1-L) on gene expression, respectively. Interestingly, HN1-L was found to produce less protein than HN1-S in all three tested human cell lines (293T, A549, and HUVEC), as evaluated by dual-luciferase reporter assay ( Figure 2E), suggesting 3′ UTR lengthening of HN1 contributes to reduced protein abundance.

Knockdown of HN1 induces cellular senescence in both normal and cancer cells
Given 3′ UTR lengthening of HN1 contributed to the down-regulation of protein production, we knocked down the expression of HN1 in both normal cells (293T, and HUVEC) and cancer cells (A549) to mimic the consequence of 3′ UTR lengthening. Lentiviralmediated HN1 knockdown (KD) through two short hairpin RNAs (shHN1_#1, shHN1_#2) was confirmed by both qRT-PCR and Western blot in HUVEC ( Figure  3A and 3B), A549 ( Figure 3H and 3I), and 293T cells (Supplementary Figure 7A and 7B). Interestingly, a series of senescence-associated cellular phenotypes emerged when HN1 was stably knocked down. First, all three cell lines showed decreased proliferation rate ( Figure 3C, 3J and Supplementary Figure 7C) and increased percentage of senescence-associated SA-β-Gal positive cells ( Figure 3D-3E, 3K-3L, Supplementary Figure 7D and 7E) [26,27]. Second, decreased DNA synthesis rate was observed in HN1-KD cells by the EdU incorporation assay ( Figure 3F-3G, 3M-3N), which could be regarded as another  [23]. The red line and green line within a rectangle represent tumor tissues and normal tissues, respectively. Median expression values were denoted with the crossed black short line. Significantly increased and decreased expressions in tumor comparing to matched normal tissue indicated as red and green fonts at the top, respectively. (B) qRT-PCR measured HN1 expression levels in paired normal and tumor tissues of 16 HCC patients. The numbers on X axis represents the same labeling ID of a patient as described in Figure 1D. Left part to dashed line represented patients with upregulation of HN1. (C) HN1 expression in various senescent cells based on public RNAseq datasets [24], indicated by RPKM (reads per kilobase per million reads). Lower HN1 expression was observed in five human senescent models (error bars and t-test were based on three biological replicates. **, p < 0.01; ***, p < 0.001). (D) HN1 expression changes between young (passage 6) and senescent (passage 15) HUVECs evaluated by qRT-PCR. * stands for p < 0.05 (t-test with three PCR replicates). (E) Dualluciferase reporter assay to test the influence of 3′ UTRs (HN1-S and HN1-L) on protein production in HEK293T, A549, and HUVEC cells. Relative luminescence of Renilla luciferase was normalized using the reference Firefly luciferase activity, as shown in the left panel. *** represents a p value less than 0.001 in t-test with four biological replicates.
AGING indicator for senescence [28]. Next, HN1-KD induced G2/M cell cycle arrest in HUVEC cells and G1 arrest in  [29], showed significantly decreased expression in both HN1-KD HUVECs and A549 cells (Supplementary Figure  6C, 6H). What's more, the level of intracellular reactive oxygen species (ROS), a known senescence-associated phenotype and also a contributor to senescence [30], slightly increased in HN1-KD cells (Supplementary Figure 6D, 6I). Finally, several other senescenceassociated molecular markers, including CDK1, CCNB1, and IL6 [31], were also used to evaluate the senescence in HN1-KD cells. The result showed that HN1-KD induced down-regulation of CDK1, CCNB1 and up-regulation of IL6 in both HUVECs and A549 cells (Supplementary Figure 6E, 6J). These above combined to support the notion that HN1-KD could induce senescence-associated phenotypes in both normal and cancer cells.
Since HN1-KD could induce cellular senescence in a variety of cell types, we then went on to test whether this senescent state was HN1-dependent or not. Interestingly, overexpression of HN1 in HN1-KD HUVECs and HEK293T cells could alleviate the senescence state to some extent ( Figure 3O-3Q, Supplementary Figure 7F-7H). Similar results were also observed in A549 cancer cell line ( Figure 3R-3T). This further confirms that HN1 can act as a regulator of cellular senescence.

The 3′ UTR length of HN1 was regulated by HNRNPA1
We next searched the upstream regulators responsible for APA-mediated 3′ UTR length changes in HN1. Since RBPs were widely reported to regulate alternative polyadenylation of certain genes [32], we tried to narrow down the potential RBPs with the following criteria: 1) have potential binding motifs located at the 3′ UTR; 2) have relatively high expression in HUVEC, 293T and A549 cells; 3) opposite expression trends between cancer and senescence; 4) with the reported function to regulate either alternative polyadenylation (APA) or alternative splicing (AS) due to the crosstalk between APA and AS. Nine candidate RBPs (HNRNPA1, HNRNPAB, HNRNPA2B1, HNRNPH1, HNRNPD, HNRNPM, CPSF5, CPSF6, and PABPC1) were screened out to examine their effect on the relative abundance of HN1 transcripts with alternative 3′ UTR caused by APA. Knockdown of HNRNPA1 or HNRNPM could promote the relative expression of HN1-L compared to total expression (HN1-L and HN1-S, labelled as HN1-T), but HNRNPA1 demonstrated the maximun effect ( Figure 4A), so HNRNPA1 was choosed for further investigation. The ability of HNRNPA1 to regulate APA in HN1 was further validated separately in HNRNPA1-KD HEK293T cells (Supplementary Figure Figure  8E and 8F). These above results gave the strong evidences that splicing factor HNRNPA1 could promote the relative selection of proximal pA site in HN1, thereby adjusting the corresponding 3′ UTR length.
To further test whether HNRNPA1 could bind to mRNA of HN1, RNA immunoprecipitation coupled with reverse transcription PCR (RIP-PCR) was performed in cells expressing HA-tagged HNRNPA1. RNAs interacting with HNRNPA1 were pulled down with HA antibody and negative control IgG, then the level of HN1 mRNA was quantified by both RT-PCR and qRT-PCR. The result showed that HNRNPA1 was highly enriched on mRNA of HN1 compared to control samples in all three tested human cells (

Down-regulation of HNRNPA1 promotes senescenceassociated phenotypes through reduced HN1 expression
Since HNRNPA1-KD resulted in the higher proportion of HN1-L, which produced less protein and gave rise to senescence-related phenotypes, one would expect that down-regulation of HNRNPA1 could also induce cellular senescence. To test this hypothesis, we knocked down HNRNPA1 in HUVEC and A549 cells ( Figure 5A, 5B, 5F and 5G), and examined some senescence-associated phenotypes. Interestingly, HNRNPA1-KD led to a higher percentage of positive SA-β-Gal stained cells and slower cell growth rate in both cell types ( Figure 5C-5E, 5H-5J). Besides, HNRNPA1-KD HUVECs and A549 cells also showed arrested G2/M phase, decreased LMNB1 expression, increased ROS production, and altered expression of senescence-associated molecular markers (including CDK1, CDK2, CCNB1, and IL6) (Supplementary Figure 9A-9J), resembling to the effects caused by HN1-KD. To investigate whether HNRNPA1-KD-induced senescence was dependent on decreased HN1 expression, we overexpressed HN1 in HNRNPA1-KD cells, and found that most of the SA-β-Gal staining was reversed ( Figure 5K-5M, 5N-5P). These above results strongly suggest that APA-mediated reduction of AGING AGING HN1 protein contributes, at least in part, to HNRNPA1-KD induced cellular senescence.

Down-regulation of HN1 and HNRNPA1 inhibits cancer-related phenotypes
The fact that cancer cells exhibited 3′ UTR shortening of HN1 and up-regulation of HNRNPA1 while senescent cells displayed 3′ UTR lengthening of HN1 and downregulation of HNRNPA1 (Supplementary Figure 10A-10D) can be well explained by the HNRNPA1-mediated APA regulation of HN1. These results prompted us to hypothesize that HNRNPA1-mediated changes in HN1 expression not only can regulate cellular senescence but also have the potential to regulate cancer-associated phenotypes. To test this, we performed colony formation and cell migration assays in the lung cancer cell line A549 before and after knocking down HN1 or HNRNPA1. Interestingly, HN1 deficiency reduced the colony formation efficiency ( Figure 6A) and cell migration ability ( Figure 6B, Supplementary Figure 11A). Coincidentally, knockdown of HNRNPA1 in A549 cells also resulted in similar phenotypes as that in HN1-KD cells ( Figure 6C and 6D, Supplementary Figure 11B). Moreover, HN1-KD and HNRNPA1-KD also showed decreased cell migration rate in HUVEC cells ( Figure 6E and 6F, Supplementary Figure 11C-11D). Consistent with the previous observation that patients with higher HN1 expression have worse survival prognosis (Supplementary Figure 5), higher HNRNPA1 expression was also associated with lower survival rates in multiple cancer types (Supplementary Figure 12). Altogether, these results mRNA levels (L/T) evaluated by qRT-PCR upon knockdown of various genes encoding candidate RNA binding proteins in HEK293T. ** represents p < 0.01 based on t-test with three qPCR reactions. (B) Knockdown of HNRNPA1 by two shRNAs (shHNRNPA1_#1, shHNRNPA1_#2) was confirmed by qRT-PCR in A549 cells. *** represents p < 0.001 based on t-test with three qPCR reactions. (C) Higher usage of HN1-L to total HN1 mRNA (L/T) upon HNRNPA1 knockdown in A549 cells was evaluated by qRT-PCR analysis. ** and *** represent p < 0.01 and p < 0.001, respectively, based on t-test with three qPCR reactions. (D) 3′ RACE assay showed higher abundance of HN1-L compared to HN1-S upon knockdown of HNRNPA1. Relative mean gray value of HN1-L compared with HN1-S was shown at the upper panel by ImageJ. (E-F) HNRNPA1 overexpression reversed HNRNPA1-KD induced 3′ UTR lengthening of HN1. Overexpression of HNRNPA1 was confirmed in both HNRNPA1-KD and control A549 cells (E). Decreased usage of HN1-L compared to total HN1 mRNA (L/T) was detected by qRT-PCR upon overexpression of HNRNPA1 in A549 cell (F). * and *** represent p < 0.05 and p < 0.001, respectively, based on t-test with three qPCR reactions. (G) RNA binding protein immunoprecipitation coupled with PCR (RIP-PCR) was performed in A549 cells transfected with the HAtagged HNRNPA1-overexpression plasmid. Immunoprecipitated RNA with either HA-antibody or control IgG was reversely transcribed and amplified with primer pairs specific for HN1 mRNA. Agarose gel of three independent RIP-PCR experiments (rep1, 2, 3) was shown at left and mean gray value was shown at right. *** represents p < 0.001 based on t-test with three RIP-PCR reactions at the left. (H) Quantitative PCR for immunoprecipitated RNA (RIP-qPCR) to test the enrichment of HN1 mRNA in HA-antibody compared to control IgG in A549 cells. ** represents p < 0.01 based on t-test with three qPCR reactions. AGING suggested that down-regulation of HN1 and HNRNPA1 could inhibit cancer-related cell phenotypes and contribute to patient survival.

DISCUSSION
Our recent work indicates that 3′ UTR lengthening is a novel mechanism in regulating cell senescence [16]. However, whether a single gene shows 3′ UTR lengthening in senescent cells and 3′ UTR shortening in cancer cells can function in cell fate decision is unknown. By combining integrative analyses and experimental validations, we screened out HN1, whose UTR-APA demonstrate regulatory roles in both senescence and cancer. We further showed that HNRNPA1 expression increased in cancer and decreased in senescence well explained the related APA-mediated 3′ UTR length changes in HN1. The discovery that HNRNPA1-mediated 3′ UTR length changes in HN1 contributed to cancer-and senescence-associated phenotypes largely expands our knowledge in post-transcriptional regulation in cancer and senescence ( Figure 7). HN1 has been reported to have important biological functions. HN1 was conservatively expressed in multiple species [33], and was known to be associated with neural development [34], retina regeneration [35], and tumor progression [36]. HN1 could promote breast cancer progression by increasing MYC activity [37], and contribute to cell growth and migration in prostate cancer [18]. What's more, HN1 plays an important role in androgen-receptor mediated signaling and promotes proteosomal degradation of androgen-receptor in prostate cell lines [38]. In our study, we found HN1 could also paly a role in senescence-asscociated phenotypes ( Figure  3) in addition to apoptosis and necrosis in HUVECs and A549 cells (Supplementary Figure 13).
Although miRNAs and/or RBPs located in alternative 3′ UTR can explain the differences in RNA or protein levels of their target genes [39], how they regulate HN1's expression at the post-transcriptional level deserves further study. Though transcripts with longer 3′ UTR (HN1-L) were well validated to generate less protein than the shorter ones ( Figure 2E), whether the reduced protein level is attributed to faster RNA degradation or weakened translational efficiency was unclear. To address this question, we performed RNA stability assay and found that HN1-L had a slightly faster mRNA degradation rate compared to HN1-S in 293T cells (Supplementary Figure  14), indicating that in addition to RNA stability, translational regulation may also explain the reduced AGING protein level for this situation. Interestitly, alternative 3′ UTR of HN1 can be recognized by miR-132 to repress its translation [19]. Noteworthy, various miRNAs and RBPs with the potential to bind to the alternative 3′ UTR of HN1 were predicted with TargetScan [40] (Supplementary Figure 15) and RBPmap [41] (Supplementary Figure 16), respectively. Further experimental screening is needed to figure out detailed regulators responsible for the reduced protein production of HN1-L.
In the present study, we first explored the correlation between the changes of 3′ UTR length and expression of HN1 through the published next-generation sequencing data in cancer [17]. HN1's 3′ UTR length has a general negative correlation with its expression in six types of cancers (LUSC, LUAD, UCEC, BLCA, BRCA, and KIRC, see). We next examined in HCC tumor in Chinese patients and found a similar correlation between 3′ UTR length change and expression alteration at RNA level.
However, the degree of 3′ UTR length change seems not always correlated well with gene expression alteration, as exemplified by sample 1 and 34 ( Figure 1D and Figure  2B). We speculate that two possible reasons may explain this. The first one is the heterogeneity of HCC tumors. The tumor tissue from different patients may contain different percentage of contamination of normal cells, and the degree of infiltration of immune cells could also be different among samples. These heterogeneous cells may partially contribute to the weak correlation between 3′ UTR length changes and expression alteration. The second possible explanation is that alternative polyadenylation (APA) is not the only contributor to steady-state mRNA level of HN1. Transcriptional regulation and/or RNA degradation could also play a role, which deserves future study.
The downstream mechanism underlying HN1-KD induced senescence deserves further investigation. Graphic abstract for HNRNPA1-mediated 3′ UTR length changes contributes to cancer-and senescence-associated phenotypes. In cancer cells, upregulated HNRNPA1 leads to higher usage of proximal pA site of HN1 and thus increases the level of its transcripts with shorter 3′ UTR, which produces more protein than the longer one, and ultimately promotes cancer-associated phenotypes. In senescent cells, HNRNPA1 is downregulated and causes higher usage of distal pA site of HN1. Such regulation lengthens the 3′ UTR of HN1 and generates less protein, which in turn promotes senescence-associated phenotypes in both normal and cancer cells.

AGING
Interestingly, anti-HN1 pull-down coupled with mass spectrum showed high enrichment of ribosomal-related proteins (Supplementary Figure 17A), indicating that reduced HN1 may contribute to senescence-associated phenotypes through translational level rather than RNA level. Considering that inhibition of ribosomal RNA (rRNA) biosynthesis by chemical inhibitor CX-5461 or other factors reduces global protein production and promotes senescence-associated phenotypes [42,43], we examined whether HN1-KD induced senescence could be explained by reduced translation. Notably, decreased total protein levels and HN1 expression were observed in CX-5461-treated HUVECs and A549 cells (Supplementary Figure 17B-17E). Moreover, the rRNA synthesis inhibitor CX-5461 could also induce the increased senescenceassociated SA-β-Gal activity in both HUVEC and A549 cells (Supplementary Figure 17F-17I). Intriguingly, down-regulation of HN1 with two different shRNAs both reduced total protein abundance (Supplementary Figure 17J). However, the translation-dependent mechanism of HN1 in explaining senescence required future study.
As an RNA binding protein, HNRNPA1 has been reported to interact with many different RNAs and be involved in multiple aspects of RNA metabolism. HNRNPA1 may interact with target RNA through a sequence-based way. The telomere RNA containing UUAGGG repeats can interact with HNRNPA1 to regulate the telomere length and function [44,45]. HNRNPA1 could also bind to RNAs with other sequence motifs, such as the UAGA motif [46], AUrich or GU-rich motif in the 3′ UTR [47]. Moreover, HNRNPA1 could also interact with RNA through a structure-based way. For example, it could bind to the G-quadruplex loops in the telomere RNA and in turn regulate telomere maintenance [48,49]. Considering that telomere dysfunction is a hallmark of aging [50], the results above provide credible clues that HNRNPA1 may play a role in cellular aging. According to the iCLIP datasets in POSTAR2 [51], HNRNPA1 could potentially target to thousands of genes, suggesting that HNRNPA1 has the potential to function in various molecular and biological processes.
HNRNPA1 is a well-known splicing factor involved in the regulation of human global alternative splicing events [52]. Comprehensive misregulation of HNRNP proteins and alternative splicing disorders have been shown in many types of cancers [53]. Therefore, the role of HNRNPA1 in cancers has been thoroughly studied. For example, HNRNPA1 is highly expressed in hepatocellular carcinoma [54] and lung adenocarcinoma [55] to promote tumor progression. Down-regulation of HNRNPA1 could suppress tumorigenesis by cell cycle arrest [55] or by inducing apoptosis in cancer cells [56]. Defects of HNRNPA1 could also cause aging-related diseases, such as amyotrophic lateral sclerosis (ALS) [57] and Alzheimer's disease [58]. In addition to the splicing-dependent function that may explain cancerrelated phenotypes and other age-related diseases, the splicing-independent function of HNRNPA1 is also important for a well-understood regulatory mechanism. Indeed, previous studies have demonstrated multiple splicing-independent functions of HNRNPA1, such as transcription, mRNA transport, stability and translation [59]. Here, we identified HNRNPA1 as an important APA regulator of HN1.
Altogether, we provided evidence that HNRNPA1mediated APA regulation of HN1 can contribute to HNRNPA1-KD-induced senescence. Targeting the HNRNPA1-HN1 axis may provide a new perspective for the potential treatment of cancer and other ageassociated diseases.

3′ RACE and Sanger sequencing
Total RNA was extracted with TRIzol Reagent (Sigma) according to the manufacturer's protocol. 500 ng RNAs were reversely transcribed into cDNAs with sequencecoupling oligo (dT), using the FastQuant RT kit (TianGen). PCR was performed with primers specific to given 3′ UTR of HN1. Then the 3′ RACE products were cloned into pGEM-T-easy vector and sequenced by Sanger's method.

Vector construction for overexpression and dual luciferase assay
To construct overexpression plasmids for HN1 and HNRNPA1, cDNAs were amplified from HEK293T and cloned into pCDH vectors with HA tags or pCMV vector with FLAG tags. To test the efficiency of two transcripts of HN1, different 3′ UTR sequence was amplified from human genomic DNA and cloned into psiCHECK2 vector (Promega, cat. no. C8021) using the XhoI and PmeI restriction enzyme sites located at the 3′ end of the Renilla gene. All construct sequences were confirmed by Sanger sequencing. Primers used for cloning are listed in Supplementary Table S1.

Dual luciferase reporter assay
Cells (HEK293T, A549, and HUVEC) in 24-well plate were transfected with luciferase reporter plasmids (psiCHECK2 and those inserted with long or short 3′ UTR of HN1) for four replicates and harvested after 24 hours. 100 μl lysis buffer was added to each well and the plate was shaken at room temperature for 15 minutes (min). Then the luciferase activities in the cell lysates were measured using the dual-luciferase reporter 1000 assay system (Promega) according to the standard protocol.

Quantitative reverse transcription PCR
Total RNA was extracted by TRIzol Reagent (Sigma). cDNA synthesis was then carried out from 500 ng DNA-free total RNA using random hexamers and FastQuant RT kit (Tiangen). Triplicate samples were subjected to quantitative PCR analysis using SYBR Green (Vazyme) for HN1, HNRNPA1, and two isoforms of HN1 with different 3′ UTR length (HN1-S, HN1-L). GAPDH served as an endogenous control. The information of these primers and senescent markers (CDK1, CDK2, CCNB1, CCNE1, IL6, RB1, and p27) were listed in Supplementary Table S1.

Western blot
Total proteins were extracted by TRIzol Reagent (Sigma) according to vendor's protocol and resolved in 5% SDS solution. 20 μg of total proteins were subjected to SDS-PAGE, then transferred to nitrocellulose membranes. Membranes were incubated with anti-HN1 (Rabbit mAb, Abcam, ab126705), anti-HNRNPA1 (Rabbit mAb, Abcam, ab177152), and anti-GAPDH (Mouse mAb, Abcam, ab9484) separately at room temperature for 2 hours (h), then washed with TBST buffer, and incubated with corresponding secondary antibodies conjugated with horseradish peroxidase (HRP) (Anti-rabbit IgG, Cell Signaling Technology, #7074; Anti-mouse IgG, Cell Signaling Technology, #7076) for 1 h. Then protein levels were detected by ECL (Tanon) and images were captured using an imaging system (SageCreation).

mRNA stability assay
Cells were treated with 5 ng/ml of Actinomycin D (Act D, which can stop new transcription; Sigma-Aldrich, A4262) for 0, 2, 4, 6, 8 10, 12 and 24 h and harvested at each time point. RNA was extracted and reversely transcribed into cDNA. Transcripts with different 3′ UTR (HN1-S, HN1-L) were subjected for qRT-PCR to measure mRNA level of the corresponding isoform of HN1 at each time point.

RNA immunoprecipitation
48 hours after transfection of HNRNPA1-overexpression plasmid in 293T and HUVEC, cells were washed with ice-cold PBS and harvested into two Eppendorf tubes. The collected cells were then centrifuged and resuspended in an equal pellet volume of RIP buffer (150mM KCl, 25mM Tris pH 7.4, 5mM EDTA, 0.5mM DTT, and 0.5% NP-40) supplemented with fresh RNase inhibitor (Applied Biosystems, cat. no. N8080119) and protease inhibitor (Roche, cat. no.4693116001) and were put on ice for 10min. 20% of the supernatant was saved as RNA input and the remaining lysate was used for IP with 10 μg anti-HA antibody or negative control IgG at 4°C overnight. Protein G beads were washed three times with RIP buffer on a magnetic concentrator and then added to each IP sample to incubate with rotation for 2 hours at 4 °C. Then wash beads three times with cold RIP buffer and once with PBS. Afterward, the RNA-protein complexes were digested with proteinase K at 65 °C for 1 h. RNA was purified using TRIzol Reagent and analyzed by RT-PCR or qRT-RCR.

SA-β-Gal staining
Cells were seeded into 24-well/12-well plate with about 60% cell confluence one day in advance. Then the standard procedure of SA-β-Gal staining kit (Sigma, Biovision, cat. no. K320-250) was performed. After removing the culture medium, cells were washed twice with PBS, then fixed for seven minutes in fixation solution (1X), followed by three times washes with PBS, then incubated at 37 °C overnight in the freshmixed staining buffer. The color stained images were captured under the microscope (Leica).

Cell proliferation and EdU incorporation assays
Cell proliferation rate was assayed using Cell Counting Kit-8 (CCK-8) (Dojindo, Japan) and EdU incorporation AGING assay was carried out using kFluor488 Click-iT EdU image kit (KeyGen BioTECH). For the CCK-8 assay, cells were seeded in a 96-well plate with 100 μl culture medium and at least 2,000 cells per well. Then 10 μl CCK-8 solution was added to each well, followed by 2hour incubation, and the absorbance at 450 nm was measured using a microplate reader (TECAN). This measurement was performed every 24 hours and cell growth curve was drawn according to the absorbance value at each time point. For EdU incorporation assay, cells were seeded in a 24-well plate and incubated with 10 μM EdU for 2 hours. Changes in cell growth were assayed following the kFluor488 Click-iT EdU image kit instruction. Then images were captured by fluorescence microscope (Leica).

Colony formation assay
For each well in a 6-well plate, 2 ml complete medium containing 200 cells was prepared. After culturing cells at 37 °C with 5% CO2 for 14 days, the supernatant was discarded and the plate was washed three times with PBS. The cells were then fixed with 4% paraformaldehyde for 10 min and stained with 1% crystal violet (Sangon) for 5min. Colonies were counted under a microscope.

Cell migration assay
The 24-well transwell plate (8-μm pore size, Corning Life Sciences, USA) was used to perform cell migration assay according to the manufacturer's instruments. Briefly, cells (5 × 10 4 /well) were seeded into the upper chamber of a transwell filter with medium containing 1% FBS. Medium containing 10% FBS was added to the lower plate. After 72-hour or 18-hour incubation, cells migrated to the lower filter were fixed with 4% paraformaldehyde, washed with PBS, then stained with 1% crystal violet and counted under the microscope. Each experiment was repeated three times.

Reactive oxygen species detection
The levels of intracellular reactive oxygen species (ROS) was measured by membrane-permeable probe 2′,7′-dichlorodihydrofluorescin diacetate (DCFH-DA) (KeyGen BioTECH), which can be hydrolyzed into non-fluorescent DCFH, and then oxidized by ROS to produce fluorescent 2′,7′-dichlorofluorescin (DCF). Briefly, HUVEC and A549 cells were seeded in a sixwell plate, washed with serum-free medium, loaded with 10 μM DCFH-DA in serum-free DMEM. Following incubation in the dark at 37 °C for 30min, cells were harvested and fluorescence intensity was measured by flow cytometry (BD Bioscences).

Cell cycle and cell death analysis
Cells were trypsinized, collected by centrifugating at 500 g for 5 min, washed twice with PBS, and then resuspended with PBS containing 0.03% Triton X-100 and 50 μg/ml Propidium Iodide (PI). Following incubation in tha dark for 10 min, cell cycle assay was performed using a BD Flow Cytometer. For cell death analysis, cell apoptosis was detected by fluorescein isothiocynate (FITC)-conjugated Annexin V and propidium iodide (PI) double staining according to manufacturers' protocol (BD Bioscience). Treated cells were harvested, washed, and stained with 5 μl PI and 5 μl FITC-AnnexinV in the dark for 10min, followed by assessment performed in a BD Flow Cytometer. Both results were analyzed using ModFit and CellQuest Pro software respectively. Each sample was tested three times.

Gene expression and survival analysis of TCGA datasets
The mRNA expression of HN1 and HNRNPA1 in many cancer types and the overall survival probability were measured using the TCGA datasets from GEPIA website [23].

Human liver cancer sample collection
Hepatocellular carcinoma (HCC) samples and matched controls were obtained from Chinese HCC patients. All samples were collected with the informed consent of the patients and the experiments were approved by the ethics committee of Second Military Medical University. Clinical information was collected from patient records.

Mass spectrometry analysis
The coding sequence of HN1 was ligated to the pCMV-FLAG overexpression plasmid, and FLAG-tagged HN1 was prepared after transfecting the construct in HEK293T cells (8 x100 mm dishes). Two days later, cells were lysed in 0.1% NP-40 buffer and pCMV-FLAG-HN1 was immunopurified by anti-FLAG M2 agarose beads (Sigma), followed by SDS-PAGE and coomassie blue staining. The band corresponding to FLAG-HN1 was analyzed by liquid chromatography tandem mass spectrometry in the Proteomics platform of School of Life Science, Fudan University.

Statistical analysis
Analysis was performed using GraphPad Prism software. Unpaired t-test was used to analyze the statistical significance of differences between the means of independent groups evaluated by qRT-PCR. Error AGING bars reflect the standard error of the mean (SEM) for three replicates. Results are presented as mean ± SEM. * represents p < 0.05; ** represents p < 0.01; *** represents p < 0.001.  [2]. Significantly higher and lower mRNA levels of HNRNPA1 when comparing tumor to matched normal tissue were shown as red and green fonts at the top, respectively. Each red line within the rectangle represents tumor tissues while the green line stands for normal tissues. Median TPM expression values were denoted as crossed black short line. (B) qRT-PCR assay to test total mRNA levels of HNRNPA1 between normal and tumor tissues in 16 pairs of HCC patients. The number in X axis represents the same labeling ID of a patient as described in Figure 1D. Left of dashed line showed patients with upregulation of HNRNPA1. (C) The mRNA levels of HNRNPA1 based on public RNA-seq datasets of various senescent cells. Lower expression was observed in five human senescent models, including BJ, human foreskin fibroblasts (HFF), WI-38, MRC-5 and IMR90. RPKM was used as an indicator of expression. ** and *** represent p < 0.01 and p < 0.001, respectively, based on t-test with three biological replicates. (D) Changes in HNRNPA1 expression between young (passage 6) and senescent (passage 15) HUVECs were evaluated by qRT-PCR. * represents p < 0.05 based on t-test with three qPCR reactions. Supplementary Figure 15. The alternative 3′ UTR region of HN1 has potential binding sites for regulatory microRNAs. Multiple conserved microRNAs (miR-455-3p, miR-29-3p and miR-338-3p) were predicted to bind to the alternative 3′ UTR region of HN1 by TargetScan [3].

Abbreviations
Supplementary Figure 16. The alternative 3′ UTR region of HN1 has potential binding sites for regulatory RBPs. Multiple RNA binding proteins including HNRNPA1 were predicted to bind to the alternative 3′ UTR region of HN1 by RBPmap [4]. AGING Supplementary Figure 17. Downregulation of HN1 may contribute to rRNA biogenesis-related cellular senescence. (A) Mass spectrometry detection of proteins after HN1 immunoprecipitation, followed by functional analysis using the Datasets for Annotation, Visualization and Integrated Discovery (DAVID) [5]. Top ten enriched pathways with the highest score in DAVID were shown. (B-C) Total protein levels (B) and HN1 expression (C) detected after treatment with rRNA synthesis inhibitor, CX-5461, in HUVECs. *** represents p < 0.001 based on t-test with three BCA protein assays in B and three qPCR reactions in C. (D-E) Total protein levels (D) and HN1 expression (E) detected after treatment with CX-5461 in A549 cells. *** represents p < 0.001 based on t-test with three BCA protein assays in D and three qPCR reactions in E. (F-G) SA-β-Gal staining (F) and quantitative statistics (G) in CX-5461 treated HUVECs. *** represents p < 0.001 based on t-test with three independent countings in G. (H-I) SA-β-Gal staining (H) and quantitative statistics (I) in CX-5461 treated A549 cells. *** represents p < 0.001 based on t-test with three independent countings in I. (J) Total protein levels detected by BCA protein assay in HN1-KD (two shRNAs) A549 cells. ** stands for p < 0.01 based on t-test with three replicates of BCA assay.