EPLIN-β is a novel substrate of ornithine decarboxylase antizyme 1 and mediates cellular migration

Polyamines, which include putrescine, spermidine and spermine, are bound to and modulate the functions of negatively charged molecules, such as DNA and RNA, as well as negatively charged proteins (Igarashi and Kashiwagi, 2000). They are not only essential for normal cellular growth and differentiation, but also play an important role in cellular proliferation and in the development of cancers (Nowotarski et al., 2013). Elevated levels of polyamines are associated with multiple cancer types, including breast, colon, lung, prostate and skin cancers (Nowotarski et al., 2013), and, hence, polyamine levels need to be carefully regulated.

Ornithine decarboxylase (ODC, encoded by ODC1) is the rate-limiting enzyme regulating polyamine synthesis (Palanimurugan et al., 2004). This enzyme catalyzes the decarboxylation of ornithine to form putrescine, a common step involved in polyamine biosynthesis, leading to the generation of spermidine and spermine. High ODC activity is associated with rapid proliferation of normal and cancerous cells and tissues (Porter et al., 1987; Gerner and Meyskens, 2004). ODC is negatively regulated by the ornithine decarboxylase antizyme (referred to as Oaz or Az) family, which is encoded by three genes: OAZ1 (Az₁), OAZ2 (Az₂) and OAZ3 (Az₃) in humans (Ivanov et al., 1998; Chen et al., 2002; Snapir et al., 2009). The most predominant Az protein is Az₁, which is ubiquitously expressed along with Az₂, although at higher levels than the latter (Ivanov et al., 1998). Az₃ is, however, only expressed in the testis (Ivanov et al., 2000).

Az proteins are expressed in a unique manner, as they are derived from two overlapping open reading frames (ORFs): ORF1 contains the translational start codon and also has a stop codon at a frameshift site, and ORF2 encodes most of the protein and lacks an initiation codon (Matsufuji et al., 1995). A +1 ribosomal frameshift results in the skipping of one nucleotide at the stop codon of ORF1 and leads to continued translation to the end of ORF2, which results in the production of functional full-length Az proteins (Matsufuji et al., 1995). Increased intracellular polyamine levels stimulate the +1 frameshift, which then negatively regulates ODC expression as a feedback mechanism to regulate polyamine levels in cells (Matsufuji et al., 1995; Rom and Kahana, 1994; Agostinelli et al., 2010; Coffino, 2001).

The enzymatically active ODC functions as a homodimer, but ODC monomers bind to Az proteins with higher affinity to form ODC–Az heterodimers (Matsufuji et al., 1995). Az binding promotes the degradation of ODC through the 26S proteasome, independent of ubiquitination (Murakami et al., 1992). Although all Az proteins are able to inhibit ODC activity and polyamine uptake, only Az₁ induces ODC degradation (Murakami et al., 1992; Li and Coffino, 1993; Mangold and Leberer, 2005). The role of Az₂ in ODC degradation is controversial (Chen et al., 2002; Murakami et al., 1992; Zhu et al., 1999), and Az₃ is unable to induce ODC degradation (Snapir et al., 2009). Therefore, Az₁-mediated downregulation of ODC and the consequential inhibition of polyamine biosynthesis are crucial processes in polyamine homeostasis.

Az₁ activity is regulated by an inhibitory interaction with Az inhibitor (AZIN, encoded by AZIN1), which is highly similar to ODC (Qiu et al., 2017; Nilsson et al., 2000). Given the similarity to ODC, AZIN competitively binds to Az, thereby relieving ODC from Az-mediated degradation. Thus, the Az/AZIN ratio determines cellular polyamine homeostasis, thereby regulating cellular growth (Olsen and Zetter, 2011). The levels of both Az₁ and AZIN have been evaluated in several cancers and corresponding normal tissues, which indicate that Az₁ levels are indeed reduced with a concomitant increase in AZIN levels in cancers (Olsen and Zetter, 2011), supporting a model in which a decrease in Az₁ levels and/or activity promotes cancer development.

Besides ODC, five other proteins that are upregulated in cancers have been identified as Az₁ substrates to date, including DNp73 (encoded by TP73), Aurora A (AURKA), cyclin D1 (CCND1), Mps1 and Smad1 (Lim and Gopalan, 2007; Newman et al., 2004; Kasbek et al., 2010; Gruendler et al., 2001; Dulloo et al., 2010). Our laboratory has previously shown that Az₁ is a negative regulator of DNp73 (Dulloo et al., 2010), an oncoprotein of the p53 (TP53) tumor-suppressor family, that is overexpressed in a variety of cancers and the expression of which leads to resistance to a variety of chemotherapeutic drugs and metastasis (Müller et al., 2005; Steder et al., 2013). Az₁-mediated DNp73 degradation also occurs in a ubiquitin-independent but proteasome-dependent manner and facilitates chemosensitivity (Dulloo et al., 2010), collectively suggesting that Az₁ could regulate the expression of a larger set of substrates to prevent tumorigenesis.

In order to uncover the full repertoire of Az₁ substrates, we undertook a proteomics approach using cells lacking Az₁ expression and identified EPLIN-β as a novel Az₁ substrate. EPLIN is an actin-binding protein encoded by the LIMA1 gene. It is expressed as two isoforms: EPLIN-α, a 600 amino-acid protein, and EPLIN-β, a 760 amino-acid protein, generated from an alternative pre-mRNA splicing event (Maul and Chang, 1999). Both isoforms participate in regulating actin cytoskeleton and dynamics. They contain a LIM domain, which is a cysteine-rich domain composed of two zinc fingers and functions as a modular interface to facilitate protein–protein interactions (Kang et al., 2000; Sánchez-García and Rabbits, 1994; Schmeichel and Beckerle, 1994). Most LIM domain-containing proteins are frequently present in molecules responsible for cytoskeletal organization (Brown et al., 1996; Collins et al., 2015), and EPLINs are also involved in maintaining epithelial cell junctions (Maul and Chang, 1999; Maul et al., 2003; Abe and Takeichi, 2008; Song et al., 2002).

Loss of EPLIN-α expression has been implicated in the progression of various cancers, such as oral, breast and prostate cancers, in which its expression was either downregulated or completely abolished compared to normal tissues, indicating a tumor-suppressor role for EPLIN-α (Liu et al., 2012, 2016). On the contrary, increased levels of EPLIN-β have been noted in various cancers (Maul and Chang, 1999), suggestive of an oncogenic or growth-promoting role for EPLIN-β. Our results illustrate that although both isoforms are able to interact with Az₁, only EPLIN-β is degraded by Az₁ in a ubiquitin-independent manner. Functional analysis indicated that EPLIN-β is a positive regulator of cellular migration, being upregulated in and causal to the enhanced cellular growth and migration of Az₁-null (Az₁^−/−) cells. These data therefore demonstrate that EPLIN-β is a novel Az₁ substrate mediating cellular migration.

In order to identify novel Az₁ substrates, we generated HCT116 human colorectal cells lacking Az₁ using CRISPR-Cas9 genome editing. The aim was to identify proteins with elevated expression levels in the absence of Az₁. Colorectal cells were chosen for the screen as polyamines are associated with the physiology of the human gut, and their levels are elevated in gastrointestinal tumors (Tofalo et al., 2019). Two pairs of guide (g) RNAs (lentiviral CRISPR constructs) targeting different coding regions of OAZ1 were used. A schematic diagram of the location of the gRNA targeting site and the screening primer locations are shown in Fig. 1A. The first pair of gRNAs targeted ORF1 of OAZ1, whereas the second pair targeted ORF2. gRNAs targeting ORF1 resulted in a 684-nucleotide insertion at the 238 nucleotide position within the CRISPR-Cas9 cutting region, and a consecutive 57-nucleotide deletion (corresponding to nucleotides 126–182) in one clone (Fig. 1B). The second pair of gRNAs led to the deletion of 92 nucleotides between nucleotides 1963 and 2054 (Fig. 1B,C, upper panel) in another clone. Sequencing analyses were performed to confirm these alterations in both clones (Fig. 1B,C, upper panel). OAZ2 was targeted (Fig. 1A) and two representative clones with an insertion of 396 nucleotides and a deletion of 14 nucleotides were obtained (Fig. 1B,C, lower panel).

Based on the hypothesis that increased levels of Az₁ substrate proteins could be identified from cells lacking Az₁, we employed the quantitative stable isotope labeling using amino acids in cell culture (SILAC) proteomics methodology (Ong et al., 2002) and compared the differential expression of proteins between Az₁^−/− knockout (KO) cells and their wild-type (WT) controls. Changes in expression of greater than twofold between heavy and light proteins (H/L ratios) were considered significant (Duan et al., 2010). The initial analysis led to the identification of several proteins that were significantly differentially expressed between Az₁^−/− cells and their WT controls, among which ODC ranked first, validating our experimental approach (Fig. 2A). The ratios of the expression of these proteins between KO and WT cells (KO/WT ratios) are shown in Table S1. In addition, another well-known Az₁ substrate, cyclin D1, was also identified in the analysis, albeit below the H/L ratio of 2, indicating that many more bona fide substrates could be identified at lower thresholds. Of note, EPLIN (LIMA1) was second on the list. Hence, we focused our efforts in characterizing the role of Az₁ in the regulation of EPLIN. As aforementioned, EPLIN is expressed as two isoforms generated from an alternative pre-mRNA splicing event (Fig. 2B). Interestingly, our SILAC analysis identified that only EPLIN-β, but not EPLIN-α, was upregulated in the absence of Az₁, indicating that the additional sequences in EPLIN-β might be critical for Az₁-mediated regulation.

To validate whether EPLIN-β is indeed regulated by Az₁, we undertook several approaches. Firstly, we quantified the expression of ODC as well as EPLIN-β levels in Az₁^−/− cells. Baseline ODC levels were low in HCT116 WT cells, but were significantly elevated in two Az₁^−/− clones (Fig. 2C), further establishing the utility of the Az₁^−/− cells in the study of its substrates. However, the absence of Az₂ did not affect ODC levels, highlighting a dominant role for Az₁ in ODC regulation, as reported previously (Murakami et al., 1992). Furthermore, treatment with polyamines (e.g. putrescine), which induces the frameshift of endogenous Az, caused a further decrease in ODC levels only in WT and Az₂^−/− cells, but had no major effects in Az₁^−/− cells (Fig. 2C), demonstrating the importance of Az₁ in ODC regulation.

The expression of EPLIN-β was also upregulated in the absence of Az₁ (Fig. 2D), but not in the absence of Az₂ (Fig. S1A), without any significant impact on EPLIN-β mRNA levels (Fig. S1B). In addition, EPLIN-β expression was also examined by immunofluorescence, and its expression was noted to be significantly increased in Az₁^−/− cells compared to that in WT cells (Fig. S2, top panel; experimental replicates for Fig. S2 are provided in Fig. S13), supporting the immunoblotting data. In addition, although we were unable to detect an increase in ODC fluorescence intensity in Az₁^−/− cells, there was a change in its localization, being found both in the cytoplasm and nucleus in Az₁^−/− cells, unlike in WT cells in which it was predominantly nuclear (Fig. S2, middle panel). Thus, by both assays, the absence of Az₁ led to an increase in EPLIN-β expression.

Moreover, although treatment of WT cells with polyamines (e.g. spermine) led to a decrease in EPLIN-β levels, this was not the case in cells lacking Az₁, in which EPLIN-β levels remained relatively unchanged (Fig. 2E), further confirming that EPLIN-β levels are indeed regulated in an Az₁-dependent manner. In addition, treatment of cells with cycloheximide (CHX) to block protein synthesis and subsequent chase indicated that the decay of EPLIN-β and ODC proteins was significantly delayed in the absence of Az₁ (Fig. 2F), suggesting that the effects of Az₁ on EPLIN-β and ODC levels were at the post-transcriptional level, leading to their extended half-lives.

Finally, we expressed cDNAs encoding either EPLIN-α or EPLIN-β alone or with Az₁ in H1299 cells to evaluate the direct impact of Az₁ overexpression on EPLIN levels. Co-expression with Az₁ led to a decrease in EPLIN-β but not EPLIN-α levels (Fig. 2G). However, co-expression with Az₂ did not affect EPLIN-β expression (Fig. S1C). These data collectively establish that EPLIN-β, but not EPLIN-α, is a target of Az₁-mediated degradation.

As all the reported substrates of Az₁ interact with it, we next evaluated whether EPLIN also interacts with Az₁. As shown in Fig. 3A, both EPLIN-α and EPLIN-β interacted with Az₁ when co-expressed. As both EPLIN-α and EPLIN-β contain a centrally located LIM domain, which might mediate self-dimerization or allow EPLIN to interact with other proteins (Fig. 3B) (Maul et al., 2003; Schmeichel and Beckerle, 1994), we hypothesized that EPLIN might interact with Az₁ through its LIM domain. To evaluate this possibility, we generated cDNAs encoding EPLIN-α and EPLIN-β with LIM domain deletions, which were co-expressed with Az₁. The absence of the LIM domain abrogated binding of both EPLIN isoforms to Az₁ (Fig. 3C), suggesting that the LIM domain of EPLIN is critical for its interaction with Az₁. Nonetheless, EPLIN-β lacking the LIM domain was refractory to Az₁-mediated degradation (Fig. 3C), emphasizing the importance of the Az₁–LIM interaction for the degradation of EPLIN-β. Moreover, co-expression with AZIN led to a partial rescue of EPLIN-β degradation by Az₁ (Fig. 3D), similar to observations with DNp73 and Cyclin D1 (Newman et al., 2004; Dulloo et al., 2010). Additionally, co-expression of EPLIN-α along with EPLIN-β had a similar effect as AZIN, reducing EPLIN-β degradation by Az₁ (Fig. 3E), suggesting a competitive model between the two isoforms resulting in the protection of EPLIN-β in the presence of EPLIN-α.

Using purified proteins, we explored whether the interaction between these two partners was direct. Based on our analyses, we were not able to observe a direct interaction between these two proteins, or with the LIM domain alone (data not shown). Thus, we believe that the interaction between Az₁ and EPLIN- β is likely indirect, in a ternary complex or, alternatively, in a way that the purified proteins do not reflect the conformation in vivo to reveal the binding.

Lastly, we evaluated whether Az₁-mediated EPLIN-β degradation occurs independent of ubiquitination, similar to other Az₁ substrates. To that end, we co-expressed Az₁ and EPLIN-β along with either WT ubiquitin, or one in which the seven lysine residues required for ubiquitin chain formation were substituted to arginine residues (i.e. Ubi-KO) (Stringer and Piper, 2011). EPLIN-β levels were reduced in the presence Az₁ irrespective of the ubiquitin status (Fig. 3F). As a positive control, we used Mdm2-mediated degradation of p53 to demonstrate the dependence on ubiquitin chain formation (Fig. 3G), as reported previously (Oliner et al., 1993; Haupt et al., 1997). These data together demonstrate that EPLIN-β is degraded by Az₁ in a ubiquitin-independent manner.

To determine whether the Az₁–EPLIN-β interaction has a functional role in cellular growth regulation, we first evaluated the effects of Az₁ deficiency on cellular growth and migration. Compared to control cell numbers, Az₁^−/− cell numbers increased significantly, leading to increased cellular colony formation (Fig. 4A). Moreover, the absence of Az₁ led to accelerated cellular migration in wound-healing assays (area of open wound after 48 h: control cells, 0.99 mm; Az₁^−/− clone 1, 0.53 mm; Az₁^−/− clone 2, 0.39 mm) (Fig. 4B), together confirming the inhibitory effects of Az₁ on cellular growth and motility.

We next examined whether this observed growth advantage could be due to elevated EPLIN-β levels. Although the majority of the previous studies indicated a putative tumor suppressor role of EPLINs, the roles of each individual isoform have not been fully elucidated. Hence, we silenced the expression of EPLIN-β in parental and Az₁^−/− clones (Fig. 4C). Silencing EPLIN-β led to a significant reduction in the growth of colonies in both Az₁^−/− clones, whereas the effect was marginal in HCT116 WT cells (Fig. 4D), indicating that elevated EPLIN-β expression in the absence of Az₁ indeed contributed to its accelerated growth. Moreover, a similar reversal of the accelerated migration of Az₁^−/− cells was noted in the wound healing assays (area of open wound after 48 h: control cells with control shRNA, 0.92 mm; Az₁^−/− clone 1 with control shRNA, 0.39 mm; Az₁^−/− clone 2 with control shRNA, 0.28 mm; control cells with EPLIN-β shRNA, 1.16 mm; Az₁^−/− clone 1 with EPLIN-β shRNA, 0.78 mm; Az₁^−/− clone 2 with EPLIN-β shRNA, 1.06 mm) (Fig. 4E). These results further illustrate that the increased growth and migration observed in Az₁^−/− cells is due to the upregulation of EPLIN-β.

The above data suggest that EPLIN-β might function as a cellular growth promoter rather than a tumor suppressor. To assess its expression in cancers, we evaluated and found that LIMA1 expression was higher in tumor tissues compared to that in normal tissues in colorectal cancers, based on the University of Alabama at Birmingham Cancer data analysis Portal (UALCAN) website (http://ualcan.path.uab.edu/index.html) (Fig. 5A). Furthermore, Kaplan–Meier analysis of survivability in a colorectal carcinoma dataset (GSE17536) from PrognoScan (Mizuno et al., 2009) indicated that higher levels of LIMA1 (EPLIN-β only) are correlated with a significantly lower survival probability (Fig. 5B, left panel). Interestingly, higher OAZ1 levels correlated with a higher survival probability in the same dataset (Fig. 5B, right panel), albeit with much lower significance. These data suggest that EPLIN-β, which enhances cancer cell growth and migration, is indeed associated with poorer patient survival.

This report describes EPLIN-β as a novel Az₁ substrate. This work expands the repertoire of Az₁ substrates to seven and also suggests the existence of many more substrates that are yet to be identified and characterized. Our proteomics analysis identified around ten substrates in HCT116 colorectal cells that were upregulated significantly (H/L>2) in the absence of Az₁, of which we characterized EPLIN-β here, and the others are being currently investigated. We envisage that there might be other substrates being regulated in a cell-type-specific and context-dependent manner, and these are being explored in our laboratory. The growing list of Az₁ substrates prompts us to speculate that Az₁-mediated protein degradation might be a yet underexplored regulatory mechanism that could be critical in the regulation of cellular growth, and thus in cancer development, as well as in other physiological and pathological processes.

Polyamines are ubiquitously produced, and their levels are relatively high in rapidly growing cells. In particular, increased polyamines levels have been shown to directly correlate with disease activity and tumor burden (Soda, 2011; Durie et al., 1977). Polyamines lead to the synthesis of full-length functional antizymes, which then negatively regulate polyamine production through the binding and proteasome-mediated degradation of ODC, the rate-limiting enzyme involved in polyamine biosynthesis (Kahana, 2018).

ODC, the classical bona fide substrate of Az₁, has been used to study the mechanistic basis of Az₁-mediated degradation. Binding of Az₁ to ODC leads to disruption of ODC homodimers, leading to the exposure of the C-terminal tail, which targets ODC to the proteasome in a ubiquitin-independent manner (Palanimurugan et al., 2004; Pegg, 2006; Almrud et al., 2000; Li and Coffino, 1993). For a long time, no other substrates had been identified, but five other substrates (Mps1, Smad1, cyclin D1, Aurora A and DNp73) were then found to be degraded in an Az₁-dependent manner. All of them have been shown to bind to Az₁ and are targeted for degradation in a ubiquitin-independent manner (Lim and Gopalan, 2007; Dulloo et al., 2010; Newman et al., 2004; Kasbek et al., 2010; Gruendler et al., 2001).

Interestingly, a few features appear to be common among all the seven substrates identified thus far: they act as homodimers or can heterodimerize with highly identical isoform variants, and they are all involved in promoting cellular proliferation and/or migration, being overexpressed in a variety of cancers (D'Assoro et al., 2016; Alao, 2007; Daniel et al., 2011; Chen et al., 2021; Yokomizo et al., 1999). All seven substrates of Az₁, including EPLIN-β, are negatively regulated by Az₁, consistent with the growth inhibitory properties of the latter. Nevertheless, whether Az₁-dependent degradation is indeed a relevant mechanism of proteolysis is unclear owing to the lack of further insights into its substrates. Our work was initiated to explore the Az₁ substratome using Az₁^−/− HCT116 colorectal cells. The absence of Az₁ expression expectedly led to elevated levels of ODC, which was also identified as the top candidate in the SILAC analysis. This proteomics analysis also identified a list of proteins that are differentially expressed in the absence of Az₁, suggesting that the Az₁ substratome might be much larger than expected, and is the subject of our future investigations.

Similar to the other Az₁ substrates, EPLIN-β was able to interact with Az₁ and was degraded in an Az₁-dependent manner without the need for ubiquitination. Co-expression with Az₁ or treatment with polyamines was sufficient for EPLIN-β degradation in an Az₁-dependent manner. Furthermore, the enhanced cellular growth and migration phenotypes observed in the absence of Az₁ and that were reversed by silencing of EPLIN-β demonstrate the functional interaction between Az₁ and EPLIN-β. Consistently, we also found that high EPLIN-β levels or lower Az₁ levels prognosticated for poorer survival, further confirming the functionality of the data in the clinical context.

EPLIN-β is an actin-binding protein, and its negative regulation by Az₁ extends a role for the latter in cellular migration, expanding the repertoire of tumor suppressive functions of Az₁. Az₁ has been suggested to be a tumor suppressor, with its expression being reduced in many cancer types (Abe and Takeichi, 2008). Previous analysis of cyclin D1 as an Az₁ substrate has suggested its role in cellular proliferation and growth (Newman et al., 2004). Consistently, Az₁^−/− cells showed enhanced cell growth, and we also noticed that they migrated faster in cellular wound healing assays compared to their WT counterparts. Furthermore, silencing of EPLIN-β reversed the enhanced migration of Az₁^−/− cells, further demonstrating an important role of Az₁ in regulating cellular motility.

Among the two EPLIN isoforms, EPLIN-α expression is significantly downregulated in several cancers (compared to normal tissues), leading to enhanced migration or invasion capabilities (Maul and Chang, 1999). Overexpression of EPLIN-α leads to inhibition of cellular growth (Sanders et al., 2011; Jiang et al., 2008). Hence, EPLINs were generally considered as tumor suppressors. However, it is to be noted that cancers also exhibit a concomitant increase in EPLIN-β expression (Maul and Chang, 1999), although it has not been clarified whether the reduction in EPLIN-α or the increase in EPLIN-β, or a combination of both, is causal to the enhanced cellular migration. Our results indicate that, at least in HCT116 colorectal cancer cells, EPLIN-β plays a pro-growth role, as opposed to the established tumor-suppressor functions of EPLIN-α (Sanders et al., 2011; Jiang et al., 2008). Consistently, analysis of several human colorectal cancer gene expression datasets showed that survival rates were significantly reduced among the patients with tumors expressing relatively higher levels of LIMA1 (denoting EPLIN-β) compared to those with lower LIMA1 levels. This further indicates that EPLIN-α expression is either downregulated or undetectable in colorectal cancers, whereas EPLIN-β expression is upregulated, consistent with previous observations (Maul and Chang, 1999).

Several points are worth highlighting based on our results. Firstly, as with other substrates, binding of Az₁ alone is insufficient, although it is necessary for EPLIN-β degradation. As such, the other EPLIN-α isoform, which is also able to bind to Az₁, is not degraded by Az₁. We speculate that although the binding region is common to both isoforms, the additional 160 amino acids of EPLIN-β are critical for degradation. Given the lack of these 160 amino acids in EPLIN-α, it is not degraded, albeit being able to bind. Therefore, structural predictions of the N-terminus of EPLIN-β were performed using the ‘AlphaFold Protein Structure Database’ (Jumper et al., 2021). It was found that the extra 160 amino acids of EPLIN-β contain an intrinsically disordered region (data not shown). As proteins with terminal or internal intrinsically disordered segments are easily recognized by the 26S proteasome for degradation (van der Lee et al., 2014), we assume that this disordered region in the N-terminus of EPLIN-β, which is absent in EPLIN-α, contributes to the degradation of EPLIN-β by 26S proteasome.

This observation is similar to the p73 proteins, of which DNp73, but not the TAp73 isoform, is degraded by Az₁, albeit both being able to bind Az₁ (Dulloo et al., 2010). This indicates that the presence of additional domains is required for the degradation process. Although the mechanistic basis of Az₁-mediated selective degradation of related proteins is unclear, we postulate that the additional regions are structurally altered upon Az₁ binding, and thus lead to proteasomal degradation. This is the case for ODC, where the structure of ODC changes upon binding to Az₁, resulting in the exposure of its C-terminus to the 26S proteasome (Wu et al., 2015). Interestingly, EPLIN-β does not contain any homologous regions to the Az₁-binding interface of ODC (data not shown), suggesting that the mechanism of degradation of various Az₁ substrates might be different.

Secondly, Az₁, but not Az₂, is the key mediator of EPLIN-β degradation. Both silencing and overexpression of Az₂ did not alter EPLIN-β levels. Thus, although polyamines can induce the frameshifting of both Az₁ and Az₂, only the former appears to be the major regulator of protein abundance, at least in the cellular systems used in our study. Thirdly, it is to be noted that many of the newly identified substrates have a much lower H/L ratio compared to that of ODC. This probably provides an explanation as to why other Az₁ substrates have not been identified thus far. Moreover, we detected the interaction between Az₁ and EPLIN-β using overexpression systems, as it is technically difficult to detect interaction of endogenous proteins, as antibodies against processed full-length Az₁ are currently not available commercially.

Finally, although EPLIN-β co-immunoprecipitated with Az₁, we were not able to notice a binding using the respective purified proteins, as well as the LIM domain alone (data not shown). This suggests that although they do bind to each other, the interaction does not occur under isolated situations, indicating that either (1) the purified proteins are not correctly folded or in the correct conformation to bind each other, as we needed to use modified methods for purification of Az₁/EPLIN-β, or (2) that some other unknown factors, particularly a protein co-factor, might be necessary for this interaction. Further biochemical studies are required to confirm whether this interaction is indeed direct or indirect.

Taken together, the data presented in this report evidently show that EPLIN-β is a novel substrate of Az₁, mediating Az₁-dependent cellular growth and migration (Fig. 5C). This study also highlights the existence of other substrates of Az₁, warranting further exploration of the entire Az₁ substratome, which could shed light on potential new targets that could be useful in clinical investigations. Moreover, this study also suggests that activation of Az₁ could be a strategy to inhibit the growth and migration of tumor cells.

p53-null human lung cancer H1299 cells, 293T cells and p53-proficient human colorectal carcinoma HCT116 cells were used in this study (obtained from American Type Culture Collection) (Dulloo et al., 2015). These three cell lines have been tested to be mycoplasma free. Cells were grown in Dulbecco's modified Eagle medium (DMEM; Hyclone) supplemented with 10% bovine fetal serum (FBS; Hyclone), 1% penicillin-streptomycin solution, 2 mM L-glutamine (Invitrogen, Carlsbad, CA, USA), 100 μM non-essential amino acids (Invitrogen) and 0.1 mM sodium pyruvate (Invitrogen), as described previously (Subramanian et al., 2015). The EPLIN-β knockdown cells were maintained in complete DMEM medium containing 0.5 µg/ml puromycin. Putrescine was 51799-100MG, lot BCBZ6097, Sigma-Aldrich; spermine was S3256-1G, lot BCBG8969V, Sigma-Aldrich.

The relevant plasmids were transfected using Lipofectamine 2000 (Invitrogen) in accordance with the manufacturer's instructions. Expression vectors (all in pcDNA3.0) expressing full length EPLIN-α, EPLIN-β, Az₁, ODC or AZIN have been described previously (Li et al., 2010; Dulloo et al., 2010).

HCT116 parental cells were cultured with ‘heavy’ [SILAC Dulbecco’s modified Eagle’s medium (88364, Life Technologies), 10% dialyzed fetal bovine serum (26400044, Life Technologies), ¹³C₆¹⁵N₂ L-lysine-2HCl (143 mg/ml) (0.1%) (CNLM-291-0.5, Cambridge Isotopes), ¹³C₆¹⁵N₄ L-arginine-HCl (83 mg/ml) (0.1%) (CNLM-539-0.5, Cambridge Isotopes) and 1% penicillin-streptomycin solution (Hyclone, SV30010)] or ‘light’ [SILAC Dulbecco’s modified Eagle’s medium and 10% dialyzed fetal bovine serum, ¹²C₆¹⁴N₂ L-lysine-HCl (143 mg/ml) (0.1%) (L8662-25G, Sigma-Aldrich), ¹²C₆¹⁴N₂ L-arginine-HCl (83 mg/ml) (0.1%) (A8094-25G, Sigma-Aldrich) and 1% penicillin-streptomycin solution] medium and the corresponding Az₁-KO cells were cultured with ‘light’ or ‘heavy’ medium for ∼4–5 doublings, at a cell density of around 20–30% confluence. The cells were harvested, and normalized protein extracts were used for SILAC analysis. LDS buffer (Invitrogen) and reducing agent (Invitrogen) were added to the lysates and boiled for 5 min. The proteins were then separated on 4–12% NuPage Novex Bis–Tris Gels (Invitrogen); stained with a colloidal blue staining kit (Invitrogen) and digested with trypsin using in-gel digestion procedures (Shevchenko et al., 2006).

Tryptic peptides were analyzed using an EASY-nLC 1000 coupled to a Q Exactive Hybrid Quadrupole-Orbitrap (Thermo Fisher Scientific). The peptides were resolved and separated on a 50 cm analytical EASY-Spray column (ES803, Thermo Fisher Scientific) equipped with a pre-column over a 120 min gradient ranging from 8 to 38% of 0.1% formic acid in 95% acetonitrile/water at a flow rate of 200 nl/min. Survey full-scan mass spectrometry (MS) spectra (mass-to-charge ration or m/z, 310–2000) were acquired with a resolution of 70,000, an automatic gain control (AGC) target of 3×10⁶ and a maximum injection time of 10 ms. The top twenty most intense peptide ions in each survey scan were sequentially isolated to an AGC target value of 5×10⁴ with a resolution of 17,500 and fragmented using a normalized collision energy of 25. A dynamic exclusion of 10 s and isolation width of 2 m/z were applied. SILAC peptide and protein quantification was performed with MaxQuant version 1.5.0.30 (https://maxquant.org/) using default settings. Database searches of MS data were performed using UniProt human fasta (2017) (https://www.uniprot.org/proteomes/UP000005640) with tryptic specificity allowing a maximum of two missed cleavages, two labeled amino acids and an initial mass tolerance of 4.5 ppm for precursor ions and 0.5 Da for fragment ions. Cysteine carbamidomethylation was searched as a fixed modification, and N-acetylation and oxidized methionine were searched as variable modifications. Labeled arginine and lysine were specified as fixed modifications. Maximum false discovery rates were set to 0.01 for both proteins and peptides. Proteins were considered identified when supported by at least one unique peptide with a minimum length of seven amino acids. All mass spectrometry data have been deposited in the PRIDE repository with the accession number PXD039865.

Cell lysates were prepared in lysis buffer containing 0.5% Nonidet P-40 (NP-40) with protease inhibitor cocktail and dithiothreitol as described previously (Dulloo and Sabapathy, 2005). The total protein amounts were quantified and boiled in 4× SDS sample buffer, followed by separation on SDS polyacrylamide gels. Immunoblotting was performed with the following antibodies: anti-actin (A2066_.2 ml, lot 058M4812V, Sigma-Aldrich) and anti-ODC (O1136_.2 ml, lot 034M4836V, clone ODC-29) from Sigma-Aldrich (St. Louis, MO, USA); anti-FLAG (PA1-984B, lot WG319616) from Thermo Fisher Scientific; anti-EPLIN (sc-136399, lot H2117), anti-Myc (9E10) (sc-40, lot D0419) and anti-Mdm2 (sc-965, lot F2718; 1:200 for western blotting) from Santa Cruz Biotechnology, Dallas, TX, USA; anti-mouse secondary antibody (7076S, lot 36), anti-rabbit secondary antibody (7074S, lot 29), anti-HA-tag (C29F4) (3724S, lot 10) and anti-His-tag (27E8) (2366S, lot 14) from Cell Signaling Technology, Danvers, MA, USA. The antibody dilution factors are shown in Table S3. Uncropped images of the blots shown in Figs 2, 3 and 4 are shown in Figs S3, S6 and S9, respectively; experimental replicates of these blots are shown in Figs S4, S7 and S10, respectively, and uncropped images of the replicate blots are shown in Figs S5, S8 and S11, respectively. Uncropped images of the blots shown in Fig. S1 are shown in Fig. S9; experimental replicates of these blots in Fig. S10, and uncropped images of the replicate blots in Fig. S11.

Cell lysates, prepared as described above, were used for co-immunoprecipitation (IP) assays as described previously (Li et al., 2010). Briefly, cell lysates were immunoprecipitated with the indicated antibodies (both anti-FLAG and anti-Myc antibodies were used at a concentration of ∼5 μg/ml for IP) for 3 h at 4°C and then incubated with protein G-agarose (GE Healthcare Life Sciences) for 2 h at 4°C. Protein G-agarose was washed with NP-40 buffer three times and then analyzed by western blotting with the indicated antibodies.

To knockout Az₁ or Az₂ in HCT116 cells, CRISPR-Cas9 system was used as previously published (Sanjana et al., 2014). Briefly, sgRNAs targeting Az₁ (Az₁ gRNA1_Forward, 5′-caccgCTAAGCCTGCACAGCCGCGG-3′; Az₁ gRNA1_Reverse, 5′-aaacCCGCGGCTGTGCAGGCTTAGc-3′; PCR screening primer_Forward1, 5′-GCAGCGGATCCTCAATAGCCAC-3′; PCR screening primer_Reverse1, 5′-CTTCTGGAAGCTTCGGACGG-3′; Az₁ gRNA2_Forward, 5′-caccgGACTATTCCCTCGCCCACCT-3′; Az₁ gRNA2_Reverse, 5′-aaacAGGTGGGCGAGGGAATAGTCc-3′; PCR screening primer_Forward2, 5′-GACCTGCATCATCTTCAGTTCC-3′; PCR screening primer_Reverse2, 5′-CTCAGGCAACGCCTGGGCTG-3′) or Az₂ (Az₂ gRNA_Forward, 5′-caccgCCACCGGGGATCTTCGACAG-3′; Az₂ gRNA_Reverse, 5′-aaacCTGTCGAAGATCCCCGGTGGc-3′; PCR screening primer_Forward, 5′-CACCCTCTCTGTCTCTTGCAGTAG-3′; PCR screening primer_Reverse, CAGTGGAGTCTGAGAAAGCTCCAAC-3′) were designed using the online tool CRISPOR (http://crispor.tefor.net/) and cloned into the vectors lentiCRISPRv2 (Addgene plasmid #52961) using Bbs1 enzyme. After validation by DNA sequencing, the sgRNA vectors were co-transfected with the corresponding packaging plasmids into 293T cells, and the virus supernatant was prepared. HCT116 cells were then transduced with polybrene (4 mg/ml, Sigma-Aldrich) and viral supernatant, and selected on 0.5 μg/ml puromycin-containing medium for 10–12 days. Single colonies were randomly picked up into 96-well plates for genotyping by PCR and the knockouts were further validated by PCR using the Az₁- or Az₂-specific genomic primers.

To determine EPLIN-β half-life, HCT parental and Az₁ KO cells were seeded into six-well plates at ∼60% confluence. After 24 h, the cells were treated with the protein synthesis inhibitor cycloheximide (Sigma-Aldrich) for the indicated durations before harvesting and analysis by immunoblotting.

For EPLIN-β ubiquitination analysis, H1299 cells were transfected with His-ubiquitin (pCDNA3.0-6xHis-ubiquitin, cloned in our laboratory), Myc-Az₁ (as above) or FLAG-EPLIN-β (as above). After 24 h of transfection, cells were lysed in NP-40 lysis buffer and then incubated with anti-FLAG antibody (Thermo Fisher Scientific; 5 μg/ml) for 3 h and protein G-agarose beads for 2 h at 4°C. After washing three times, the ubiquitinated EPLIN-β was detected by immunoblotting using anti-His monoclonal antibody.

HCT116 parental and Az₁-KO cells were seeded into six-well plates with glass coverslips. The next day, the permeabilized cells were blocked for 1 h with PBS supplemented with 2% bovine serum albumin and 5% FBS, and incubated with primary anti-ODC and anti-EPLIN antibodies (as above) antibodies at 4°C overnight, at 1:200 dilution (Phang et al., 2015). After washing, cells were incubated with secondary antibodies [goat anti-mouse IgG, IgM (H+L), secondary antibody, Alexa Fluro^TM 488, A11001, lot: 1397999, Thermo Fisher Scientific] for 1 h at room temperature. Images were visualized and captured using a Nikon Motorized Upright microscope 90i (NIS-Elements AR3.0, Japan).

Total RNA was prepared from cells using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Then, 1.5–2 µg of total RNA was reverse transcribed into cDNA using SuperscriptII (Invitrogen). The sequences of real-time PCR primers and semi-quantitative reverse transcription-PCR analysis are the same as described (Subramanian et al., 2015).

The shRNA for EPLIN-β was cloned into pLKO.1 vector (Addgene 8453). Briefly, two restriction enzymes were used to linearize the vector followed by gel extraction. The insert shRNAs consist of two oligonucleotides that are complementary (forward and reverse primers listed in Table S2). When the two oligonucleotides annealed together, they contained the appropriate overhangs to allow cloning into the vector. The pLKO.1 vector was used as the control. The shRNAs were co-transfected with the corresponding packaging plasmids into 293T cells, and the virus supernatant was prepared. After 2 days of infection, the cells were under puromycin selection for 1 week before seeding for the experiments (Jin et al., 2018).

HCT116 cells – WT, Az₁ knockout or Az₁ knockout cells, with or without EPLIN-β knock-down cells (1000 cells per well) – were plated on six-well plates and grown for 2 weeks. After rinsing three times with PBS, cells were fixed with 4% paraformaldehyde in PBS, and stained with 1% Crystal Violet.

Similarly, the above cells were seeded in 24-well plates at a density of 1.6×10⁴ cells/well in complete DMEM and cultured to confluence. The cells were then serum-starved overnight in DMEM, and the confluent cell monolayers were scraped with a yellow pipette tip to generate scratch wounds and were washed twice with Opti-MEM (Invitrogen) to remove cell debris (Zhao et al., 2011). Cells were incubated at 37°C with the medium containing 0.5 µg/ml puromycin. Time-lapse images were captured at 0, 24, 48 and 72 h time points in the same position using a Nikon Eclipse TE2000-5 microscope.

We checked the survival status of colorectal cancer patients by using PrognoScan (Mizuno et al., 2009) as it has features of a large collection of publicly available cancer microarray datasets with clinical annotation. The Kaplan–Meir plots of LIMA1 and OAZ1 genes were analyzed by using it.

The differences in mean values were considered significant at P-values <0.01 (**) and <0.05 (*). The log-rank test was applied in the Kaplan–Meier survival examination. Statistical comparisons between two groups were carried out by two-tailed unpaired Student's t-test. A two-tailed P-value <0.05 was considered significant.

We thank Catherine Yen Li Kok and Kai Ning Chew for technical assistance.