Epigenetic Inactivation of α-Internexin Accelerates Microtubule Polymerization in Colorectal Cancer

Colorectal cancer, one of the most common cancers, represents a typical tumor entity exhibiting aberrant DNA methylation that contributes to tumor initiation and progression (1–4). The dense hypermethylation of CpG islands located in the promoter regions of tumor-suppressive genes has been established as an essential mechanism for gene inactivation (4, 5). The cumulative alterations in the form of somatic mutations and DNA methylation in colonic epithelial cells are presumed to be lynchpins during cancer initiation. We previously found that genome-wide alterations in DNA methylation drive the colorectal adenoma transformation to cancer (6). Even in the early stage, DNA methylation has been aberrant in the inflammatory colorectal epithelium and in normal colon mucosa of individuals who at high risk for colorectal cancer (7–9). Despite being such a vital phenotype, aberrant DNA methylation sustains tumor progress in a manner that remains largely obscure. Identifying functional methylated genes may help us gain insight into the epigenetic mechanisms involved in colorectal cancer pathogenesis.

Malignant proliferation and metastasis are complex processes requiring an extensive reorganization of cellular cytoskeletons (10). A high frequency of intracellular cytoskeleton reorganization has been observed in cancer, including actin cytoskeleton remodeling, microtubule rearrangement, and intermediate filament alteration (11–13). These transformations are frequently controlled by cytoskeleton-associated proteins throughout the remodeling process (14, 15). However, the function of DNA methylation on cytoskeletal reorganization in cancer has rarely been systematically considered. It is reasonable to speculate that aberrant DNA methylation in cytoskeletal genes may radically alter the composition of tumorous skeletons, evoking the concomitant rearrangement of intracellular cytoskeletons. Hence, we acquired the methylation status of cytoskeletal genes covered in an established genome-wide screening of methylated genes in colorectal cancer using HumanMethylation450 Bead Chip (HM450K) arrays, and several aberrantly hypermethylated cytoskeletal genes were identified. Among the sequences, the promoter in the INA (α-internexin) gene, a member of type IV intermediate filament, was notably hypermethylated in colon adenoma and adenocarcinoma.

INA is one of the neurofilaments in which the heteropolymer is composed of four subunits, with the others in the group being neurofilament light (NEFL), neurofilament medium (NEFM), and neurofilament heavy (NEFH). INA has been identified as a structural component of the cytoskeleton that is involved in neurogenesis (16). Data from the Human Protein Atlas show a moderate expression of INA in normal human colorectum (https://www.proteinatlas.org/ENSG00000148798-INA/tissue), but lesions of colorectal cancer and other tumor types expressing INA at low levels (https://www.proteinatlas.org/ENSG00000148798-INA/pathology). Recent studies reported that the loss of INA expression is associated with poor prognosis for people with nervous system malignancies (17–19), but its functional significance and underlying mechanisms in cancer remain undetermined. Here, we found that INA is commonly unmethylated in normal colon epithelium but densely hypermethylated in adenoma and adenocarcinoma lesions. This hypermethylation process may be an event of intermediate filament reorganization and play a potential role during colorectal cancer evolution. We conducted the first study on INA methylation in colorectal cancer to illuminate its biological functions and underlying mechanisms, as well as excavate its latent clinical implications on this malignant disease.

The process to attain 30 pairs of adenocarcinoma specimens and matched adjacent noncancer samples from donor patients with stage I–IV colorectal cancer was compliant with procedures approved by the Institutional Review Board of Sun Yat-sen University (Guangzhou, China). Colon adenoma polyps were resected from 37 patients who underwent endoscopic polypectomy. The abovementioned specimens were frozen in liquid nitrogen when we collected and then stored at −80°C. Formalin-fixed and paraffin-embedded (FFPE) blocks (n = 269) of colorectal cancer tissues were randomly collected from pathology archives in the Sixth Affiliated Hospital of Sun Yat-sen University (Guangzhou, China). All patients included in this study consented to use the specimens and signed written informed consent agreements.

Infinium HM450K arrays (Illumina) were used to profile the DNA methylome from the normal colonic mucosa (n = 41), adenoma (n = 42), and adenocarcinoma (n = 64) tissues. Detailed information on the platform, sample preparation, and genome-wide data analysis were described previously (6).

Genomic DNA from tissue samples and cultured cells was extracted using a DNA Extraction Kit (QIAGEN) following the recommended protocols in the supplemental manufacturer's instructions. An FFPE sample-specific Kit (QIAGEN) was used to extract the DNA from the FFPE tissues. Purified DNA was denatured and treated with sodium bisulfite to deaminate the unmethylated cytosine into uracil using a DNA Methylation Kit (Zymo Research, Irvine). The quantitative methylation-specific PCR (qMSP) assays were carried out using HotStarTaq DNA polymerase (QIAGEN), as we described previously (20). Briefly, the bisulfite-converted DNA was added to the reaction mix consisting of custom-designed primers (20 μmol/L), FAM/NED-labeled probes (100 μmol/L), PCR buffer (1 ×), MgCl₂ (1.5 mmol/L), dNTPs (200 pmol/L), and HotStarTaq DNA Polymerase (1.5 U). The thermocycler conditions were as follows: 95°C × 15 minutes, 45 × (94°C × 30 seconds, 57°C × 90 seconds, 72°C × 90 seconds), 72°C × 10 minutes. The reactions were run in triplicate on a QuantStudio 7 Flex PCR system. A repetitive Alu sequence (AluC4) was used as an internal loading control. The percent of methylation rate was calculated by setting the standard template as CpGenome methylated DNA (Sigma). We used duplex qMSP to detect the methylation status of INA and AluC4 simultaneously. According to a pilot study, no cross-interference was observed in the duplex qMSP with respect to single qMSP.

The bisulfite-converted DNA was amplified by forwarding primers and biotinylated reversed primers designed by PyroMark Assay Design 2.0. Biotin-labeled amplicons were captured by streptavidin-coated sepharose beads (GE Healthcare), and denaturation procedures were followed to produce single-stranded DNA. For pyrosequencing, the sequencing primer of INA was added to the mixture of pyrosequencing reactions, and the procedure was performed on a PyroMark Q96 ID system (QIAGEN). The methylated CpG sites were quantified using PyroMark CpG Software 1.0.11 (QIAGEN).

Genomic RNA extraction was carried out using a spin column RNA Purification Kit (Sangon) following its recommended protocols. A PCR Master Mix Kit (TOYOBO) was used to remove genomic DNA and transcribe RNA into cDNA. PCR using custom-designed primers and cDNA templates were performed within an SYBR Green Mix system (Roche), as we reported previously (21). GAPDH was used for normalization. A reference cDNA (TaKaRa) was served as a standard for calculating the relative quantitative PCR.

All cell lines were obtained from the ATCC, and cultured in ATCC-recommended media (Gibco) supplementing with FBS (Gibco; 10%, v/v), penicillin (10 U/mL), and streptomycin (10 μg/mL). The cells were authenticated by blasting the short tandem repeat (STR) loci in the ATCC STR database, and no cross-contamination was found. We also confirmed that the cells were Mycoplasma-free by PCR evaluation using a Mycoplasma detection set (TaKaRa) within 6 months of their use.

Extracted cell lysates were separated by SDS-PAGE and immunoblotted following classic procedures, as we described previously (22). Primary antibodies targeting INA (2E3; Thermo Fisher Scientific), GAPDH (1E6D9; Proteintech), and GFP (Sigma) were used for immunoblotting.

Cells were treated with 5-Aza-2′-deoxycytidine [5-AZA (Sigma)] following previously published protocols (20).

To silence INA, we constructed INA-targeted short hairpin RNA (shRNA) or scrambled shRNA into a pLVX-shRNA2-puro lentiviral vector between BamH I and EcoR I sites as we described previously (21). In the reconstitution study, commercial lentiviral vectors expressing wild-type INA (NM_032727.3), INA-GFP, mutant INA (10–153 deletion), TUBB-GFP (NM_178014), EB1-GFP (NM_012325.2), and matched empty vectors were custom created by Sino Biological. The sequences expressing the INA-tubulin binding site (aa 4–51) and sequence-rearranged control tagged with StrepII were synthesized (Sino Biological) and inserted into the pLV plasmid. The constructs were confirmed by direct sequencing in an independent third party (Sangon) before use.

For lentivirus production, the constructed vectors were cotransfected with psPAX2 and pMD2G into 293T cells by supplementing polyethyleneimine (Sigma), which two particles were kindly provided by Huanliang Liu (Sun Yat-sen University, Guangzhou, China). The virus in the cultured supernatant was collected at 48 hours for transfection. The cells were infected in the medium containing virus for 72 hours. The transfected cells were grown for 10–14 days in medium supplemented with puromycin to select the stably transfected cells.

In 96-well plates, 5,000 cells per well were plated and cultured for 24 hours. Cell confluence monitoring was performed intermittently for 2 hours over 5 days using an IncuCyte system. The confluence of the cells relative to the whole single-well area was calculated using a cell analysis module in the complementary software of the IncuCyte system.

Transfected cells were seeded in 6-well plates at a concentration of 200 cells per well and cultured from 10 to 14 days, and then fixed with 4% paraformaldehyde. Stained colonies with crystal violet (Sigma) were visualized and imaged by an Olympus DP27 microscope.

Standard protocols have been described previously (22).

Male BALB/c nude mice (4–6 weeks old) were purchased from the Vital River. The mice were fed and treated following the guidelines approved by the Sun Yat-sen University Institutional Animal Care and Use Committee. INA-shRNA–based stably transfected CACO2 cells (1 × 10⁷ cells/mouse), INA-overexpressing DLD1 cells (5 × 10⁶ cells/mouse), and their parental cells mixing with Matrigel (25%, v/v) were subcutaneously seeded into mice. Tumor diameters were measured with an electronic caliper, and tumor volumes were calculated with a standard formula that has been described previously (22). Excised tumor tissues were weighed, and then IHC staining was conducted using the Ki67 antibody (8D5; Cell Signaling Technology).

In the vein metastatic model, transfected DLD1 cells (1 × 10⁶ cells/mouse) expressing GFP-INA or GFP were injected into the mice tail veins for a month. The resected lung tissues were harvested and scanned with an IVIS system (Caliper Life Sciences) to determine the GFP fluorescence. Hematoxylin–eosin staining was conducted to detect lung metastases. The pulmonary tumor burden was quantified by ImageJ software following previously reported protocols (23).

A Tubulin Polymerization Assay Kit (Cytoskeleton) was used according to the manufacturer's directions to measure microtubule polymerization. To determine the activity of INA and tubulin-binding peptides on microtubule polymerization, pure tubulin proteins (3 mg/mL) were diluted in the tubulin polymerization buffer and pipetted into the half area 96-well plates. Recombinant GST-tagged INA (Creative Biomart), GST (Sino Biological), custom-created peptides (Sangon), and nocodazole (Selleck) were added to the mixtures. Signals were recorded within 60 minutes after reaction initiation according to optical density (OD)-based measurements taken every minute with a Varioskan Flash reader (Thermo Fisher Scientific) at 37°C.

The cells were fixed with paraformaldehyde, and the membrane was permeated with Triton X-100, followed by blocking using goat serum (Thermo Fisher Scientific). α-tubulin antibody (Proteintech) and fluorescent secondary antibody were used to label intracellular microtubules. The nuclei of the cells were stained with DAPI. The fluorescence was captured by a Leica TCS SP8 microscope.

The monitoring of dynamic end-binding protein 1 (EB1) has been established to measure microtubule end-plus assembly rates (24). For detecting the microtubule assembly rates in transfected DLD1 cells, EB1-GFP comets in living cells were monitored by time-lapse imaging in 40 frames (0.68 frames/second) using a Leica TCS SP8 microscope. The maximum projections of the EB1-GFP time-lapse tracking in 40 frames were merged with Leica LAS AF Lite software. The trajectory length of each comet was measured by using the Mtrack2 plugin in ImageJ software (https://imagej.nih.gov/ij/).

The peptide arrays were produced by INTAVIS. Partially overlapping peptides (15 aa) with a 3-aa frameshift from the N-terminus of INA to the C-terminus were synthesized to spot onto arrays. The identified tubulin-binding sites of mouse neurofilaments (NEFL, NEFM, and NEFH) were set as positive controls (25). Three scrambled peptides (G23, H4, and H15) were also spotted on the arrays. All the peptides were acetylated in the N-termini and covalently linked to cellulose on the surface of membranes through C-terminal amino acids. The specific sequencing peptides and distribution are represented in Supplementary Table S1, and the theoretical isoelectric point (pI) of each peptide was computed by ExPASy (https://web.expasy.org/compute_pi/). The binding assay was used to create a profile according to previously reviewed protocols (26). In brief, the slides were saturated overnight at 4°C with a blocking reagent containing 10% skimmed milk, followed by the incubation of the tested proteins that diluted in TBS buffer with 5% skimmed milk for 1 hour at room temperature. The recombinant proteins for the binding assay were as follows: α-tubulin (Proteintech), β-tubulin (Abcam), βIII-tubulin (Novus), His (Sangon), and GST (Sino Biological). The peptide–protein binding was detected by following the procedures of standard immunoblotting methods using primary antibodies included β-tubulin (Proteintech), βIII-tubulin (D71G9; Cell Signaling Technology), α-tubulin (1E4C11; Proteintech), His-tag (D3I10; Cell Signaling Technology), and GST (91G1; Cell Signaling Technology). Sequences with at least three consecutive peptides positively bound to all three tubulin subtypes were considered valid binding sites.

The statistical analyses of the differences in the intergroup comparisons were based on a two-tailed Student t test, while one-way ANOVA with Tukey post hoc test was applied to analyze data that included more than two groups. For evaluating the clinical significance of INA methylation, the points with the most significant split in the log-rank test were selected as the cut-off in the two cohorts, as we reported previously (27). On the basis of the cut-off, each cohort was divided into two groups: the high and low INA methylation. Kaplan–Meier analysis was performed to plot survival curves, and a log-rank test was used to compare the statistical significance. A χ² test was performed to statistically analyze discrete variables. Univariate and multivariate analyses based on the Cox proportional hazards regression model were used to estimate the potential risk factors associated with the clinical variables. Statistical tests of INA methylation and its expression or patient age were analyzed on the basis of Pearson correlation. All statistical analyses were performed using SPSS 21.0. Data are represented as the means ± SD, unless otherwise stated. *, P < 0.05; **, P < 0.01; and ***, P < 0.001 are considered statistically significant, and “ns” indicates not statistically significant (P > 0.05).

The complete details of the resources are available in the Reagent Table in the Supplementary Information.

First, we looked for abnormally methylated genes encoding cytoskeletal proteins in colorectal cancer. Then, we filtered the data from the HM450K arrays that had been profiled to assess aberrantly methylated genes in normal mucosa, adenoma, and adenocarcinoma tissues (6). The methylation status of the probes located in promoters, gene bodies, and intergenic regions was assessed. In a total of 577 probes targeting 37 cytoskeletal genes, we found 227 differentially methylated probes (DMP) in adenoma and 101 DMPs in adenocarcinoma tissues with a P < 1E-5 value when compared with the normal panel (Supplementary Fig. S1). Notably, the CpGs of ACTA1 (actin, alpha 1), DES (desmin), NEFL (neurofilament light), and INA (α-internexin) located in islands were frequently hypermethylated in the colon neoplasms (Supplementary Fig. S1). Among these screened candidates, INA was one of the top deregulated genes with a high fold change in colorectal cancer (Fig. 1A). According to the probes targeting the INA gene located on 10q24.33, INA was intensively hypermethylated in the promoter region in both adenoma and adenocarcinoma specimens, especially at the CpG islands, while it was ordinarily methylated at low levels in normal tissues (Fig. 1A). Inversely, the methylation levels of three probes located in the S-Shore, S-Shelf, and 3′ untranslated region (UTR) did not show aberrant hypermethylation (Fig. 1A).

To validate the methylated alteration of INA, we analyzed INA methylation status in biospecimens included normal colon (n = 30), adenoma (n = 37), and adenocarcinoma (n = 30) using qMSP. In line with the results from the HM450K arrays, both adenoma and adenocarcinoma carried markedly hypermethylated INA in the promoter region (Fig. 1B). We further assessed INA methylation derived from The Cancer Genome Atlas (TCGA) datasets that were obtained using Illumina Infinium (HumanMyethylation 27) arrays (2). As expected, the methylation status of a probe (probe no. cg24680586) covering the amplicons of the custom-designed qMSP was also abnormally high in cancer samples when compared with the matched normal colon tissues (Fig. 1C). A higher proportion of methylated CpGs in representative adenoma (67.08%) and adenocarcinoma (57.79%) than that in a normal epithelium (15.14%) was observed when the methylated CpGs were quantitatively measured by the pyrosequencing (Fig. 1D). In colorectal cancer cell lines, INA methylation data were retrieved from the Cancer Cell Line Encyclopedia (CCLE) databases that the DNA methylation was profiled by HM450K arrays (28), showing that CpGs located in the islands (chr10:105036628–105038084) were densely methylated in a majority of colorectal cancer cell lines (Supplementary Fig. S2). We further used pyrosequencing to confirm the dense INA hypermethylation in HT-29 (90.34%), DLD1 (89.55%), and RKO (91.24%) cells (Fig. 1D).

INA methylation in a panel of 269 colorectal cancer samples was determined using qMSP to evaluate its clinical significance. A cut-off with the most significant split in the log-rank test was selected to divide the cohort into the high and low INA methylation groups. We firstly analyzed the correlation between INA methylation and multiple clinical manifestations (Table 1). Noticeably, the proportions of elderly patients (P = 0.003) and large tumor sizes (P < 0.0001) in the INA-hypermethylated group were aberrantly high at a statistical level (Table 1). Clinically, large tumors are generally more invasive, leading to worse clinical outcomes. We therefore performed an intergroup comparison with analyze the variable INA methylation status in large and small tumors in both the adenocarcinoma and adenoma panels. As expected, the INA-hypermethylation burden in the large adenocarcinomas (largest diameter ≥ 3 cm) was worse than that in the small adenocarcinomas (largest diameter < 3 cm; P < 0.0001; Fig. 1E). A similar comparison highlighted the aggravated INA hypermethylation when compared advanced adenomas (largest diameter ≥ 1 cm) to early adenomas (largest diameter < 1 cm; P = 0.0051; Fig. 1F). Those results illustrate that large adenomas and adenocarcinomas carry heavier INA hypermethylation burden than small-sized tumors.

We next asked whether INA methylation is associated with the survival of patients. The Kaplan–Meier analysis showed that patients whose specimens belong to the INA-hypermethylated tumors exhibit poor overall survival with a decline survival curve (P = 0.0273; Fig. 1G). The worse outcome of those exhibiting high methylation levels was confirmed in the TCGA cohort (P = 0.0339; Fig. 1H). According to the univariate Cox regression analysis, age, pN status, liver metastasis, and INA methylation were associated with prognosis (Table 2). Multivariate Cox regression analysis revealed INA methylation as an independent prognostic factor for poor survival after the potential confounding factors including age, pN status, and liver metastasis were controlled (Table 2).

DNA hypermethylation in promoter regions generally contributes to expressive suppression of tumor-suppressive genes. To investigate the impact of INA hypermethylation on expression, we first assessed the relationship between INA methylation and mRNA expression in the TCGA datasets. The methylation status was inversely correlated with the corresponding mRNA expression in colorectal cancer tissues (P < 0.0001; Fig. 2A). We tested the INA methylation status and expression in a panel of colorectal cancer cell lines using qMSP, RT-PCR, and Western blotting, respectively. According to the CCLE data (Supplementary Fig. S2), the CpG islands in HS698T, SNU503, and HS675T cells were low methylated, these cell lines would be ideal candidates for INA expression detection, however, their slow growth characteristics limited their use. Although HEK293 is a nonneoplastic cell line, it was used as the alternative to serve as the INA-overexpressed cells considering its abundant INA expression that was found in a previous study (29). Out of the 11 cell lines, high INA expression in low-INA methylated HEK293 cells could be detected (Fig. 2B). Inversely, a large majority of the tested colorectal cancer cell lines (HCT-8, DLD1, HCT-15, SW48, COLO205, HT-29, RKO, SW480, and SW620) held INA hypermethylation, showing with low protein and mRNA expression (Fig. 2B). We also found that the INA gene promoter in CACO2 cells was moderately methylated in the qMSP analysis, which was further validated with pyrosequencing (Fig. 2B; Supplementary Fig. S3). These results indicate that the hypermethylation of INA gene may silence its expression. To test this hypothesis, we treated three INA-hypermethylated cells (DLD1, SW620, and RKO), the moderate-INA methylated CACO2 cells, and the low-INA methylated HEK293 cells with 5-AZA, a DNA methyltransferase inhibitor, to demethylate the overall levels of DNA methylation. INA methylation levels were markedly decreased after 5-AZA treatment in the INA-methylated cells (DLD1, SW620, RKO, and CACO2), but only slightly reduced in the low-INA methylated HEK293 cells (Fig. 2C). INA mRNA expression levels were markedly restored in 5-AZA-treated DLD1 (30.1 folds), SW620 (756.1 folds), and RKO (1064.8 folds) cells, but the expression was only slightly upregulated in CACO2 cells (2.3 folds) and does not significantly change in HEK293 cells (0.96 folds; Fig. 2D), indicating that the promoter hypermethylation silences INA expression. In support of this idea, we assessed the INA mRNA expression levels in clinical biospecimens. Compared with the status of INA mRNA in normal colon epitheliums, a pronounced loss of INA expression could be observed in the colon adenoma polyps and cancer tissues (Fig. 2E). These results were correlated well with the data from the TCGA datasets (Fig. 2F), supporting an idea that the epigenetic inactivation of INA is an early event in colorectal cancer progression.

To elucidate the biological function of INA in colorectal cancer, we stably transfected the shRNA vector targeting INA into CACO2 and COLO320DM cells, both of which are INA-expressing cells (Fig. 3A). Strikingly, the INA-repressed CACO2 and COLO320DM cells were more unchecked growth than the cells expressing scrambled shRNA (Fig. 3B). In the colony formation assays, the large-sized colonies and increased clone-forming rates were observed in the INA-depleted CACO2 cells (Fig. 3C and D). The wound scratch assays displayed a marked gain in the wound-healing ability of the INA-shRNA–transfected CACO2 cells (Fig. 3E; Supplementary Fig. S4E). Moreover, downregulation of INA markedly potentiated the capacity of CACO2 cells to invade or migrate across microporous membranes (Fig. 3F and G; Supplementary Fig. S4H). These results demonstrate that INA depletion promotes colorectal cancer cell proliferation, invasion, and migration. Furthermore, INA was reconstituted into DLD1 and RKO cells for additionally validating its biological functions (Fig. 3A). Even though the viral INA overexpression did not affect cell proliferation in vitro (Supplementary Fig. S4A–S4D), the reconstitution significantly decreased the wound-healing ability, as well as the invasive and migratory capacity of the DLD1 and RKO cells (Fig. 3H–J; Supplementary Fig. S4F, SG, SI, and SJ).

The biological effects of INA were further tested on immunodeficient BALB/c nude mice. Notably, the xenograft tumors derived from INA-depleted CACO2 cells exhibited heightened acceleration on tumor growth (Fig. 3K and L). Analyses of the proliferating cells via staining Ki67 in the xenograft tumor tissues showed densely positive staining in the INA-depleted cells (Supplementary Fig. S4K). We confirmed the maintenance of INA knockdown in INA-shRNA–expressed xenograft tumors (Supplementary Fig. S4L). In subcutaneous xenografts derived from transfected DLD1 cells, the growth was not markedly altered upon INA reconstitution, according to the growth curves, tumor weights, and Ki67 staining of the xenograft tumors (Fig. 3M and N; Supplementary Fig. S4M and S4N). The results were in full concordance with those of the functional assays in vitro. Intriguingly, after DLD1 cells expressing INA-GFP were injected into the tail veins of athymic mice, the INA reconstitution robustly depressed the ability of the cells to colonize the lungs, forming metastatic lesions (Fig. 3O). The tumor burdens of metastatic DLD1 cells overexpressing INA were consistently reduced in the lungs (Fig. 3P). Taken together, these results reveal INA as a tumor suppressor in vitro and in vivo.

As a starting point, tubulin polymerizing activities of mouse intermediate filaments have been identified, including neurofilaments (NEFL, NEFM, and NEFH) and keratin (25). In light of their structural similarities, we evaluated the effect of INA on microtubule polymerization using recombinant human INA proteins. Nocodazole, a well-known antimitotic agent that disrupts microtubule polymerization, potently inhibited tubulin polymerization in the tubulin polymerization assays (Fig. 4A). Similarly, the assays showed a potent inhibitory effect of recombinant INA proteins on polymerization in a concentration-dependent manner, while GST fragments tagged on the recombinant proteins did not markedly affect the assembly (Fig. 4A). Considering the in vitro inhibitory activity of INA, we next investigated its effect on intracellular microtubules in colorectal cancer cells. The immunofluorescence staining α-tubulin showed many intracellular microtubules were disrupted upon INA overexpression in the DLD1 and RKO cells (Fig. 4B and C). We also tracked cellular EB1-GFP, a plus end-binding protein that binds to plus-end dynamic microtubules during elongation, in living colorectal cancer cells to evaluate the impact of INA on intracellular microtubule plus-end assembly rates. The merged maximum intensity projection of the time-lapse EB1-GFP comets in the INA-overexpressing cells displayed visibly shorter motion trajectories than empty vector–expressing cells when the time in which 40 frames of images are merged (Fig. 4D). A quantitative measure of the EB1-GFP comets' trajectory length exhibited a small but significant decrease in their length upon INA overexpression (Fig. 4E). These results suggest that INA blocks tubulin polymerization in vitro and hinders intracellular microtubule polymerization.

Next, we sought to identify the microtubule-binding sites within INA. The partial overlapping peptides of INA sequence and the well-identified tubulin-binding sites of NEFL, NEFM, and NEFH were spotted on the peptide arrays, followed by cotreatment with three tubulin subtypes (β-tubulin, βIII-tubulin, and α-tubulin), and those peptide-binding tubulins were detected using the classic Western blotting protocols (Fig. 5A and B). Consequently, these tubulin subunits were successfully bound to most peptides at the identified tubulin-binding sites in NEFL, NEFM, and NEFH, while the binding affinity of the scrambled peptides was relatively weak (Fig. 5C). The peptide arrays were also incubated with GST and His tags that were tagged in the recombinant tubulin subunits to exclude the possibility of nonspecifically binding, consequently, none of those fragments bound to the peptides on the arrays (Supplementary Fig. S5A and S5B). Notably, three tubulin-binding motifs in the INA sequence were identified as shown by an concurrent binding to variable tubulins with at least three consecutive peptides, two of which were located in the head domain (INA-TBS.4–51 and INA-TBS.55–75), with the other was a part in the C-terminal tail domain (INA-TBS.424–447; Fig. 5C). To test whether those identified tubulin-binding sites are responsible for microtubule depolymerizing activity of INA, we artificially synthesized them to evaluate their depolymerizing activities using in vitro tubulin polymerization assays. Similar to nocodazole and full-length INA, INA-TBS.4–51 robustly inhibited microtubule polymerization in a dose-dependent fashion (Fig. 5D). However, INA-TBS.55–75 and INA-TBS.424–447, two identified tubulin-binding motifs with a short sequence in length, did not effectively block the polymerization (Supplementary Fig. S6A and S6B). Hence, we mainly focused on the role of INA-TBS.4–51 in the tumor-suppressive function. Vectors expressing INA-TBS.4–51 or a scrambled peptide, which all of them were fused with a Strep II tag, were transfected into DLD1 and RKO cells to determine the effect of INA-TBS.4–51 on microtubule dynamics and cell migration (Supplementary Fig. S7A and S7B). Similar to full-length INA overexpression, the intracellular microtubule assembly rates were inactivated upon INA-TBS.4–51 overexpression based on the analysis of dynamic EB1-GFP in living DLD1 and RKO cells (Fig. 5E and F). In addition, INA-TBS.4–51–expressing DLD1 and RKO cells exhibited weak migratory abilities when compared with the scrambled peptide-expressing cells (Fig. 5G). These observations indicate that INA-TBS.4–51 markedly inhibited intracellular microtubule assembly rates and cell migration. In contrast, a truncated INA with INA-TBS.4–51 motif deletion abolished its ability to inhibit cell migration, while wild-type INA was still sufficient to restrain migrated cells (Fig. 5H and I). The tubulin-binding activity of INA and INA-TBS.4–51 enlightened us overexpressing TUBB-GFP in well-established DLD1 and RKO cells to saturate INA tubulin-binding capacity (Fig. 5J). Expectedly, the intervention markedly restored the migration ability of INA-overexpressed DLD1 and RKO cells but had no significant effect on the parental cells (Fig. 5K). Collectively, we identify a microtubule-binding motif in the INA head domain that exerts a tumor-suppressive function via binding to unpolymerized tubulins and inhibiting microtubule polymerization in colorectal cancer.

In this study, we found an aberrantly hypermethylated phenotype in the INA gene promoter region that is a prognosis-associated marker in colorectal cancer. INA is preferentially hypermethylated in large adenomas and adenocarcinomas, and the hypermethylation is an independent predictor for poor overall survival of patients with colorectal cancer, indicating that the methylation has a critical role in colorectal cancer. Previous publications have reported that the loss of INA is associated with poor survival in patients who have nerve system cancers and neuroendocrine neoplasms (18, 30–32), suggesting that INA may be a tumor suppressor in cancer. However, the mechanisms of INA loss and its biological functions in cancer have not been determined to date. Here we show that INA hypermethylation in CpG islands causes the loss of expression in colon adenoma and adenocarcinoma tissues, the INA mRNA expression levels in adenoma polyps even lower than that in adenocarcinoma tissues. Individual differences and tumor heterogeneity may cause alterations in adenoma samples to be more obvious than those in adenocarcinoma tissues. We cannot exclude the possibility that the alterations of methylation processes contribute to the decline of INA methylation, reloading its expression after adenomas transform into cancers. Despite that, it is convincing that the epigenetic inactivation of INA is one of intermediate filament reorganization events occurs in the early stage of colorectal cancer progression, which in turn endows colorectal cancer cells with the properties necessary to migrate, invade, survive, and thrive, consequently results in poor patients' survival. Mechanistically, INA has unprecedented biological activity in inhibiting microtubule polymerization mainly via a binding motif in the head domain, thereby blocking cancer cell migration. Indeed, tumor cell metastasis is extremely sensitive to microtubule dynamics that play an essential role in cellular processes, including malignant proliferation and division, motility, and intracellular trafficking. For example, microtubule-targeting agents at the low concentrations that are not sufficient to prevent mitosis and cannot cause microtubule depolymerization, still effectively inhibit cancer cell migration by blocking microtubule dynamics (12). Our results demonstrate how INA is likely to function on microtubule assembly and microtubule dynamics, exerting its tumor-suppressive role in the early stage of colorectal cancer progression.

The increase in microtubule assembly rates represents a tumorous characteristic. It has been reported that the loss of tumor suppressor and the upregulation of oncogene increase microtubule plus-end assembly rates in colorectal cancer, subsequently triggering chromosomal instability via disrupting proper segregation of mitotic spindles (33). Similarly, here we provide evidence that INA, a newly identified tumor suppressor in colorectal cancer, hinders microtubule polymerization and intracellular microtubule dynamics. We further assessed the methylation status and expression of INA in the TCGA datasets, INA hypermethylation and the loss of INA expression were also found in other common cancer types, including glioma, breast cancer, gastric cancer, esophageal cancer, liver cancer, lung cancer, and cervical cancer (Supplementary Fig. S8). Our results demonstrate that, at least in part, the acceleration of microtubule assembly rates is triggered by the epigenetic inactivation of INA in colorectal cancer and other cancer types.

Age is a high-risk factor for colorectal cancer development (34). Indeed, age-related hypermethylation in aging colonic mucosa is highly correlated with cancer risk (9, 35, 36). Notably, we observed an abnormally high proportion of older patients with colorectal cancer (P = 0.003) in the INA hypermethylated subgroup, as indicated by the χ² test (Table 1). We further analyzed the correlation between INA methylation and chronological age in both normal epithelium and colorectal cancer. Interestingly, INA methylation is positively correlated with age in normal colon epithelium in the testing cohort (R² = 0.4733, P = 0.0007) and the TCGA cohort (R² = 0.5463, P < 0.0001; Supplementary Fig. S9A). Such age-related methylation of tumor suppressor may predispose normal colon mucosa to transform into colorectal cancer during aging. Similarly, the hypermethylation of multiple tumor-suppressive genes is shown to be age dependent in the normal colonic mucosa tissues, such as MutL homolog 1 (MLH1), estrogen receptor 1 (ESR1), insulin-like growth factor 2, and tumor suppressor candidate 3 (TUSC3; refs. 37–39). Although the mechanisms of such age-dependent DNA methylation remain unknown, most of those genes have been identified to be involved in tumorigenesis, supporting a potential role of INA in the increase of tumor susceptibility during aging (4). In the colorectal cancer samples, INA methylation is also correlated with chronological age, but the linear correlations are largely disrupted both in the testing cohort (R² = 0.0285, P = 0.0055) and the TCGA cohort (R² = 0.0435, P = 0.0005; Supplementary Fig. S9B). The trend fits a reported discovery unearthing a significant acceleration of DNA methylation age (termed “DNAm age”) that exists in colorectal cancer and other cancer types, resulting in a weak correlation between DNA methylation and chronological age (40). Identifying individuals at high risk of colorectal cancer by monitoring DNAm age in normal colonic tissues may reveal potential implications of such age-related INA methylation.

Our current results further confirm the presence of tubulin-binding motifs in neurofilaments that could impede microtubule polymerization. A model has been proposed that neurofilaments harboring unpolymerized tubulins are likely to modulate the local availability of tubulins during microtubule polymerization, mainly supported by the facts that microtubule cytoskeletons in axons are enhanced in the NEFH-deficient mice and the tubulin-binding motifs disrupt intracellular microtubules (25). In colorectal cancer cells, we also observed that INA reconstitution widely disrupts intracellular microtubule cytoskeletons. In addition, INA directly binds to unpolymerized tubulins and acts as a newly identified tubulin reservoir. This capacity enables INA to orchestrate the assembly of intracellular microtubules, although the molecular mechanisms remain undetermined. In normal cells, the presence of INA might hinder exceptional microtubule assembly to prevent malignant proliferation. Hence, the epigenetic silencing of INA, causing unpolymerized tubulin release that accelerates microtubule assembly, may be a crucial step for early cancer progression.

A total of three tubulin-binding motifs are identified in the INA sequence. However, it is not yet clear why only INA-TBS.4–51 exhibits inhibitory activity of microtubule polymerization. The effect may depend on sequence with enough length to preserve sufficient capacity binding tubulins to inhibit microtubule polymerization. Moreover, a characteristic of basic arginine or lysine-rich tubulin-binding motifs with high isoelectric point (pI) values can be observed in INA and other neurofilaments (Supplementary Fig. S10A; Supplementary Table S1; ref. 25). Tau, a critical microtubule-associated protein, also possesses three microtubule-binding domains with high pI values (41). In contrast, tails of C-terminal tubulins that project away from the microtubule lattice packed dense acidic amino acids with negatively charged chains (42). The tubulin-binding sites may electrostatically interact with the C-terminal tubulin tails under physiologic conditions. This presumption was partially supported by our preliminary findings showing that tubulin-binding activities of peptides at the tubulin-binding sites of INA, NEFL, NEFM, and NEFH are pH-dependent, as indicated when the pH of the cotreated buffer was gradually adjusted using acetic acid in the peptide arrays binding assay (Supplementary Fig. S10B). Besides, the tubulin polymerizing activity of INA-TBS.4–51 is abolished when the polymerized reaction is initiated in an acidic system, while nocodazole still able to inhibit the polymerization in this acidic condition (Supplementary Fig. S10C). Charge alterations of serine, threonine, and tyrosine by phosphorylation and dephosphorylation are essential for the regulation of protein–protein interactions, which are the basis of switch-like responses in signaling transduction networks (43). The intracellular INA-tubulin interaction is probably subjected to the regulation of phosphorylation in the tubulin-binding motif that is serine-rich, but more in-depth investigations in future studies are necessary to fully determine its underlying mechanisms.

In summary, we have identified INA as a novel tumor suppressor gene in colorectal cancer, which is frequently inactivated by DNA hypermethylation in promoter regions. We have provided evidence that INA is preferentially hypermethylated in large tumors, and the hypermethylation is associated with poor prognosis. We identified a tubulin-binding motif located in the N-terminal head domain of INA that inhibits microtubule polymerization and decreases intracellular microtubule assembly rates, exerting its tumor-suppressive function. This study adds DNA methylation to the complex mechanisms of microtubule assembly in tumorigenic process.

W.M. Grady reports other compensation from SEngine (advisory board), Freenome (advisory board), Guardant Health (advisory board), DiaCarta (consultant), and Cellixbio (consultant) outside the submitted work. No disclosures were reported by the other authors.

Y. Li: Data curation, investigation, writing-original draft. L. Bai: Resources, validation, investigation. H. Yu: Funding acquisition, methodology, writing-review and editing. D. Cai: Formal analysis, methodology. X. Wang: Data curation, supervision, funding acquisition. B. Huang: Investigation. S. Peng: Formal analysis, investigation. M. Huang: Funding acquisition. G. Cao: Conceptualization, funding acquisition. A.M. Kaz: Conceptualization, resources. W.M. Grady: Conceptualization, resources. J. Wang: Funding acquisition. Y. Luo: Conceptualization, funding acquisition, project administration, writing-review and editing.

This work was supported by the National Basic Research Program of China (973 Program; 2015CB554001 to J. Wang), National Natural Science Foundation of China (81972245 to Y. Luo; 31900505 to Y. Li; 81902877 to H. Yu), Natural Science Fund for Distinguished Young Scholars of Guangdong Province (2016A030306002 to Y. Luo), “Five Five” Talent Team Construction Project of the Sixth Affiliated Hospital of Sun Yat-sen University (P20150227202010244 to J. Wang; P20150227202010251 to Y Luo), Sun Yat-sen University Clinical Research 5010 Program (2018026 to Y. Luo), Natural Science Foundation of Guangdong Province (2018A030313567 to Y. Li; 2016A030310222 to H. Yu; 2018A0303130303 to H. Yu), Program of Introducing Talents of Discipline to Universities, and National Key Clinical Discipline (2012), and China Postdoctoral Science Foundation (2018M633241 to Y. Li).

The authors would like to acknowledge the TCGA Research Network (https://www.cancer.gov/tcga) and the Broad Institute Cancer Cell Line Encyclopedia (CCLE; www.broadinstitute.org/ccle) for the results shown here are in part based upon data they generated. We also wish to appreciate Huanliang Liu (Sun Yat-sen University, Guangzhou, Guangdong, China) who kindly shared plasmids for the lentivirus packaging.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.