Microarray experiments were analyzed with Oncomine (21) and respective data downloaded from www.oncomine.org. The Human Protein Atlas (22) served as a source for survival data (www.proteinatlas.org) with the corresponding RNA sequencing data originating from ‘The Cancer Genome Atlas’ (TCGA). To assess coexpression of ATRX and JMJD1A in colorectal adenocarcinomas, provisional TCGA RNA sequencing data were analyzed with cbioportal (www.bioportal.org).

The retroviral vector pSIREN-RetroQ (Clontech, Palo Alto, CA, USA) was linearized with the restriction enzymes BamHI and EcoRI and ligated with double-stranded oligonucleotides encoding shRNAs (23). Correct cloning was verified by DNA sequencing. The control shRNA targets the sequence 5′-CAACAAGATGAAGAGCACCAA-3′, which displays at least four mismatches to any known human gene. The human JMJD1A shRNAs target the sequences 5′-GCAGGTGTCAATAGTGATA-3′ (#1 shRNA) and 5′-GTAGACCTAGTTAATTGTA-3′ (#2 shRNA), while the human ATRX shRNAs target the sequences 5′-GGTGTTATGATCATAGGCTAT-3′ (#1 shRNA) and 5′-GGATTCAACCTCTTGAGGATA-3′ (#3 shRNA).

Human colorectal carcinoma cells HCT116 (CCL-247), SW480 (CCL-228), DLD-1 (CCL-221) and HT-29 (HTB-38) as well as human embryonic kidney 293T cells (CRL-3216; all American Type Culture Collection, Manassas, VA, USA) were grown in a humidified, 5% CO₂-containing atmosphere in Dulbecco's modified Eagle's medium (10-013-CV; Mediatech; Corning Inc., Corning, NY, USA) that was supplemented with 10% fetal bovine serum (S11150; Atlanta Biologicals, Flowery Branch, GA, USA) as previously described (24,25). Transfection of 293T cells was done by the calcium phosphate coprecipitation method (26,27), the precipitate washed off with phosphate-buffered saline (28), retrovirus collected from the supernatant over the next 48 h (29) and in some cases concentrated by precipitation with poly (ethylene glycol)-8000 (30). HCT116 cells were infected with retrovirus three times (31) and then selected with 1.5 µg/ml puromycin for 3–4 days (32). To measure growth, 2,000 or 2,500 cells were seeded in 96-wells and growth determined essentially as described (33,34). For clonogenic assays, 1,000 or 3,750 cells were seeded into 6-wells and colony formation assayed as described (30).

To generate whole cell protein extracts, cells were lysed in Laemmli sample buffer and boiled for ~10 min (35). For biochemical fractionation of cells, the NE-PER nuclear and cytoplasmic extraction kit (78833; Pierce Biotechnology, Rockford, IL, USA) was employed as described (36). Proteins were then electrophoretically separated on SDS polyacrylamide gels (37), transferred to polyvinylidene difluoride membrane (38) and incubated with primary antibodies as described (39,40). Signals on blots were revealed utilizing appropriate secondary antibodies (41) followed by detection with chemiluminescence (42). Staining of a human colon cancer tissue microarray (AccuMax A303 I, slide #40; Isu Abxis, Seongnam, South Korea) was performed employing 20 min of antigen retrieval with method 2 and a 1:100 dilution of anti-JMJD1A antibody as described (43). The stained slide was digitized and the digital image extracted with Aperio ImageScope software (Leica, Wetzlar, Germany). Diaminobenzidine staining was obtained by color deconvolution from this image, intensity of light transmission was measured with Fiji version of ImageJ software (http://fiji.sc) and the strength of JMJD1A staining was defined as 100× (log maximum intensity-log intensity). The following rabbit polyclonal antibodies were utilized: Anti-Actin (A2066; Sigma-Aldrich; Merck KGaA, Darmstadt, Germany); anti-ATRX (NBP1-83077); anti-JMJD1A (NB100-77282; both Novus Biologicals, Littleton, CO, USA); and anti-H3K27me₁ (07–448; Upstate Biotechnology, Lake Placid, NY, USA). Also used was a goat polyclonal anti-Lamin B antibody (sc-6216; Santa Cruz Biotechnology, Santa Cruz, CA, USA).

Total RNA from HCT116 cells was isolated employing TRIzol (44) following standard procedures (45). RNA was further purified with the RNeasy Mini kit (74104; Qiagen, Hilden, Germany) and sequenced in the Targeted DNA Methylation and Mitochondrial Heteroplasmy Core at the Oklahoma Nathan Shock Center of Excellence in the Biology of Aging (Oklahoma City, OK, USA). Reads were aligned in Strand NGS (Agilent Technologies, Inc., Santa Clara, CA, USA) against the hg38 human genome build (December 2013) with Ensemble gene annotations (v87, January 2017). Three bases were trimmed from 3′ and 5′ ends of reads and read quality <Q20 was discarded prior to alignment. Alignment was to the full genome build to detect novel genes and splice variants and required 90 percent identity, minimum read length of 25, and reads with multiple matches were eliminated. A screening database for Illumina adapter sequences was used to remove adapter sequences. Reads counts were normalized by DESeq and a z-test was used to determine differential expression between samples. Ingenuity Pathway Analysis (Qiagen) was performed on genes whose mRNA levels were at least 1.5-fold different upon JMJD1A downregulation compared to the control shRNA treated cells.

The human ATRX promoter spanning from −600 to +100 (the ATRX transcription start site was based on the human transcript variant 1: NCBI reference sequence NM_000489.4) was cloned into pGL2-Basic (Promega Corp., Madison, WI, USA). Human 293T and HCT116 cells, which were grown in 12-wells, were transiently transfected with 100 ng of this luciferase reporter construct, 900 ng pBluescript KS⁺, and 60 ng of Flag-JMJD1A expression plasmid or empty vector pEV3S utilizing 2 µg polyethylenimine (43). Approximately 42 h after transfection, cells were lysed as described (46) and luciferase activities determined in a luminometer (47).

Human embryonic kidney 293T cells were grown in 10 cm dishes and transfected by the calcium phosphate coprecipitation method with 3 µg ATRX luciferase reporter plasmid, 21 µg pBluescript KS⁺, and 6 µg Flag-JMJD1A expression plasmid or empty vector pEV3S. Cells were treated with formaldehyde and processed for chromatin preparation and immunoprecipitation as described (48,49). The following antibodies were used: Normal mouse IgG (sc-2025; Santa Cruz Biotechnology); anti-H3K9me₂ mouse monoclonal antibody (ab1220); and anti- H3K36me₂ rabbit polyclonal antibody (ab9049; both Abcam, Cambridge, MA, USA). Resultant DNA was then amplified by PCR, which was performed with the GoTaq DNA polymerase kit (M3008; Promega Corp.) and the following temperature program: 97°C for 2 min; 8 cycles of 97°C for 25 sec, 65°C (−1°C per cycle) for 25 sec, 72°C for 35 sec; 26 cycles of 97°C for 25 sec, 57°C for 25 sec, 72°C for 35 sec (+1 sec per cycle); 72°C for 4 min followed by cooling down to 4°C. Primers used were ATRX-for2 (5′-GTAGGTTTGTCTACCTCAGAGAGTG-3′; spanning the ATRX promoter from −332 to −308) and ATRX-rev2 (5′-ACAGCTCAAAGGCCGCTACCACTGC-3′; spanning the ATRX promoter from +117 to +141). The PCR products were electrophoresed in agarose gels and revealed by ethidium bromide staining (50).

Statistical significance was assessed with one- or two-way analysis of variance (ANOVA) with post hoc Dunnett's or Tukey's multiple comparisons test, an unpaired Student's t-test, or a log-rank test. P<0.05 was deemed to show a statistically significant difference.

To examine the expression of JMJD1A mRNA in colorectal tumors, we first interrogated publicly available databases. In a study from TCGA (51), we found significant overexpression of JMJD1A in cecum, colon and rectal adenocarcinomas and a trend towards overexpression in mucinous tumors of the colon and rectum (Fig. 1A). Importantly, high JMJD1A mRNA levels were significantly correlated with reduced survival (Fig. 1B). Further, we discovered that JMJD1A was more highly expressed in patients with recurrent disease [Fig. 1C; microarray data retrieved from reference (52)] and at metastatic sites compared to the primary colorectal tumors [Fig. 1D; microarray data retrieved from reference (53)]. Altogether, these data indicate that JMJD1A mRNA is overexpressed in many colorectal tumors and associated with a more aggressive phenotype.

We then wanted to assess JMJD1A protein expression in colorectal cancer. Expression of JMJD1A protein was confirmed in all of the four tested human colorectal cancer cell lines by western blotting (Fig. 2A). Further, we biochemically fractionated HCT116 and DLD-1 colorectal cancer cells and found that JMJD1A was present in both the cytoplasm and nucleus, while very little JMJD1A was detectable in the insoluble fraction that primarily consists of the nuclear matrix and heterochromatin (Fig. 2B). Then, we stained a human tissue microarray consisting of 32 matching normal and cancerous colorectal tissues. Consistent with our biochemical fractionation experiments, JMJD1A staining was present in both the cytoplasm and nucleus. Importantly, a significant overexpression of JMJD1A was observed in the tumors (Fig. 2C and D), further implicating that JMJD1A overexpression may contribute to the development of colorectal cancer.

To assess a potential physiological role of JMJD1A, we downregulated JMJD1A with two different shRNAs in HCT116 colorectal cancer cells. Both shRNAs induced a large reduction of JMJD1A protein levels (Fig. 3A). While JMJD1A shRNA#1 did not cause any change in HCT116 cell growth, shRNA#2 displayed a significant, yet very small reduction in cell growth (Fig. 3B). This indicates that JMJD1A downregulation has only negligible effects on HCT116 cell growth. In contrast, clonogenic activity of HCT116 colorectal cancer cells was robustly reduced upon JMJD1A downregulation with either of the two shRNAs (Fig. 3C). The latter data indicate that JMJD1A can influence the physiology of HCT116 cells in a manner that is predicted to be tumor promoting.

Next, we assessed how JMJD1A affects the transcriptome of HCT116 cells. To this end, we again downregulated JMJD1A with two different shRNAs (Fig. 4A) and performed RNA sequencing. Compared to control shRNA, we found that 281 genes were >1.5-fold upregulated with both JMJD1A shRNAs and 192 genes were >1.5-fold downregulated (Fig. 4B). Ingenuity Pathway Analysis revealed that multiple metabolic (Fig. 4C) and upstream regulatory pathways (Fig. 4D) were affected by JMJD1A downregulation. This indicates that JMJD1A pleiotropically affects the gene expression program of HCT116 colorectal cancer cells and may thereby be a determinant of their oncogenic potential.

Since JMJD1A as a histone demethylase modulates chromatin structure, we were especially interested in other proteins involved in chromatin regulation whose genes were found to be affected by JMJD1A in our transcriptome analysis. One such gene was ATRX, which encodes for a chromatin remodeler (54). Our RNA sequencing data showed that JMJD1A downregulation by shRNA#1 or shRNA#2 led to a 2.6-fold or 3.5-fold decrease in ATRX mRNA levels, respectively (Fig. 5A). Accordingly, ATRX protein levels were also reduced in the presence of JMJD1A shRNAs (Fig. 5B). Please note that ATRX has a calculated molecular weight of 282.6 kDa and that the ATRX gene is composed of 35 exons, giving rise to multiple splice variants. This complex structure, and possibly protein degradation, accounted for the fact that multiple bands were detected with the anti-ATRX antibody in our western blot analyses. In total, these data suggest that JMJD1A can stimulate ATRX gene transcription. This notion is strongly supported by the fact that JMJD1A and ATRX mRNA levels were positively correlated across normal and malignant colorectal tissue specimens (Fig. 5C) or across adenocarcinomas in another data set (Fig. 5D).

To provide further evidence for a JMJD1A-ATRX axis, we cloned the human ATRX gene promoter in front of a luciferase reporter gene. As shown in Fig. 6A, overexpression of JMJD1A in 293T cells slightly stimulated the ATRX promoter. In addition, we also overexpressed JMJD1A-H1120A/D1122G, in which two amino acids within the JMJD1A catalytic center have been mutated rendering it inactive (1). This mutant repressed the ATRX luciferase reporter gene, presumably since it prevents endogenous JMJD1A from interacting with and thereby activating the ATRX promoter. Similar results were obtained with wild-type JMJD1A and its H1120A/D1122G mutant in HCT116 colorectal cancer cells (Fig. 6B). Moreover, we found that JMJD1A overexpression expectedly reduced dimethylation of histone H3 on lysine 9, but not dimethylation on lysine 36 (Fig. 6C). On the other hand, the H1120A/D1122G mutant predictably elevated levels of dimethylation on lysine 9 of histone H3. Collectively, these data suggest that JMJD1A interacts with the ATRX promoter and induces it by a mechanism that involves reduction of H3K9me₂ levels.

The previous findings led to the question of whether ATRX is overexpressed in colorectal cancer. Indeed, similar to JMJD1A, ATRX mRNA was significantly upregulated in cecum, colon and rectal adenocarcinomas, and even mucinous adenocarcinomas of the colon and rectum displayed significant ATRX upregulation [Fig. 7A; microarray data from reference (51)]. Excitingly, high ATRX expression was positively correlated with increased disease recurrence [Fig. 7B; microarray data from reference (55)] and lethality [Fig. 7C; microarray data from reference (52)]. These data suggest that ATRX might promote colorectal cancer.

We then downregulated ATRX with two different shRNAs in HCT116 cells (Fig. 8A). This led to a significant reduction of HCT116 cell growth (Fig. 8B). In addition, both ATRX shRNAs caused a reduction in clonogenic activity of HCT116 cells (Fig. 8C). These data are further supporting the notion that ATRX is a promoter of colorectal cancer.