Retrovirus was produced according to standard procedures (33). Viral supernatant was then used to infect 2-3 times HCT116 colorectal cancer cells (CCL-247; American Type Culture Collection), followed by selection with 1-2 µg/ml puromycin (34). To validate effectiveness of viral infection, protein extracts were prepared from stably transduced cells and assessed by Western blotting (35). Rabbit polyclonal antibodies against BHLHE40 [Novus, NB100-1800), ETV1 (#959 generated by our laboratory (36)], JMJD1A (Novus, NB100-77282), JMJD2A (Bethyl, A300-861A) and actin (Sigma, A2066) were used as primary antibodies. The following shRNAs were utilized: shBHLHE40#1 (GGAGACCTACAAATTGCCG), shBHLHE40#2 (GAACCGGATGCAGCTTTCG), shBHLHE40#4 (GAGCTCAGACTCAGAGGAA), shADAM19#2 (GTTGTACTGTGAACTGGAA), shADAM19#3 (GAGTGTGACTGTGGAGAAG), shKLF7#2 (GGAGTCTACAGTCCTTACT) and shKLF7#6 (GCCTTATAAGTGCTCATGG). The shRNA sequences for ETV1, JMJD1A and JMJD2A have been published before (16, 36, 37).

RNA was isolated with Trizol (38) and further purified with the RNeasy Mini kit (Qiagen, #74104). Utilizing random p(dN)₆ primers, RNA was reverse transcribed (GoScript Reverse Transcription; Promega, #A5000) and then examined in real-time with the AccuPower 2xGreenstar qPCR kit (Bioneer, #K-6251) in a BioRad light cycler according to the manufacturer’s recommendations. Sequences of RT-PCR primers are provided in Supplementary Table 1 . RNA sequencing was performed by Novogene and analysis of data done as described (39).

HCT116 cells were treated with formaldehyde and chromatin immunoprecipitations then performed as described (40). Antibodies used were control rabbit IgG (Santa Cruz, sc-2027), rabbit polyclonal anti-JMJD1A antibodies (Novus Biologicals, NB100-77282), anti-JMJD2A antibodies (Bethyl, A300-861A) or previously reported (36) anti-ETV1 antibodies. BHLHE40 promoter fragments were amplified with nested PCR (16) and then visualized through ethidium bromide staining (41) after agarose gel electrophoresis (42). Primers used for amplification are listed in Supplementary Table 2 .

HCT116 cells were grown in 12-wells and transiently transfected with the help of 2 µg polyethylenimine (43). A total of 0.25 µg luciferase reporter plasmid, 0.9 µg pBluescript KS⁺, and 0.1 µg of empty vector pEV3S or ETV1 expression vector were used for each transfection when comparing pBL2-Basic to BHLHE40-luc, while 0.1 µg BHLHE40 luciferase reporter plasmid, 0.9 µg pBluescript KS⁺, 80 ng empty vector pEV3S or ETV1 expression vector, and 60 ng of empty vector pEV3S or indicated JMJD expression vector were employed otherwise. Eight hours afterwards, cells were washed with phosphate-buffered saline, and cell extracts were prepared 1.5 days later (44). Luciferase activity was then determined in a luminometer as described (45).

Human embryonic kidney 293T cells (CRL-3216; American Type Culture Collection) were cultured in poly-L-lysine coated 6-cm plates (46). Upon reaching 25% confluency, cells were transiently transfected by the calcium phosphate coprecipitation method (47). Two days after transfection, cell extracts were prepared (48) and immunoprecipitations performed as previously described (49). Precipitated proteins were run on SDS polyacrylamide gels (50), transferred to polyvinylidene fluoride membrane (51), and then challenged with indicated antibodies (52). After incubation with corresponding secondary antibodies coupled to horseradish peroxidase (53), signals were revealed by chemiluminescence (54) and exposure to film (55). Likewise, endogenous proteins were coimmunoprecipitated from human HCT116 colorectal cancer cells, utilizing control rabbit IgG (Santa Cruz, sc-2027) or rabbit polyclonal anti-JMJD1A antibodies (Bethyl, A301-538A) for immunoprecipitation followed by Western blotting with anti-JMJD2A antibodies (Bethyl, A300-861A).

Transduced HCT116 cells were seeded into 96-well plates at a density of 2800 cells/well and their growth measured as described (56). For clonogenic assays, 2000 cells were seeded into 6-well plates, grown for approximately 10 days, and then stained with crystal violet (57).

Previously, we performed RNA sequencing on ETV1-depleted human HCT116 colorectal cancer cells and found downregulation of BHLHE40 mRNA (16). To corroborate this finding, we performed Western blotting with HCT116 cells in which ETV1 was downregulated with two different shRNAs. And we expectedly observed that BHLHE40 protein levels became reduced upon ETV1 ablation ( Figure 1A ). Likewise, we downregulated the histone demethylase JMJD1A, which is capable of binding to ETV1 and also reportedly affects BHLHE40 transcription (16), and observed concurrent decreases of JMJD1A and BHLHE40 protein levels ( Figure 1A ). Like JMJD1A, the histone demethylase JMJD2A also directly binds to ETV1 (37), prompting us to assess if BHLHE40 would similarly be regulated by JMJD2A. And indeed, BHLHE40 was downregulated at both the protein and mRNA level upon expression of two different JMJD2A shRNAs, and the degree of this downregulation was similar to that observed with ETV1 or JMJD1A shRNAs ( Figures 1A, B ). These data imply that BHLHE40 gene transcription can be jointly regulated by ETV1, JMJD1A and JMJD2A.

Consistently, chromatin immunoprecipitation assays revealed that all three of these proteins bound to several regions within the human BHLHE40 gene promoter ( Figure 2A ). Notably, the BHLHE40 promoter contains numerous potential ETV1 binding sites throughout its length ( Supplementary Figure 1 ), providing a possible explanation why ETV1 bound to all four tested regions of the BHLHE40 promoter. Further, we cloned the BHLHE40 promoter in front of a luciferase gene and then determined that ETV1 was capable of activating this reporter construct ( Figure 2B ). In contrast, neither ectopic JMJD1A nor ectopic JMJD2A alone activated the BHLHE40 luciferase reporter gene, but either one of these histone demethylases cooperated with ETV1 in doing so ( Figure 2C ). In contrast, catalytically inactive mutants of JMJD1A and JMJD2A were unable to cooperate with ETV1, and their expression alone even suppressed the BHLHE40 luciferase reporter gene ( Figure 2C ), possibly by competing with and thus precluding endogenous JMJD1A and JMJD2A to cooperate with endogenous ETV1. Overall, these data suggest that ETV1, JMJD1A and JMJD2A are jointly involved in the upregulation of the BHLHE40 gene promoter. In further support of this notion, BHLHE40 mRNA levels are positively correlated with ETV1, JMJD1A and JMJD2A in the 592 colorectal adenocarcinomas represented in The Cancer Genome Atlas (TCGA; Supplementary Figure 2 ).

Because ETV1 can bind to both JMJD1A and JMJD2A (16, 37), their joint regulation of BHLHE40 transcription may involve the formation of a ternary complex of all these three factors. This would be consistent with the fact that all of these three transcription factors interacted with the same regions A and B of the BHLHE40 gene promoter in our chromatin immunoprecipitation experiments ( Figure 2A ). This additionally prompted us to examine if JMJD1A and JMJD2A would also form complexes with each other and not only with ETV1. To this end, we coexpressed Flag-tagged JMJD1A and Myc-tagged JMJD2A, performed immunoprecipitations with anti-Flag antibodies and probed with anti-Myc antibodies if JMJD2A would coimmunoprecipitate. And this was the case ( Figure 3A ), and this interaction between JMJD1A and JMJD2A could be confirmed by switching the tags in another coimmunoprecipitation experiment ( Figure 3B ) as well as upon assessing the complex formation between endogenous JMJD1A and endogenous JMJD2A in HCT116 cells ( Figure 3C ). We then determined that the N-terminal half of JMJD1A (amino acids 2-650), which does not contain the catalytic JmjC domain nor a Zn-finger also required for catalytic activity, primarily mediated the interaction with JMJD2A ( Figure 3D ). Deletion of the N-terminal JmjN domain that is required for the catalytic activity of JMJD2A did not abolish binding to JMJD1A (see 61-1064 in Figure 3E ), but further deletion of the catalytic core JmjC domain did (see 301-1064). While this shows that the JmjC domain of JMJD2A is required for binding to JMJD1A, it is not sufficient since JMJD2A amino acids 2-350 did not bind to JMJD1A ( Figure 3E ). But JMJD2A amino acids 2-703 interacted with JMJD1A, indicating that the JMJD2A JmjC domain together with about 400 amino acids C-terminal of it jointly mediate the interaction with JMJD1A.

Ablation of ETV1 (16), JMJD1A (58–60) or JMJD2A (61) alone reportedly suppressed growth or clonogenic activity of HCT116 cells, indicating that all three factors are required for the full oncogenic potential of these colorectal cancer cells. Hence, we tested whether their joint target gene, BHLHE40, would also promote the growth and/or clonogenic activity of HCT116 cells. To this end, we downregulated BHLHE40 with three different shRNAs and observed that this resulted in both reduced cell growth and clonogenic activity ( Figure 4 ). While this manuscript was under preparation, another study similarly showed that a BHLHE40 siRNA was capable of reducing HCT116 cell growth (62). These data implicate that BHLHE40 in its own right is a tumor promoter in colorectal cancer. In agreement with this, analysis of published microarray data (63, 64) revealed that BHLHE40 mRNA levels are upregulated in colorectal tumors ( Figure 5A and Supplementary Figures 3A-D ) and may be a predictor of disease recurrence ( Figure 5B and Supplementary Figures 3E, F ) and disease-free survival ( Figure 5C ). Altogether, these data indicate that upregulation of BHLHE40 in human colorectal tumors may promote their growth and aggressiveness.

To gain mechanistic insights how BHLHE40 could affect colon cancer cells, we performed RNA sequencing on HCT116 cells that expressed three different BHLHE40 shRNAs. We found that this caused the down-/upregulation of 275/146 genes ( Figures 6A, B ). The fact that only 1.4% of all transcripts were significantly altered implies that BHLHE40 has a strong selectivity in its function as a transcriptional regulator. We then validated the RNA sequencing data by quantitative RT-PCR with a small selection of the differentially expressed genes (all downregulated in RNA sequencing data; Figure 6C ). Amongst those, we selected ADAM19, which encodes for a metalloproteinase, and KLF7, which encodes for a DNA-binding transcription factor, for further analysis, because their expression was positively correlated with BHLHE40 in 592 colon adenocarcinomas ( Figures 6D, E ) and little was known about their role in colon cancer.

First, we determined in published (63) microarray data that KLF7 mRNA is significantly upregulated in various subtypes of colorectal cancer ( Figure 7A ). Likewise, analysis of public RNA sequencing data revealed upregulation of KLF7 mRNA in colorectal adenocarcinomas at the primary tumor site ( Figure 7B ). Moreover, we found a significant association of KLF7 mRNA with disease recurrence in another published (65) microarray data set ( Figure 7C ). Further, high KLF7 levels are associated with reduced disease-free survival ( Figure 7D ). Altogether, these data suggest that KLF7 may promote colon tumorigenesis and may be a predictor of worse outcome. Similarly, we found that ADAM19 is significantly overexpressed in colorectal adenocarcinomas ( Figure 7E ), corroborating a previous study comparing a small number (n=14) of colorectal tumors to normal tissue (66). And high ADAM19 expression appears to also be predictive of worse disease-free survival, although this does not reach statistical significance ( Figure 7F ), suggesting that ADAM19 exerts pro-tumorigenic functions in colon cancer.

We then tested how downregulation of KLF7 or ADAM19 would affect HCT116 cells. Using two independent shRNAs, we achieved efficient ADAM19 downregulation ( Figure 8A ). Both of these ADAM19 shRNAs caused a significant decrease in HCT116 cell growth and clonogenic activity ( Figures 8B, C ), which is similar to what we observed upon BHLHE40 downregulation (see Figure 4 ). In contrast, downregulation of KLF7 did not affect HCT116 cell growth, but still led to a reduction in clonogenic activity ( Figures 8D-F ), albeit less so than compared to ADAM19 downregulation. These data suggest that KLF7 and especially ADAM19 are downstream effectors of BHLHE40.