Expression of SULT1C2 is correlated with overall LUAD patient survival and cigarette smoke exposure, and cigarette smoke condensate can induce SULT1C2 in lung cell lines
To determine what role SULT1C2 may play in lung adenocarcinoma (LUAD) we first utilized large-scale publicly available datasets of gene expression levels in LUAD tumors to determine if SULT1C2 expression was related overall patient survival (OS). Because of the known involvement of SULT1C2 as a xenobiotic metabolism enzyme, we first split patients based on smoke-exposure status, then plotted survival as a function of SULT1C2 expression using KMplot(45) (Fig. 1A). The effect of SULT1C2 on LUAD patient OS appeared to be dependent on cigarette-smoke exposure; never-smoker patients showed no effect of SULT1C2 on OS, whereas patients with smoke exposure showed significantly better survival with SULT1C2 expression. To further characterize the relationship between SULT1C2 and LUAD, we analyzed publicly available data from The Cancer Genome Atlas (TCGA) on LUAD expression in tumor and unmatched adjacent tumor normal (AdjNTL). This indicated that SULT1C2 expression levels were significantly elevated in LUAD tumors relative to AdjNTL (Fig. 1B). To determine the effect of the subjects’ cigarette smoke exposure on SULT1C2 mRNA levels, LUAD tumor samples were split based on patients’ smoke exposure status into 4 major TCGA-annotated categories: never smokers (category 1), former smokers who quit more than 15 years prior (category 2), former smokers who quit less than 15 years prior (category 3), and current smokers (category 4). Plotting expresion of SULT1C2 relative to smoking status indicated that SULT1C2 expression levels were significantly inversely correlated to duration of smoke exposure (p = 9.09 x 10− 6, Fig. 1B, Table 2). To mechanistically investigate the mechanism by which SULT1C2 expression levels are affected by smoke exposure, we carried out in vitro experiments.
We selected three cell lines with varying levels of SULT1C2 endogenous expression: immortalized non-cancerous lung epithelial cell line BEAS-2B and two lung adenocarcinoma cell lines, H2347 and PC3_LUAD Publicly available RNAseq profiles of all three were downloaded from the Gene Expression Omnibus(31, 46) and aligned to the hg19 genome prior to determine reads-per-kilobase of gene per millions mapped (RPKM) levels. BEAS-2B did not express detectable levels of SULT1C2 (Fig. 1C). H2347 robustly expressed SULT1C2 while PC3_LUAD expressed barely detectable levels of SULT1C2 levels as measured by RNAseq (Fig. 1C). To determine if cigarette smoke was able to affect transriptional levels of SULT1C2, each of these cell lines was treated with cigarette smoke condensate (CSC). SULT1C2 RNA levels were measured alongside the positive control gene for CSC exposure, CYP1B1. Untreated cells were maintained in a separate incubator so they would not be affected by secondary areosolized CSC. All cell lines were treated with 10, 20, 40, and 80 ug/ml of CSC for 24 hours. We used 10 and 20 ug/ml to simulate second hand smoke exposure and 40 and 80 ug/ml to simulate a smoker and heavy smoker enviornment, respectively, and to account for potential substrate inhibition of the phase I and phase II xenobiotic metabolizing enzymes (47). We observed a dose-dependent effect on transcription levels of SULT1C2 in BEAS-2B and H2347 (Fig. 1D). In contrast, PC3_LUAD cells showed no significant dose-dependent response. All lung cell lines tested showed a significant transcriptional response of CYP1B1 at 24 hours (Figs. 1E). This raised the question of why we observed differential induction of SULT1C2 in the tested LUAD cell lines and what could account for differental levels of SULT1C2 expression prior to CSC exposure.
Methylation of SULT1C2 promoter is altered in human lung
We hypothesized that the differential endogenous expression of SULT1C2 and response to CSC exposure may be due to differing epigenetic states in the SULT1C2 promoter-spanning regulatory region. SULT1C2 is classified as having a CpG-poor promoter, with sparse CpG dinucleotide occurrence. However, multiple studies support a role of CpG-poor promoters in tissue-specific expression(48), (49). To determine the role CpG methylation in the SULT1C2 promoter plays in LUAD, we first examined DNA methylation profiles of 390 LUAD patients and 26 AdjNTL controls generated by the TCGA on the Illumina Infinium HumanMethylation450 Beadchip(50). One probe on the array, cg13968390, was located within the SULT1C2 promoter. Methylation at this probe location was used to split LUAD patients into methylation high and methylation low groups, and OS was compared between them. We observed that methylation levels at cg13968390 are significantly inversely correlated with patient overall survival (Fig. 2A). Next, we evaluated if cg13968390 methylation was altered in LUAD tumors compared to AdjNTL. To do so, we again utilized the TCGA LUAD dataset and observed that DNA methylation levels at cg13968390 were significantly lower in LUAD tumors as compared to AdjNTL (two-tailed unpaired T test; p = 2.6x10− 6, Fig. 2B). To validate these findings, we used a secondary, independent dataset generated by the Early Detection Research Network (EDRN), that profiled 59 LUAD tumors alongside matched AdjNTL(42). We found that cg13968390 was also hypomethylated in LUAD tumors vs. AdjNTL in this dataset (two-tailed paired T test, p = 5.2x10− 4, Fig. 2B). To further validate this finding, we utilized a third, independent study that profiled 27 LUAD tumors and AdjNTL derived from the Ontario Tumor Bank tissue repository. We performed a one-sided paired T-test and found that cg13968390 was significantly hypomethylated in LUAD tumor tissue from this source as well (p = 1.9x10− 3, Fig. 2B). Thus, hypomethylation of cg13968390 in tumor compared to non-tumor lung appears to be a common feature of LUAD.
Next, we set out to determine if CpG methylation of the SULT1C2 promoter was causally related to SULT1C2 expression. We examined the correlation between expression and methylation of the SULT1C2 promoter region in samples from the TCGA database for which matched RNA expression and DNA methylation data was available and found a significant inverse correlation between methylation of cg13968390 and SULT1C2 expression (Fig. 2C). Linear regression was used to determine the significance of that association (p = 1.53x10− 19). To confirm this observation, a secondary dataset derived from patients in the EDRN collection (51) consisting of 60 tumors and paired adjacent non-tumor lung (AdjNTL) was also examined for methylation state vs. matched SULT1C2 expression. We again found a statistically significant inverse correlation (cor = -0.346) between cg13968390 DNA methylation and expression of SULT1C2 (p = 8.37x10− 6, Fig. 2D). In sum, multiple LUAD patient cohorts exhibited hypomethylation of cg13968390 and a significant inverse relationship to SULT1C2 expression.
DNA methylation represses transcription of SULT1C2 promoter
Having observed a highly significant correlation between SULT1C2 promoter hypomethylation in LUAD and SULT1C2 expression levels, we sought to functionally test this relationship. To do so, we used publicly available RNAseq data for BEAS-2B, H2347 and PC3_LUAD cell lines (31, 46) as well as shotgun whole genome bisulfite sequencing (sWGBS) and reduced representation bisuifite sequencing (RRBS) data made available on the Database of Transcriptional Start Sites (DBTSS)(38) and the Gene Expression Omnibus (GEO)(36), respectively. Visualization of the SULT1C2 promoter using bisultife mode in IGV revealed that H2347 cells have unmethylated CpGs throughout the SULT1C2 promoter and robust expression of SULT1C2 in untreated cells (Fig. 3A), and that PC3_LUAD cells displayed high levels of promoter methylation and low levels of expression, consistent with promoter CpG methylation blocking transcription of the adjacent gene. This may also account for the previously observed inability of CSC to upregulate SULT1C2 in PC3_LUAD cells (Fig. 1E). In contrast, RRBS data generated on BEAS-2B cells indicated that the CpGs present in the RRBS data were unmethylated, however BEAS-2B cells lack expression of SULT1C2. This could be due to a number of factors, such as the requirement for specific transcription factors not expressed in BEAS-2B under untreated conditions, or repression of enhancers whose association are required for basal activation of the SULT1C2 promoter. In order to test whether SULT1C2 promoter methylation functionally affects transcriptional activity of the adjacent SULT1C2 gene, the promoter region from − 1.27 kb to + 535 surrounding the transcriptional start site (TSS) of SULT1C2 was cloned into a CpG-less vector(52) containing the luciferase reporter gene. The CpG-less vector is devoid of CpG dinucleotides. Therefore, the only CpGs present in the construct are within the SULT1C2 promoter. The in vitro SssI-methylated SULT1C2 promoter plasmid was transfected into all three cell lines with subsequent CSC treatment to determine the effect of the methylation state of the SULT1C2 promoter on downstream gene expression levels.
In all three lung cell lines tested, the SULT1C2 promoter showed baseline activity in the unmethylated state, and this transcriptional activity was repressed by in vitro SssI methylation (Fig. 3B). Our results are therefore in agreement with Han et al., showing that DNA methylation can directly silence CpG-poor promoters(53). We then tested whether the addition of CSC could activate the SULT1C2 promoter. Consistent with the transcript level data in Fig. 1, the addition of CSC induced transcriptional activity of the unmethylated SULT1C2 promoter in BEAS-2B cells. However, we did not observe significant induction by CSC the two LUAD cell lines, suggesting that the while cloned promoter can drive baseline expression, it may lack certain regulatory elements (such as enhancers) mediating CSC induction (Fig. 3B).
While the results of the in vitro SssI methylated SULT1C2 promoter construct indicated that methylation can affect downstream transcriptional activity, this did not test whether alteration of DNA methylation levels in vivo at the endogenous SULT1C2 locus could alter transcriptional activity of SULT1C2. In order to directly test the effect of DNA methylation on the activity of the endogenous promoter, we used the DNA methylation inhibitor 5-Aza-deoxycytidine (5-aza-CdR) to block DNA methylation and subsequently measured SULT1C2 transcription. PC3_LUAD cells were the only cell line showing endogenous SULT1C2 promoter methylation (Fig. 3C) and were therefore the only line used for this experiment. PC3_LUAD cells were exposed to two doses of 5-aza-CdR, 0.15 µm (half the clinical dose) and 0.3 µm (clinical dose)(54), for 24 hours, after which the cells underwent recovery for three days to allow for DNA replication to incorporate the drug into the daughter cells and block DNA methylation (55). Post treatment with 5-aza-CdR, MethyLight(56) was used to determine methylation status of the SULT1C2 promoter alongside qRT-PCR to evaluate SULT1C2 expression levels. Treatment of PC3_LUAD cells with 5-aza-CdR resulted in a dose-independent decrease in methylation at the endogenous SULT1C2 promoter (Fig. 3C). The percent methylated reference (PMR) decreased and SULT1C2 gene expression increased in a dose-dependent manner and reached significance at the clinical dose of 0.3 µm. Taken together, these results suggest that methylation plays a significant role in the regulation of SULT1C2 expression in lung cell lines, which agrees with our findings in human samples that showed an inverse correlation between methylation and RNA expression (Fig. 2C).
SULT1C2 expression is elevated in adjacent non-tumor lung of Asians relative to Caucasians and is significantly correlated to SULT1C2 promoter methylation levels.
Now that we had established a direct relationship between SULT1C2 promoter methylation and SULT1C2 expression, we wanted to further understand what underlying features of the cell models could contribute to the differential promoter methylation observed between PC3_LUAD and H2347. PC3_LUAD and H2347 cell lines were both derived from female patients, and no significant difference was observed in SULT1C2 levels based on sex or age of the patient (Table 2).
Table 2
SULT1C2 expression in TCGA LUAD. Multiple linear regression was used to include all listed clinical features into one model. Bolded p values were considered significant.
SULT1C2 Expression
|
Number of patients
|
Estimate
|
P-value
|
Sample type
|
|
Normal
|
53
|
0.60538
|
0.0479
|
|
Tumor
|
429
|
Gender
|
|
Male
|
210
|
-0.13455
|
0.4926
|
|
Female
|
272
|
Age
|
|
By Year
|
|
|
482
|
0.01207
|
0.3031
|
Smoking
|
vs. Never Smoker
|
|
Never Smoker
|
71
|
--
|
--
|
|
Former Smoker ( > = 15 years )
|
129
|
-0.05579
|
0.8595
|
|
Former Smoker (< 15 years )
|
169
|
-0.70430
|
0.0187
|
|
Current Smoker
|
113
|
-1.03470
|
0.0016
|
Race
|
|
vs. Caucasian
|
|
Caucasian
|
422
|
--
|
--
|
|
Black or African American
|
52
|
0.39562
|
0.2068
|
|
Asian
|
7
|
1.63644
|
0.0414
|
|
American Indian or Alaskan Native
|
1
|
-0.12833
|
0.9512
|
However, expression of SULT1C2 did vary significantly based on the race of the patient in the TCGA cohort. We then tested the cohort to determine if methylation levels in the SULT1C2 promoter varied significantly based on race as well. We observed that methylation levels were significantly lower in Asian relative to Caucasian patients (Table 3).
Table 3
The effect of race on SULT1C2 promoter methylation levels in TCGA LUAD dataset. Univariate linear regression was used on the indicated clinical feature. Bolded p value was considered significant.
cg13968390 Methylation
|
|
|
|
Race
|
|
vs. Caucasian
|
|
Caucasian
|
360
|
--
|
--
|
|
Black or African American
|
51
|
0.005096
|
0.8020
|
|
Asian
|
6
|
-0.130799
|
0.0197
|
|
American Indian or Alaskan Native
|
0
|
N/A
|
N/A
|
However, the TCGA cohort contained relatively few Asian patients, and the DNA methylation data was derived from LUAD which is subject to a host of molecular alterations during tumor formation. To further validate these findings in a secondary dataset with a larger number of Asian patients as well as data from non-tumor adjacent normal tissue, we used the AdjNTL subset of the EDRN cohort, which contained expression and methylation data from adjacent normal lung of 22 Asian and 37 Caucasian patients. We found that SULT1C2 expression was significantly elevated in Asian relative to Caucasian patients (Fig. 4A), and that there was a concomitant lower level of DNA methylation of the SULT1C2 promoter in Asian patients relative to Caucasians (Fig. 4B). Indeed, methylation of cg13968390 within the SULT1C2 promoter was the most significantly differential methylation event genome-wide between Asians and Caucasians in the EDRN AdjNTL cohort (Fig. 4C). To determine if methylation and expression were inversely related according to the race of the patient, we plotted methylation vs. expression of the EDRN dataset, which resulted in two clusters with minimal overlap, one consisting mainly of Asian patients, the other consisting primarily of Caucasian patients (Fig. 4D).
CSC alters AHR occupancy at the SULT1C2 promoter
Now that we had established a direct relationship between SULT1C2 promoter methylation levels and expression as well as CSC-mediated transcriptional activation, we wanted to understand the mechanism by which CSC mediates activation of the SULT1C2 promoter. To do so, we performed transcription factor binding site analysis on the area surrounding CpG sites within the SULT1C2 promoter using Biobase(57) (Fig. 5A). We identified the aryl hydrocarbon receptor (AHR) as a likely binding candidate to a site carrying two CpGs within the SULT1C2 promoter. AHR is a ligand-activated transcription factor(58) whose transcriptional activity is induced by xenobiotic chemicals, among which polycyclic aromatic hydrocarbons (PAH) such as benzo(a)pyrene found in cigarette smoke(59). It is well established that aryl hydrocarbon receptor acts as a transcriptional activator for phase I detoxifying enzymes such as CYP1B1 when induced by exogenous ligands(60). However, there is very little evidence available implicating AHR-mediated activation of phase II enzymes. We therefore hypothesized that AHR was the major transcription factor bridging CSC-induced transcriptional activity and SULT1C2 upregulation in cells with unmethylated SULT1C2 promoters.
If SULT1C2 is a transcriptional target of AHR in LUAD, we would expect to see a positive correlation between AHR and SULT1C2 expression in LUAD patient data cohorts. To test this, we utilized RNA expression levels of AHR and SULT1C2 generated by TCGA PanCancer study through the TIMER2.0 portal(34). Indeed, SULT1C2 expression levels were significantly positively correlated to AHR expression in LUAD (cor = 0.173, p = 8.1e− 5, Fig. 5B). This trend was also observed in the known AHR target gene CYP1B1 (cor = 0.359, p = 2.83e− 17).
We then sought to test if AHR bound differentially to the SULT1C2 promoter in the presence of cigarette smoke. In order to do so, we first analyzed RNAseq from BEAS-2B, H2347, and PC3_LUAD for levels of AHR expression. Indeed, all three cell lines expressed AHR and its dimerization partner, ARNT (Fig. 5C). We then performed chromatin immunoprecipitation of AHR from cell lines treated with vehicle or 20 ug/ml CSC. CSC treatment resulted in significant enrichment of AHR at the SULT1C2 promoter in all three cell lines as compared to DMSO (Fig. 5D). In all cases, enrichment of AHR binding to the SULT1C2 promoter was greater than the enrichment at a previously described AHR binding site near CYP1B1(60).