Inherited Alterations of TGF Beta Signaling Components in Appalachian Cervical Cancers

Abstract

PURPOSE:

This study examined targeted genomic variants of transforming growth factor beta (TGFB) signaling in Appalachian women. Appalachian women with cervical cancer were compared to healthy Appalachian counterparts to determine whether these polymorphic alleles were over-represented within this high-risk cancer population, and whether lifestyle or environmental factors modified the aggregate genetic risk in these Appalachian women.

METHODS:

Appalachian women’s survey data and blood samples from the Community Awareness, Resources, And Education (CARE) CARE I and CARE II studies (N=163 invasive cervical cancer cases, 842 controls) were used to assess gene-environment interactions and cancer risk. Polymorphic allele frequencies and socio-behavioral demographic measurements were compared using t-tests and chi-square tests. Multivariable logistic regression was used to evaluate interaction effects between genomic variance and demographic, behavioral, and environmental characteristics.

RESULTS:

Several alleles demonstrated significant interaction with smoking (TP53 rs1042522, TGFB1 rs1800469), alcohol consumption (NQO1 rs1800566), and sexual intercourse before the age of 18 (TGFBR1 rs11466445, TGFBR1 rs7034462, TGFBR1 rs11568785). Interestingly, we noted a significant interaction between “Appalachian self-identity” variables and NQO1 rs1800566. Multivariable logistic regression of cancer status in an over-dominant TGFB1 rs1800469/TFBR1 rs11568785 model demonstrated a 3.03-fold reduction in cervical cancer odds. Similar decreased odds (2.78-fold) were observed in an over-dominant TGFB1 rs1800469/TGFBR1 rs7034462 model in subjects who had no sexual intercourse before age 18.

CONCLUSIONS:

This study reports novel associations between common low-penetrance alleles in the TGFB signaling cascade and modified risk of cervical cancer in Appalachian women. Furthermore, our unexpected findings associating Appalachian identity and NQO1 rs1800566 suggests that the complex environmental exposures that contribute to Appalachian self-identity in Appalachian cervical cancer patients represents an emerging avenue of scientific exploration.

Introduction

Cancer of the uterine cervix remains a significant cause of cancer death in women worldwide [1–3]. In the U.S. an estimated 13,170 women will be diagnosed with invasive cervical cancer in 2019 and 4,250 will die from the disease. While the incidence of cervical cancer has decreased markedly over the last three decades, these advances have stabilized over the last several years. Globally, nearly 570,000 cases and more than 311,000 deaths are expected with the vast majority (84%) expected to come from less developed regions [1–3].

Despite some of the best healthcare available in the world, health disparities in underserved populations remains an ongoing battle in the U.S. Specifically, Appalachia is a 205,000 square-mile region in the U.S. that follows the contours of the Appalachian Mountain range from New York to Mississippi, touching regions in 13 states along the way [4–6]. Historically characterized as a mostly white, non-Hispanic population living in economically depressed rural settings known for mining, forestry, and chemical manufacturing, the Appalachian region has demonstrated a remarkable recovery over the last fifty years. While the poverty rate in this largely rural region has decreased from 31% to 17.2% during this period, 91 counties are still listed as high-poverty with rates greater than 1.5 times the US average [5–6]. In the U.S. there are increased incidence and death rates of rural Appalachian women from cervical cancer compared to non-Appalachian women, demonstrating a relative risk of 1.29 (95% CI, 1.21–1.38). Appalachian women are recognized to participate in lifestyle behaviors such as increased smoking, energy imbalance, and risky sexual behaviors which place them at greater risk for developing cervical cancer [6–9]. Furthermore, fewer Pap tests, elevated rates of Human Papillomavirus (HPV) infection, and an overall diminished access to health care contribute to the increased incidence of cervical cancer in these rural regions [8,10]. However, it is possible that these factors alone do not fully explain the increased risk of cervical cancer in rural Appalachia. For example, while HPV infections have been causally linked to cervical cancer development, the majority of HPV-infected women do not progress and present malignant lesions. Consequently, other factors must alter biological mechanisms that ultimately tilt the homeostatic balance of the cervical microenvironment toward cancer.

Emerging evidence suggests that the missing risk liability is associated with gene-environment interactions. It is well established that high-penetrance somatic gene mutations are associated with the development of cancers, and this damage is a hallmark feature of multistep epithelial carcinogenesis. But what about inherited genetic variants? Polymorphisms are genetic variants that are detected in >1% of a population and often of unknown biological significance. Millions of single nucleotide polymorphisms (SNPs) have been identified by massively parallel sequencing projects, but far fewer have been strongly associated with specific disease states such as cancer, and even less are established as part of a functional mechanism. Consequently, because there is little known about the role of genetic susceptibility and gene-environment interactions on the increased rate of cervical cancer development in Appalachia, we implemented a targeted profiling of genetic variation in Appalachian women in an effort to address this unknown genetic risk component.

The transforming growth factor-beta (TGFB) signaling cascade has been well established as a hallmark regulatory mechanism for controlling cell growth, proliferation, and migration [11]. Consequently, when components of this central pathway are impaired, cells become insensitive to TGF-ß-mediated growth controls and carcinogenic progression occurs [12]. Key members in this signaling series include the secreted cytokines and TGFB ligands (TGFB1, TGFB2, TGFB3) as well as their cognate receptors (TGFBR1, TGFBR2, TGFBR3) at the point of engagement at the cell surface. The events that follow ligand binding, receptor activation, and downstream signaling place TGFB-mediated events at the forefront for both tumor suppression and tumor promotion by altering the cell microenvironment to impact cell growth, differentiation, and migration. Importantly, numerous studies have demonstrated that polymorphic alleles of TGFB pathway components significantly increase the risk of cancer development, including HPV-associated cancers [reviewed in 13,14]. For example, SNPs in the TGFB1, TGFB2, TGFB3 cytokine genes and the TGFBR1 and TGFBR2 receptor genes have been associated with altered risk in several cancers, including those of breast, colon, lung, bladder, pancreas, liver, head and neck, ovary, endometrium, and cervix (15–24)

Furthermore, the polymorphic TGFBR1 gene presents in a number of cancers as a high-frequency, low-penetrance allele (TGFBR1*6A) that confers increased cancer susceptibility. The most common TGFBR1 allele contains an amino acid repeat containing nine alanines (TGFBR1*9A) within the coding sequence at the 3’-end of exon 1. The TGFBR1*6A hypomorphic allele contains a deletion of three alanines within this repeat region which has been proposed to encode the signal sequence [25]. Studies and meta-analyses have subsequently shown that TGFBR1*6A is (i) a candidate tumor susceptibility allele, (ii) present in a large proportion of the general population, (iii) significantly elevated (20%) in cases compared to controls, and (iv) significantly increases cancer risk by 20–22% [26,27]. Importantly, at least 14% of the general population carries at least one copy of the TGFBR1*6A allele. Recently, Levovitz et al. [28] reported on the fundamental importance of TGFBR1 as an immune susceptibility gene in HPV-associated cancers, as well as the consistent presence of other TGFB-associated signaling components in HPV-mediated cancers.

Therefore, while the polymorphic alleles within the TGFB signaling cascade have been implicated in modifying cancer susceptibility, the contributions of these factors within a gene-environment model have not been characterized for cervical cancer within the elevated risk Appalachian population. Consequently, a series of pathway-based candidate gene variants previously associated with altered cancer risk were examined within the framework of Appalachian cervical cancer patients versus regionally-matched women presenting with normal Pap tests. In addition, polymorphisms demonstrated to be associated with cervical cancer risk in the general population were characterized for their contribution in this specific at-risk population. Furthermore, key clinicopathologic and social-demographic features were interrogated in relation to these genetic cofactors in order to explore novel gene-environment interactions within this unique underserved population.

Materials and Methods

Study Design:

As part of the National Institutes of Health Centers for Population Health and Health Disparities (CPHHD) Initiative, The Ohio State University CPHHD Center was developed to focus on the health of underserved areas in Appalachia. Through this initiative the Community Awareness, Resources, And Education (CARE I [29] and CARE II [30]) studies were established to address high cervical cancer incidence and mortality in the southern North Central Appalachia region (Ohio, West Virginia) and Central Appalachian region (West Virginia, Kentucky) [6].

One source of data and samples was a case-control study in CARE II, included 163 women diagnosed with invasive cervical cancer (ICC) and 79 control women who had normal Pap tests from Appalachian Ohio, West Virginia, and Kentucky (CARE II 2011–2015) [30]. A second source of data and samples from women with normal Pap tests (control group) were included from southern Appalachian Ohio, CARE I (2005–2009, N=763) [29]. Study methods of CARE I have been described in great detail previously [31].

Women with prevalent or newly diagnosed ICC were recruited during the CARE II study period. Inclusion criteria for participants were (i) women residing in Appalachian counties who were ≥18 years, (ii) spoke English, (iii) not cognitively impaired, and (iv) able to provide informed consent. Blood samples were collected at the time of recruitment into each study and used for genomic DNA isolation. A self-administered baseline questionnaire was collected from all participants at the time of recruitment. In all studies, blood samples were collected at the time of recruitment into each study and used for genomic DNA isolation. A self-administered baseline questionnaire was collected from all participants at the time of recruitment. Questions included those related to demographics, social environment, general health history, behavioral, psychosocial and sexual health.

Statistical Analysis:

Demographic, behavioral and environmental characteristics were compared across arms using t-tests and chi-square tests as appropriate. Several different genetic models of inheritance to estimate cancer risks were defined. For example, rs1234567890 is a SNP (T/C) with “T” representing the most frequent allele and “C” representing the variant allele which modifies the risk of developing cervical cancer. The risk of developing cancer is dependent on the number of C alleles present in the genotype of interest. Genetic models were defined as follows: (i) In a Dominant Model the presence of a single risk allele C is sufficient to modify risk. Consequently, both the heterozygous T/C genotype and the homozygous C/C genotype have the same risk, and in summation are compared to the homozygous most frequent allele T/T genotype; (ii) In a Co-dominant Model each genotype presents as distinct and non-additive. The heterozygous T/C genotype is compared to the homozygous most frequent allele T/T genotype, and the homozygous risk allele C/C genotype is compared to the homozygous most frequent allele T/T genotype. Co-dominant models allow for a hypo/hypermorphic gradient such that a functional continuum can exist [CC>T/C>T/T or CC<T/C<TT]; (iii) In an Over-dominant Model the heterozygous T/C genotype is compared to the summation of the homozygous most frequent allele genotype T/T and the homozygous risk allele genotype C/C. In this model, the heterozygous T/C allele is dominant to either homozygous allele; (iv) In a Recessive Model the presence of two risk alleles C/C is necessary to modify the risk for cervical cancer. Consequently, both the heterozygous T/C genotype and the homozygous most common allele T/T genotype have the same risk, and the summation of these genotypes is compared to the homozygous variant allele C/C genotype. Of particular interest was the potential effect modification of polymorphism effects by smoking status, alcohol consumption, self-reported Appalachian identity [32,33], and risky sexual behavior adjusted for age. These were assessed via multivariable logistic regression models. Secondary analyses used multivariable logistic regression models to examine the effects of multiple polymorphisms, significant confounders identified a priori, as well as their interactions. Analyses were performed in SAS version 9.4 (SAS institute, Cary, NC). All p-values and confidence intervals are not adjusted for multiple comparisons.

Genomic DNA Isolation:

Venous blood was collected directly into a PreAnalytix PAXgene Blood DNA Tube for the stabilization and storage of samples prior genomic DNA isolation. All samples were stored at −70°C until processed with the PreAnalytix PAXgene Blood DNA Kit. Purified genomic DNA samples were assessed for quality and quantity using an Agilent NanoDrop spectrophotometer.

Gene Variant Profiling:

A pathway-based targeted genomic variance analysis of 9 SNPs and one polymorphic repeat variant was conducted on blood DNA (Table 1). Gene Variant Profiling: A pathway-based targeted genomic variance analysis of 9 SNPs and one polymorphic repeat variant was conducted on blood DNA (Table 1). 100ng of genomic DNA was used for real-time PCR amplification using predesigned validated TaqMan SNP Genotyping Assays (Life Technologies). Each genotyping assay contained allele-specific VIC- or FAM-dye labeled probes containing minor groove binder (MGB) probe for improved hybridization stability and non-fluorescent quencher (NFQ) for decreased fluorescence background. Following amplification as suggested by the manufacturer, genotyping data was analyzed using TaqMan Genotyper Software (Life Technologies). The polymorphic repeat in TGFBR1 rs11466445 was genotyped and assessed for loss of heterozygosity (LOH) as previously reported by us [27].

Table 1.

Candidate Polymorphic Alleles Associated with Cervical Cancer Risk in Appalachian Women

Gene Name	Variant ID b	Type of Polymorphism	Cancer Association
Transforming Growth Factor Beta 1	rs1800469	single nucleotide variant	Cervical, Endometrial, Oropharyngeal, Lung, Pancreatic, Colorectal, Hepatocellular
Transforming Growth Factor Beta 1	rs1800470	single nucleotide variant	Cervical, Nasopharyngeal, Breast, Lung, and Neck, Prostate
Transforming Growth Factor Beta 3	rs3917200	single nucleotide variant	Cervical, Vulvar
Transforming Growth Factor Beta Receptor 1	rs868	single nucleotide variant	Colorectal, Bladder, Hepatocellular
Transforming Growth Factor Beta Receptor 1	rs7034462	single nucleotide variant	Breast, Colorectal
Transforming Growth Factor Beta Receptor 1	rs11568785	single nucleotide variant	Colorectal
Transforming Growth Factor Beta Receptor 1	rs11466445	Trinucleotide microsatellite	Cervical, Ovarian, Breast, Colorectal, Head Neck, Bladder
CD83 Molecule	rs750749	single nucleotide variant	Cervical, Vulvar
NAD(P)H Quinone Dehydrogenase 1	rs1800566	single nucleotide variant	Cervical, Breast, Colorectal, Prostate, Lung, Bladder, Hepatocellular, Lymphoma
Tumor Protein p53	rs1042522	single nucleotide variant	Cervical, Endometrial, Ovarian, Breast, Colorectal, Gastric, Head and Neck, Esophageal, Nasopharyngeal, Hepatocellular, Neuroblastoma, Mesothelioma, Leukemia

^aGene ID from HUGO Gene Nomenclature Committee (HGNC)

^bVariant ID from NCBI dbSNP (https://www.ncbi.nlm.nih.gov/snp)

^cCitations available via PubMed link within dbSNP

Results:

Characterization of the Study Participants:

Survey assessments and biologicals were obtained from 163 women with cervical cancer and from 832 women without cervical cancer. Table 2 describes all participant demographics, disease status (ICC, normal cervical cytology). Women with ICC tended to be older and married, less often reported consuming alcohol in the last month, and more often reported previous abnormal Pap test than control participants. Frequencies of polymorphic variants are presented in Table 3 and Online Resource 1.

Table 2.

Demographics and Clinical Characteristics of Appalachian Study Population.

Variables	CARE II a ICC (N=163)	CARE II b,c Normal Cervical Cytology Controls (N=79)	CARE I Normal Cervical Cytology Controls (N=763)
Age, mean (sd)*	52.8 (12.4)	36.9 (11.7)	35.3 (13.2)
White	148 (97%)	51 (94%)	716 (95%)
Perceived stress, mean (sd)*	18.2 (7.9)	22.5 (7.8)	17.3 (7.5)
CES-D, mean (sd) d*	17.7 (13.1)	23.9 (13.9)	15.3 (11.7)
Smoking Status
Current	52 (34%)	23 (45%)	294 (39%)
Former	33 (22%)	6 (12%)	105 (14%)
Ever (Current + Former)	85 (56%)	29 (57%)	399 (53%)
Never	66 (44%)	22 (43%)	356 (47%)
Consumed alcohol in last month*	46 (32%)	22 (42%)	413 (55%)
Had abnormal Pap test*	125 (88%)	27 (54%)	298 (41%)
Have current insurance coverage*	140 (93%)	43 (80%)	170 (80%)
Marital status
Married	74 (49%)	21 (40%)	337 (45%)
Divorced	36 (24%)	11 (21%)	94 (13%)
Widowed	23 (15%)	2 (4%)	15 (2%)
Separated	2 (1%)	3 (6%)	22 (3%)
Single	14 (9%)	12 (23%)	217 (29%)
Living together	2 (1%)	4 (8%)	57 (8%)
Residence State (N=987)*
OH	32 (22.1%)	0 (0%)	763 (100%)
WV	113 (77.9%)	79 (100%)	0 (0%)
Born State (N=412)*
OH	32 (21.5%)	2 (3.8%)	186 (88.2%)
WV	90 (60.4%)	41 (78.8%)	5 (2.4%)
KY	12 (8.1%)	3 (5.8%)	1 (0.5%)
PA	5 (3.4%)	2 (3.8%)	8 (3.8%)
Other	10 (6.7%)	4 (7.7%)	11 (5.2%)
Parents from Appalachian county (N=319)*
Both parents	95 (84.1%)	27 (73.0%)	118 (69.8%)
One parent	12 (10.6%)	8 (21.6%)	19 (11.2%)
None	6 (5.3%)	2 (5.4%)	32 (18.9%)
Appalachian Identity (N=953) e*
Yes	99 (76.2%)	41 (93.2%)	219 (29.3%)
No	31 (23.8%)	3 (6.8%)	528 (70.7%)
Sexual Behavior (Intercourse)
Age at first sex, mean (sd)*	17.4 (3.2)	16.0 (3.6)	16.7 (2.8)
Total partners, median (IQR)	4 (5)	6 (6)	5 (6)
Had sex before age 18*	85 (59.9%)	38 (77.6%)	501 (67.9%)
Had ≧4 partners	87 (66.9%)	29 (74.4%)	425 (62.3%)
Partners ever treated for STD	11 (7.4%)	9 (17.0%)	69 (9.2%)
Self ever treated for STD	21 (14.1%)	7 (13.0%)	75 (10%)

^aclinicaltrials.gov ID ; Community Awareness, Resources and Education (CARE I)[17]

^bclinicaltrials.gov ID ; Community Awareness, Resources and Education (CARE II)[18]

^cnegative cytology based upon Pap smear at time of blood collection

^dCenter for Epidemiologic Studies Depression Scale (CES-D) [60]

^eAppalachian Identity, self-identified inclusion based on geographic, cultural, and SES perceptions

^*p-value <0.05 for comparison among three groups.

Table 3.

Genotype Frequencies of Polymorphic Variants in the Appalachian Study Population.

Variables	CARE II a Cases (N=163)	CARE II Controls (N=79)	CARE I b Normal Pap (N=763)
TGFB1 rs1800469
A/A	15 (9%)	3 (4%)	72 (9%)
A/G	75 (46%)	32 (41%)	310 (41%)
G/G	73 (45%)	44 (56%)	381 (50%)
TGFB1 rs1800470
A/A	52 (32%)	28 (35%)	294 (39%)
A/G	84 (52%)	43 (54%)	348 (46%)
G/G	27 (17%)	8 (10%)	121 (16%)
TGFB3 rs3917200
A/A	135 (83%)	64 (81%)	644 (84%)
A/G	26 (16%)	14 (18%)	111 (15%)
G/G	2 (1%)	1 (1%)	8 (1%)
TGFBR1 rs868
A/A	107 (66%)	56 (71%)	495 (65%)
A/G	52 (32%)	23 (29%)	235 (31%)
G/G	4 (2%)	(0%)	33 (4%)
TGFBR1 rs7034462
C/C	136 (83%)	67 (85%)	635 (83%)
C/T	25 (15%)	12 (15%)	122 (16%)
T/T	2 (1%)	0 (0%)	6 (1%)
TGFBR1 rs11568785
A/A	135 (83%)	70 (89%)	645 (85%)
A/G	26 (16%)	9 (11%)	113 (15%)
G/G	2 (1%)	0 (0%)	5 (1%)
TGFBR1 rs11466445
Non 9A	34 (21%)	10 (13%)	141 (19%)
9A	129 (79%)	69 (87%)	618 (81%)
9A/9A	129 (79%)	69 (87%)	618 (81%)
9A/6A	32 (20%)	10 (13%)	136 (18%)
6A/6A	2 (1%)	0 (0%)	5 (0.7%)
CD83 rs750749
C/C	5 (3%)	3 (4%)	26 (3%)
C/T	49 (30%)	25 (32%)	269 (35%)
T/T	109 (67%)	51 (65%)	468 (61%)
NQO1 rs1800566
A/A	4 (3%)	1 (1%)	17 (2%)
A/G	56 (34%)	18 (23%)	256 (34%)
G/G	103 (63%)	60 (76%)	490 (64%)
TP53 rs1042522
C/C	84 (52%)	43 (54%)	398 (52%)
C/G	63 (39%)	32 (41%)	317 (42%)
G/G	16 (10%)	4 (5%)	48 (6%)

^aclinicaltrials.gov ID ; Community Awareness, Resources and Education (CARE II) [30]

^bclinicaltrials.gov ID ; Community Awareness, Resources and Education (CARE I) [29]

Examination of Allele Distribution for Hardy-Weinberg Equilibrium.

No statistically significant deviations from the expected Hardy-Weinberg (H-W) predictions were noted for any of the polymorphic alleles examined (Table 4), with the exception of NQ01 (p-value=0.02).

Table 4.

Allele Frequencies of Polymorphic Variants in the Appalachian Study Population and association between genotypes and cancer status, adjusted for age

	Case	Control	ORage a	HWE p-value b
TGFB1 rs1800469
A/G	75 (46%)	342 (41%)	1.0	0.60
A/A	15 (9%)	75 (9%)	0.73 (0.36 – 1.47)
G/G	73 (45%)	425 (50%)	0.75 (0.5 – 1.12)
TGFB1 rs1800470
A/G	84 (52%)	391 (46%)	1.0	0.57
A/A	52 (32%)	322 (38%)	0.79 (0.52 – 1.2)
G/G	27 (17%)	129 (15%)	0.85 (0.49 – 1.48)
TGFB3 rs3917200
A/G	26 (16%)	125 (15%)	•	0.19
G/G	2 (1%)	9 (1%)
A/A	135 (83%)	708 (84%)
TGFBR1 rs868
A/G	52 (32%)	258 (31%)	•
G/G	4 (2%)	33 (4%)
A/A	107 (66%)	551 (65%)
TGFBR1 rs7034462
C/T	25 (15%)	134 (16%)	•	0.89
C/C	136 (83%)	702 (83%)
T/T	2 (1%)	6 (1%)
TGFBR1 rs11568785
A/G	26 (16%)	122 (14%)	•	0.93
G/G	2 (1%)	5 (1%)
A/A	135 (83%)	715 (85%)
TGFBR1 rs11466445
9A/9A	129 (79%)	687 (82%)	1.14 (0.71 – 1.84)	0.33
9A/6A	32 (20%)	146 (18%)	1.0
6A/6A	2 (1%)	5 (1%)
CD83 rs750749
C/T	49 (30%)	294 (35%)	1.0	0.10
C/C	5 (3%)	29 (3%)	0.98 (0.32 – 3.07)
T/T	109 (67%)	519 (62%)	1.29 (0.85 – 1.96)
NQO1 rs1800566
A/G	56 (34%)	274 (33%)	•	0.02
G/G	103 (63%)	550 (65%)
A/A	4 (2%)	18 (2%)
TP53 rs1042522
C/G	63 (39%)	349 (41%)	1.0	0.12
C/C	84 (52%)	441 (52%)	1.20 (0.8 – 1.79)
G/G	16 (10%)	52 (6%)	1.27 (0.6 – 2.67)

^aOdds ratios are reported as OR (95% CI).

^bHWE P-values were obtained from Hardy-Weinberg equilibrium of the control group.

^•Odds ratio not provided due to small cell size.

Associations Between Genotypes, Behavioral/Environmental Factors, and Cancer Status.

Several polymorphic alleles demonstrated an interaction after adjusting for age with the recognized behavioral hazards of smoking (TP53 rs1042522, TGFB1 rs1800469), alcohol consumption (NQO1 rs1800566), and sex before the age of 18 (TGFBR1 rs11466445, TGFBR1 rs7034462, TGFBR1 rs11568785). In an over-dominant allele model adjusting on age for TGFB1 rs1800469, never-smokers with homozygous genotypes A/A+G/G were 2.5-fold (OR=0.4, 95% CI: 0.22–0.73, p-value=0.003) less likely to have cervical cancer compared to never-smokers with the heterozygous A/G genotype. This effect was not observed in ever-smokers (interaction p-value=0.02). A significant 3.1-fold increase in the odds of cervical cancer was estimated with TP53 rs1042522 G/G dominant model compared to C/C+C/G genotypes in never-smokers (aOR=3.1, 95% CI: 1.1–8.5, p-value=0.030), but not ever-smokers, with a marked interaction of smoking by genotype (p-value=0.02).

Risky sexual behavior also had an interaction effect on the association between several SNPs and cancer risk. Subjects who had no sex before 18 and with heterozygous non-9A genotype were 2.3-fold (OR=2.3, 95% CI: 1.07–4.89, p-value=0.033) more likely to have cervical cancer compared to subjects who had no sex before 18 and with 9A genotype. In a recessive model for TGFBR1 rs7034462, subjects who had no sex before 18 with C/T+T/T genotype were 2.3-fold (OR=2.31, 95% CI: 1.03–5.22, p-value=0.044) more likely to have cervical cancer compared to subjects who had no sex before 18 with C/C genotype. A significant 2.8-fold decrease in the odds of cervical cancer was observed in the over-dominant model in subjects who had no sex before 18 and with C/C+T/T genotype (OR=0.36, 95% CI: 0.16–0.84, p-value=0.017). Similar effects were also observed with TGFBR1 rs11568785. In dominant model, subjects who had no sex before 18 and with A/A genotype were 2.7-fold (OR=0.37, 95% CI: 0.16–0.83, p=0.016) less likely to have cervical cancer compared to subjects who had no sex before 18 and with G/G+A/G genotype. The odds for subjects with A/A+G/G genotype is 3.1-fold (OR: 0.32, 95% CI: 0.14–0.74, p=0.0075) less likely compared to subjects with A/G genotype in over-dominant model. All the effects were not observed in subjects who had sex before 18 (all interactions p-value<0.05). Unexpectedly, we noted a marginal interaction with the variables designed to assess self-identified “Appalachian Identity” and the NQO1 rs1800566 polymorphic allele. Adjusted genetic models for each cancer risk factor are presented in Table 5, Online Resource 1).

Table 5.

Association Between Genotypes and Cancer Status, Adjusted for Participant Age

Interaction Effect tested: Smoking Status
	Never smokers (N=443)			Ever smokers (N=511)
	Control	Case	OR (95% CI)	Control	Case	OR (95% CI)	p-value
TGFB1 rs1800469 Over-dominant Model
A/G	156 (41%)	38 (58%)	1.0 (Referent)	169 (40%)	32 (38%)	1.0 (Referent)
A/A-G/G	222 (59%)	27 (42%)	0.4 (0.22 – 0.73)	258 (60%)	52 (62%)	1.02 (0.59 – 1.75)	0.02
TP53 rs1042522 Dominant Model
C/C-C/G	361 (95%)	57 (88%)	1.0 (Referent)	395 (93%)	78 (93%)	1.0 (Referent)
G/G	17 (5%)	8 (12%)	3.08 (1.11 – 8.50)	32 (7%)	6 (7%)	0.55 (0.19 – 1.56)	0.02
Interaction Effect tested: Alcohol Consumption
	No alcohol consumption last month (N=468)			Consumed alcohol last month (N=480)
	Control	Case	OR (95% CI)	Control	Case	OR (95% CI)	p-value
NQO1 rs1800566 Recessive Model
G/G	246 (66%)	59 (60%)	1.0 (Referent)	276 (64%)	33 (72%)	1.0 (Referent)
A/G-A/A	124 (34%)	39 (40%)	1.4 (0.83 – 2.36)	158 (36%)	13 (28%)	0.58 (0.28 – 1.19)	0.05
NQO1 rs1800566 Over-dominant Model
A/G	116 (31%)	37 (38%)	1.0 (Referent)	148 (34%)	12 (26%)	1.0 (Referent)
A/A-G/G	254 (69%)	61 (62%)	0.70 (0.41 – 1.18)	286 (66%)	34 (74%)	1.72 (0.82 – 3.60)	0.05
Interaction Effect tested: Sex before 18y
	No sex before 18y (N=305)			Sex before 18y (N=621)
	Control	Case	OR (95% CI)	Control	Case	OR (95% CI)	p-value
TGFBR1 rs11466445 Polymorphism
Non 9A	48 (20%)	17 (30%)	2.29 (1.07 – 4.89)	93 (17%)	15 (18%)	0.74 (0.37 – 1.49)	0.034
9A	198 (80%)	40 (70%)	1.0 (Referent)	443 (83%)	68 (82%)	1.0 (Referent)
TGFBR1 rs7034462 Recessive Model
C/C	207 (83%)	44 (77%)	1.0 (Referent)	449 (83%)	70 (84%)	1.0 (Referent)
C/T-T/T	41 (17%)	13 (23%)	2.31 (1.03 – 5.22)	89 (17%)	13 (16%)	0.66 (0.32 – 1.36)	0.025
TGFBR1 rs7034462 Over-dominant Model
C/T	36 (15%)	12 (21%)	1.0 (Referent)	88 (16%)	13 (16%)	1.0 (Referent)
C/C-T/T	212 (85%)	45 (79%)	0.36 (0.16 – 0.84)	450 (84%)	70 (84%)	1.49 (0.72 – 3.09)	0.013
TGFBR1 rs11568785 Dominant Model
G/G-A/G	37 (15%)	14 (25%)	1.0 (Referent)	81 (15%)	13 (16%)	1.0 (Referent)
A/A	211 (85%)	43 (75%)	0.37 (0.16 – 0.83)	457 (85%)	70 (84%)	1.52 (0.73 – 3.2)	0.012
TGFBR1 rs11568785 Over-dominant Model
A/G	34 (14%)	13 (23%)	1.0 (Referent)	79 (15%)	13 (16%)	1.0 (Referent)
A/A-G/G	214 (86%)	44 (77%)	0.32 (0.14 – 0.74)	459 (85%)	70 (84%)	1.46 (0.69 – 3.09)	0.0083
Interaction Effect tested: Self-reported Appalachian Identity
	Non-Appalachian Identity (N=561)			Appalachian Identity (N=357)
	Control	Case	OR (95% CI)	Control	Case	OR (95% CI)	p-value
NQO1 rs1800566 Recessive model
G/G	330 (62%)	23 (74%)	1.0 (Referent)	178 (68%)	57 (59%)	1.0 (Referent)
A/G-A/A	200 (38%)	8 (26%)	0.52 (0.22 – 1.24)	82 (32%)	40 (41%)	1.55 (0.90 – 2.68)	0.04

Note: Interaction effects with ≧4 partners and parent county were also tested. Only significant interaction effects were reported here. Interaction test were not performed on specific models if any cell size <5.

Multivariable Genetic Models for Cancer Risk Estimates.

Results from multivariable genetic model with TGFB1 rs1800469 and TGFBR1 rs11568785 were presented in Table 6 (Online Resource 2, Online Resource 3). After adjusting for factors identified from previous genetic models, including age, smoking status, and risky sexual behavior, subjects who had no sexual intercourse before age 18 and with homozygous A/A+G/G genotype for TGFBR1 rs11568785 were 3.03-fold (OR: 0.33, 95% CI: 0.14–0.78) reduced in odds of having cervical cancer compared to A/G genotype. This effect was not observed in subjects who had sex before age 18 (interaction effect p-value=0.0078). The significant main effect of TGFB1 rs1800469 indicated that subjects with A/A+G/G genotype were 1.56-fold (OR: 0.64, 95% CI: 0.42–0.98) reduction in their odds of having cervical cancer compared to A/G genotype (p-value=0.037). Similar effects were observed in multivariable genetic model with TGFB1 rs1800469 and TGFBR1 rs7034462 (Table 6, Online Resource 2). After adjusting for age, smoking status, and risky sexual behavior, a significant 2.78-fold (OR: 0.36, 95% CI: 0.15–0.84) decrease in the odds of cervical cancer were observed in subjects who had no sex before age 18 and with homozygous C/C+T/T genotype compared to heterozygous C/T genotype. Such effect was not observed in subjects with risky sexual behavior (interaction effect p-value=0.0099).

Table 6.

Multivariable Logistic Regression Results Examining the Association Between TGFB1–TGFBR1 Genotypes and Cancer Status.

	Levels	OR (95% CI)	p-value
Age	10y Increase	2.75 (2.3 – 3.28)	<0.0001
Smoking Status	Never Smoker Ever Smoker	0.79 (0.51 – 1.23) 1.0	0.3
TGFB1 rs1800469	A/A+G/G A/G	0.64 (0.42 – 0.98) 1.0	0.037
TGFBR1 rs11568785 by Risky Sexual Behavior interaction			0.0078
TGFBR1 rs11568785 when subject had no sex before 18y	A/A+G/G A/G	0.33 (0.14 – 0.78) 1
TGFBR1 rs11568785 when subject had sex before 18y	A/A+G/G A/G	1.58 (0.74 – 3.35) 1.0
Model with patient’s age, smoking status, risky sexual behavior before 18 years of age, TGFB1 rs1800469 genotype, TGFBR1 rs11568785 genotype and its interaction with risky sexual behavior before 18 years of age. (N=918)

	Levels	OR (95% CI)	P-value
Age	10y Increase	2.76 (2.31 – 3.30)	<0.0001
Smoking Status	Never Smoker Ever Smoker	0.77 (0.50 – 1.20) 1.0	0.25
TGFB1 rs1800469	A/A+G/G A/G	0.63 (0.41 – 0.95) 1.0	0.029
TGFBR1 rs7034462 by Risky Sexual Behavior interaction			0.0099
TGFBR1 rs7034462 when subject had no sex before 18y	C/C+T/T C/T	0.36 (0.15 – 0.84) 1.0
TGFBR1 rs7034462 when subject had no sex before 18y	C/C+T/T C/T	1.60 (0.76 – 3.36) 1.0
Model with patient’s age, smoking status, risky sexual behavior before age 18, TGFB1 genotype rs1800469, TGFBR1 genotype rs7034462 and its interaction with risky sexual behavior before age 18. (N=918)

Discussion

People living in the Appalachian Region continue to be at higher risk for the development of cancers, including cervical cancer, compared to the rest of the U.S. [34]. Some risk factors, such as high incidence of tobacco smoking, recovering but still significant levels of poverty, and improving but still poor access to healthcare certainly contribute to these higher incidences of cancer (Table 2). While targeted efforts have narrowed the gap in health disparities between Appalachian and non-Appalachian regions, overall cancer incidence remains elevated in the Appalachian counties. In fact, as suggested by Wilson et al. [34], poorer rates of cervical cancer screening may further underestimate the cancer burden in Appalachian counties. However, another factor that can modulate cancer risk is susceptibility mediated by genomic variation in the form of polymorphic alleles [35] (Table 3). Genetic variants can alter biological responses depending upon the gene-environment interactions present [36]. These studies report for the first time the role of genetic variation as a risk cofactor for the development of invasive cervical cancer in Appalachian women.

The TGFB1 rs1800469 polymorphism is associated with a decreased risk of developing cervical cancer in Appalachian women.

We targeted genomic variants within the TGFB signaling cascade in response to this pathway’s established history with EMT and epithelial carcinogenesis. Polymorphism in TGFB cytokines as well as their cognate receptors would allow mechanisms to potentially mediate susceptibility of cervical epithelia initiation, promotion, and transition into malignant invasive disease (Table 4). The TGFB1 rs1800469 polymorphism, commonly reported as C-509T is a single nucleotide variation in the TGFB1 gene that functionally alters promoter region activity. The variant T allele is common with a MAF=0.3680 and TGFB1 rs1800469 demonstrates an average heterozygosity of 0.465±0.127 (dbSNP) [37]. The C-509T variation has been reported to be associated with disease states ranging from asthma [38] to aplastic anemia [39]. Several studies support the finding that the homozygous C/C genotype is associated with low expression of TGFB1 and the homozygous variant allele T/T genotype is marked by TGFB1 cytokine over-expression, possibly by enhanced transcription factor binding [38,40] in the promoter. The heterozygote C/T genotype is often of an intermediate phenotype for TGFB1 expression levels. Interestingly, Jin et al. [41] found that the C/T heterozygote was in fact independently associated with a significant decreased gastric cancer risk, and in a co-dominant genetic model the T/T+C/T genotypes were significantly associated with a decreased risk of gastric cancer compared to C/C homozygotes.

Within the cervical cancer microenvironment, mesenchymal stromal cells (MSCs) can differentiate and drive tumorigenesis by promoting proliferation, angiogenesis, and migration as part of EMT. The increased expression of TGFB1 and other immune suppressors initiates a microenvironment that is resistant to the cell mediated tumor response of cytotoxic T cells (CTLs) and natural killer cells. These high levels of TGFB1 coupled with elevated IL-10 leads to suppression of HLA-1 and the necessary trajectory for a cell to escape from immune surveillance and destruction [42]. We report that in an over-dominant model the C/C+T/T genotypes demonstrate a 2.5-fold decreased risk for developing cervical cancer compared to the heterozygous C/T genotype (Online Resource 3), and that this effect is specific for never-smokers. How genotypes that have previously associated with either basal levels (C/C) or over-expression (T/T) of TGFB1 confer a decreased risk of cervical cancer in this Appalachian population remains unanswered and may reflect the unique gene-environment interactions of this population.

The TP53 rs1042522 polymorphism is associated with an increased risk of developing cervical cancer in Appalachian women.

The p53 protein encoded by TP53 has been characterized as the “guardian of the genome” due to its intrinsic role in regulating the cell cycle and maintaining genomic integrity [43]. Due to this essential role in suppressing cell growth when genetic damage is detected, the discovery that p53 was mutated in about 50% of all cancers, including cervical cancer, was striking. Common activators of p53 include DNA-damaging chemical exposures, low oxygen conditions, and other metabolic stressors [44]. The amino acid change from proline to arginine that results from the single nucleotide change in codon 72 of TP53 rs1042522 is one of the most studied functional polymorphic events in the human genome. The ancestral C allele (encoding proline) demonstrates a diminished proapoptotic activity level compared to the variant G allele (MAF=0.4571, dbSNP) [37]. The high frequency variant G allele (encoding arginine) has been associated with advanced cancer staging and metastatic potential in addition to its enhanced proapoptotic activities [45].

While the impact of TP53 rs1042522 on cervical cancer promotion remains an active area of debate, consistent with our results, a meta-analysis of 70 case-control studies (14999 cases, 8195 controls) by Jee et al. [46] reported that G/G homozygotes demonstrated an increased risk relative to C/G heterozygotes (OR: 1.2, 95% CI 1.1–1.3, p-value=0.001). However, a subsequent meta-analysis with pooled data from 49 case-control studies (7946 cases, 7888 controls) proposed that following subgroup analyses, the impact of the G/G genotype was more likely due to small case numbers and deviations from HWE rather than biology [47]. We report that in a dominant model, G/G homozygotes have a 3.1-fold increase in the odds of having cervical cancer compared to either C/C or C/T genotypes. Similarly, in a co-dominant model, individuals with the G/G genotype are 3.65-fold more likely to have cervical cancer than those that are C allele carriers (Online Resource 3). Consistent with non-Appalachian general population studies, the genetic liability affiliated with the TP53 rs1042522 G risk allele is strongly maintained in Appalachian populations.

HPV and Smoking.

The human papillomavirus (HPV) is a dsDNA virus whose high-risk variants are causally associated with the vast majority of cervical cancers. While most HPV infections are transient and self-resolve, some become chronic and the presence of HPV in ICC is >99% [reviewed in 48]. Importantly, the E6 and E7 proteins of HPV are able to functionally interact with the p53 and pRB tumor suppressors, respectively, and alter the regulation of cell cycle progression checkpoints. Specifically, E6 can bind to p53 and promote ubiquitin-mediated proteasomal degradation [reviewed 49]. Consequently, in the absence of p53 genetically compromised cells are allowed to progress through the cell cycle. The unrepaired DNA damage can become fixed, and subsequent mutations contribute to the initiation of cervical carcinogenesis [49].

Tobacco smoking has been causally associated with the etiology of several cancers, including those of the lung, head and neck, esophagus, stomach, liver, pancreas, bladder and cervix [reviewed in 50–52; 53–56]. Exposure to tobacco smoke introduces carbon monoxide, hydrogen cyanide, nicotine, polyaromatic hydrocarbons such as benzo[a]pyrene (B[a]P), tobacco-specific carcinogenic nitrosamines like 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) and N’-nitrosonornicotine (NNN), as well as other toxic or carcinogenic compounds into the body. Some of these compounds (NNN, NNK, B[a]P) form DNA adducts that ultimately damage DNA repair genes, inhibit tumor suppressor genes and activate oncogenes, leading to loss of homeostatic growth control [57,58]. Others tobacco compounds (nicotine) facilitate deregulated growth by directly binding to cell receptors and promoting inappropriate cell proliferation and migration. Directly or indirectly, often via the generation of ROS and RNS, many of these compounds induce proinflammatory states that provide a favorable tumor microenvironment to facilitate cervical cancer development.

However, debate has continued as to the possible confounding role of HPV-positivity in studies where the virus is considered causally associated. Importantly, a meta-analysis of case-control studies by Plummer et al. [59] supported an independent role for smoking by demonstrating a relative risk of 2.17 (95% CI: 1.46–3.22) for HPV-positive cases relative to HPV-positive controls. We report a striking difference in cervical cancer risk and genetic variants in Appalachian women depending upon the smoking status of the cases and controls. For both TGFB1 rs1800469 and TP53 rs1042522 the contribution of the genomic variant was seen only in never-smokers. According to the CDC, despite a decrease over the last 10 years, 16.8% of the adult U.S. population currently smokes cigarettes. Importantly, 14.8% of adult women and 26.3% of adults below the poverty level in the general population smoke. In our study, 34.4% of women with cervical cancer were current smokers, while 45% of non-cancer controls reported themselves as current smokers (Table 1). The powerful health modifying events that occur with tobacco use may explain the inability to detect a cervical cancer risk modifying component for the TGFB1 and TP53 genomic variants in ever-smokers (Online Resource 2).

Interactions between Genomic Variants, Social-Behavioral Measurements, and Clinicopathologic Characteristics.

We further investigated what additional social-behavioral characteristics may contribute to overall cervical cancer risk as part of Appalachian gene-environment interactions. Using a model including age, smoking status, risky sexual behavior before 18 years of age, TGFB1 rs1800469 genotype, and TGFBR1 rs11568785 genotype, we show that there is an association between the TGFB1 ligand promoter variant rs1800469 and the TGFBR1 intronic variant rs11568785 only when there was no risky sexual behavior (Table 5, Online Resource 3). Additionally, using a similar model interrogating TGFB1 rs1800469 and the intronic TGFBR1 rs7034462 we demonstrate that there is an association only in the absence of risky sexual behavior (Table 6, Online Resource 3).

The combination of common low-penetrance genetic variants, multifactorial traits, and population heterogeneity makes definitive genome association studies challenging in defining an accurate cervical cancer genetic risk [60,61]. In many parts of Appalachian, these challenges are potentially confounded by unique socioeconomic factors, risky social behaviors, healthcare insufficiency, and incompletely surveyed environmental exposome diversity [62–65]. Prior studies have already suggested a strong relationship between socioeconomic status (SES) and depression in Appalachian female smokers [66], and provides a foundational list for potential modifiers and mediators within such an Appalachian exposome [63,64,66]. Emerging transdisciplinary geospatial approaches will further define the complex interactions between genomics, environmental exposures, and social behavior risks while engaging community-based participatory research partnerships [62,65,67–69]. The complex representation of “Appalachian identity” is, as elegantly described by BE Smith [70] as “the impossible necessity of Appalachian studies.” Our findings that an interaction effect is present between the NAD(P)H quinone oxidoreductase 1 (NQO1) variant rs1800566, self-reported Appalachian identity, and cervical cancer risk (Table 5) provides intriguing insight into the complex dynamics of defining cancer risk in this population.

Abnormal NQO1 activity has been associated with cancer risk for a number of malignancies (bladder, breast, colon, cervical, lung, and pancreas) [71–75] and is likely due to the enzyme’s ability to mediate reactive oxygen species availability and stabilize tumor suppressor proteins, including TP53 [76,77], by inhibiting proteasomal degradation. This ubiquitous but readily inducible enzyme may represent a useful genetic prognostic biomarker of cervical cancer risk in Appalachian women as the meaningful environmental exposures [78,79] that couple its association with indices of Appalachian self-identity are defined with increased resolution. Again, it is intriguing, but with caution that a possible association between a social identifier such as Appalachian identity and a functional SNP involved with detoxification must be discussed. Previously, the A/A NQO1 rs1800566 susceptibility allele has been associated with increased cervical cancer risk [80], consistent with an increased OR 1.55 (09.0–2.68) presented here (Table 5) for participant identifying as Appalachian. It would require higher resolution mapping of geographic localization, workplace risks, access to healthcare services, environmental exposome, and family health history, to reasonably speculate on the mild interaction effect seen (p-value=0.04), while noting the comparatively small numbers available for analysis. One hypothesis would be that self-identified Appalachians have an increased likelihood of exposure to the well-documented cancer risk factors endemic to rural Appalachian regions. The association of Appalachian identity and TP53 genomic variance as a social determinate of health is likely to be as an aggregate biomarker of cancer susceptibility rather than as genetic driver of disease risk per se. It is possible that as a functional SNP the cumulative environmental triggers that permit the TP53 phenotype to be potentially impactful during cervical carcinogenesis may be unique within the Appalachian population.

It should be pointed out that there are notable differences in Appalachian self-identification between CARE I controls and CARE II controls (29.3% vs 93.2%)(Table 2), and that the pooled data demonstrating an interaction between self-reported Appalachian identity and NQO1 rs1800566 makes use of a comparatively small number of cases for the genetic models (Table 5). These findings further the need to interrogate the behavioral and genetic risk factors potentially contributing to the enhanced cervical cancer risk presented across the Appalachian region, notably in the Central and Southern Appalachian Regional Commission (ARC) subregions addressed in this study. It is important to note that CARE I and CARE II were complementary but non-overlapping studies with distinct primary outcome measures. Consequently, there are different catchment clinics used for participant enrollment for CARE I and CARE II, although all clinics were from the same Appalachian geographic regions as previously noted. Additionally, there are clear differences in age between CARE II controls (36.9±11.7) versus ICC cases (52.8±12.4), with the controls between CARE I and CARE II showing only nominal differences. The risk for cervical cancer increases with age with concomitant environmental exposures such as HPV infection and cancer risk-enhancing behaviors (smoking, early sexual activity and early full-term pregnancy)[48]. Chronic persistent HPV infection increases with age, possibly confounding the associations suggested for the interactions between common low-penetrance susceptibility alleles and cervical cancer: the exposure and opportunity for non-cleared persistent HPV infection is greater in older women. The ability to better match participants would further strengthen the confidence in our conclusions.

Persistent HPV infection, exposure to tobacco smoke and alcohol, and risk-enhancing sexual behaviors clearly drive the cumulative risk for developing cervical cancer. The genetic susceptibility role that common, low-penetrance alleles play in overall risk may ultimately be minor compared to these other dominant risk factors for cervical cancer. However, that role is not trivial, especially in the case of known functional SNP phenotypes that provide both a basis for prognostic genetic biomarker screening, as well as offer the potential for predictive biomarker assessment and possible intervention measures.