Human papillomavirus oncogenes reprogram the cervical cancer microenvironment independently of and synergistically with estrogen

Significance

A subset of human papillomaviruses (HPVs) causes 5% of human cancers, including virtually all cancers of the cervix. In a mouse model of cervical cancer, estrogen is a necessary cofactor that contributes to disease by signaling through the underlying tumor microenvironment. In this study, we discovered that epithelial expression of the HPV oncoproteins reprograms the cervical tumor microenvironment and its response to estrogen. These changes involve the elicitation of paracrine-acting factors implicated in carcinogenesis, and the expression of a subset of these factors was also induced in cocultures of human cervical cancer cells and stromal fibroblasts. We hypothesize that HPV oncogenes cause cancer in part by creating a unique tumor microenvironment that synergizes with estrogen in the cervix.

Abstract

High-risk human papillomaviruses (HPVs) infect epithelial cells and are causally associated with cervical cancer, but HPV infection is not sufficient for carcinogenesis. Previously, we reported that estrogen signaling in the stromal tumor microenvironment is associated with cervical cancer maintenance and progression. We have now determined how HPV oncogenes and estrogen treatment affect genome-wide host gene expression in laser-captured regions of the cervical epithelium and stroma of untreated or estrogen-treated nontransgenic and HPV-transgenic mice. HPV oncogene expression in the cervical epithelium elicited significant gene-expression changes in the proximal stromal compartment, and estrogen treatment uniquely affected gene expression in the cervical microenvironment of HPV-transgenic mice compared with nontransgenic mice. Several potential estrogen-induced paracrine-acting factors were identified in the expression profile of the cervical tumor microenvironment. The microenvironment of estrogen-treated HPV-transgenic mice was significantly enriched for chemokine/cytokine activity and inflammatory and immune functions associated with carcinogenesis. This inflammatory signature included several proangiogenic CXCR2 receptor ligands. A subset of the same CXCR2 ligands was likewise increased in cocultures of early-passage cells from human cervical samples, with levels highest in cocultures of cervical fibroblasts and cancer-derived epithelial cells. Our studies demonstrate that high-risk HPV oncogenes profoundly reprogram the tumor microenvironment independently of and synergistically with estrogen. These observations illuminate important means by which HPVs can cause cancer through alterations in the tumor microenvironment.

Viruses cause ∼15% of human cancers (1), and high-risk human papillomaviruses (HPVs) alone are responsible for nearly 5% of human malignancies (2). HPVs infect poorly differentiated, basal keratinocytes within stratified squamous epithelia where they can establish persistent infections. The high-risk mucosotropic HPVs cause malignancies in the stratified epithelia lining the cervix, vagina, and other organs of the lower female reproductive tract, as well as the anus, penis, and epithelia lining the oropharynx. We previously developed transgenic mice expressing the high-risk type HPV16 oncogenes E6 and E7 (3, 4). In these mice, the keratin 14 (K14) promoter directs expression of the HPV viral oncogenes in the basal layer of stratified squamous epithelia, which is the natural site of papillomavirus infection. We have used this in vivo model to show how HPV E6 and E7 oncogenes and cellular cofactors contribute to HPV-associated carcinogenesis in various anatomical sites (5–8), including the cervix (9, 10).

Despite the strong etiological link between HPV and cervical cancer, persistent HPV infection is not sufficient for human cervical cancer development (11, 12), and thus other cofactors likely contribute to carcinogenesis. One such cofactor is estrogen (17β-estradiol) (13, 14). Elevated levels of circulating estrogens have been detected in women with HPV⁺ lesions and cancers (15, 16), and multiparity and long-term oral contraceptive use are significant cervical cancer risk factors (17–20). Experiments in HPV transgenic mice have directly linked estrogen and cervical cancer. In K14E6/E7 mice, expression of HPV16 E6 and E7 in the cervical epithelium is essential but not sufficient to cause cervical cancer, and additional systemic delivery of exogenous 17β-estradiol and expression of its receptor ERα are required for the onset, maintenance, and progression of neoplastic disease (9, 10, 21, 22). Furthermore, treatment with estrogen-signaling inhibitors promotes regression of cervical cancer and precancerous lesions in these mice (23, 24). While caution is reasonable in extrapolating the efficacy of drugs in murine models to therapeutic treatment of human disease (25), human population-based data do correlate long-term antiestrogen use with lower risk of cervical neoplasia (26).

An important observation is that estrogen contributes to cervical carcinogenesis through the underlying stroma. Continued expression of ERα in the cervical stroma of HPV transgenic mice is required for the maintenance of neoplastic disease in the cervical epithelium (27). In human cervical cancers, expression of ERα is, in fact, retained in the stroma but is lost in epithelial cancer cells (28), and human cervical cancer-associated fibroblasts have been shown to mediate estrogen-dependent signaling (29). Moreover, stromal–epithelial interactions mediate the effects of estrogen on female reproductive tract morphogenesis, and stromal ERα facilitates the proliferative effects of estrogen on the adjacent epithelium (30–33). The tumor microenvironment directs the biology of many cancers, as bidirectional communication between epithelial cells and the tumor microenvironment affects tumor initiation, neoplastic progression, metastasis, and therapeutic response (34). These insights suggest that stromal estrogen signaling drives cervical carcinogenesis in concert with epithelial HPV oncogene expression. In the present report, we describe how we used our estrogen-dependent K14E6/E7 cervical cancer mouse model to investigate the independent and synergistic effects of epithelial HPV oncogene expression and estrogen signaling on genome-wide gene-expression levels in the cervical stroma. We discovered an extensive reprogramming of the cervical stroma in the context of mice expressing the HPV16 E6 and E7 oncogenes in the epithelia. Unique changes were also noted in the microenvironment of these mice when given exogenous estrogen to promote cervical carcinogenesis. Paracrine factors were identified whose expression was increased in the cervical stroma of K14E6/E7 mice and was enhanced in these mice when treated with estrogen. The expression of these same paracrine factors, which were previously implicated in carcinogenesis, was further enhanced when human cervical cancer-derived cell strains were cocultured with cervical fibroblasts, defining potential modulators of cervical carcinogenesis arising from the tumor microenvironment.

Results

Gene-Expression Profiling of Cervical Epithelium and Adjacent Stroma.

We used our mouse model of cervical carcinogenesis to identify HPV oncogene- and estrogen-driven gene changes in the stroma associated with cervical cancer development. Four groups of adult female mice representing different grades of disease were included: (i) nontransgenic mice (WT; no disease); (ii) nontransgenic mice treated with estrogen (hereafter abbreviated as “E2”) for 1 mo, a duration sufficient to induce benign hyperplasia (WT+E2; hyperplasia); (iii) K14E6/E7 bitransgenic mice [E6/E7; largely cervical intraepithelial neoplasia grade 1/2 (CIN1/2) precancerous lesions]; and (iv) K14E6/E7 bitransgenic mice treated with estrogen for 6 mo, a duration necessary to induce cervical cancers (E6/E7+E2; cancer) (Fig. S1A and Table S1). From the reproductive tracts of the above mice, 20 epithelial regions of histopathological interest (i.e., no disease, hyperplasia, precancerous lesions, or cancer) and matched regions of proximal stroma were laser-capture microdissected. Pre–laser-capture microdissection (LCM) and post-LCM images were used to assess the accuracy of tissue capture before RNA extraction (Fig. S1B). RT-PCR was used to measure levels of the K14E6/E7 transgene (Fig. S1C), expression of which should be restricted to the epithelial compartment due to the tissue specificity of K14 promoter activity (3, 4). The average number of transgene transcripts was significantly higher in E6/E7 (P = 0.0007) and E6/E7+E2 (P = 0.03) cervical epithelia than in the adjacent stroma, confirming accurate capture of epithelia and stroma. A significantly higher level of transgene-specific mRNA was measured in E6/E7 epithelia than in E6/E7+E2 epithelia (P = 0.006), likely because the squamous cell carcinomas in the E6/E7+E2 group are a more heterogeneous mixture of cells compared with the more homogenous stratified epithelium in the E6/E7 group. Total RNA was then analyzed using Affymetrix GeneChip Mouse Genome 430 2.0 microarrays comprehensively measuring nearly 40,000 murine transcripts. Significant gene-expression changes between pairwise group comparisons were defined as having a false-discovery rate (FDR) ≤0.05 and a fold change ≥2.

Fig. S1.

Experimental overview of gene-expression profiling of cervical stroma. (A, Left) Reproductive tracts were harvested from four groups of mice, and histopathology analysis was performed on H&E-stained murine endocervical tissue sections. (Right) Representative regions of interest (ROIs) for LCM. Matched sets of ROIs share the same numerical label, with epithelial ROIs indicated by numbers only (i.e., 1 and 2) and stromal ROIs labeled with a number and the letter “s” (i.e., 1s and 2s). (B) Representative pre-LCM and post-LCM images showing the accuracy of LCM of stromal tissue. Extracted stromal tissue is shown on the far right. (C) Results of RT-PCR analysis specific for the HPV transgene performed using RNA extracted from epithelial (black bars) and stromal (gray bars) tissue isolated from FVB or K14E6/E7 transgenic mice. Values were normalized to those for GAPDH and then averaged among groups. A Wilcoxon rank-sum test was used to compare epithelial versus stroma values within each group. *P < 0.03 compared with the stroma sample within each group. Error bars indicate SD.

Table S1.

Overview of samples included in analysis

Genotype	Compartment	Treatment (duration)	No. of samples	Grade of worst disease*	Pathology	Location	GEO database accession number (GSE102232)
FVB	Epithelia	No estrogen	5	No disease	Normal	Endocervix	GSM2730977
				No disease	Normal	Outer cervix	GSM2730978
				No disease	Normal	Endocervix	GSM2730979
				No disease	Normal	Endocervix	GSM2730980
				No disease	Normal	Outer cervix	GSM2730981
K14E6/E7	Epithelia	No estrogen	7	CIN2	CIN1/2	Endocervix	GSM2730982
				CIN2	CIN1/2	Outer cervix	GSM2730983
				No disease	Normal	Endocervix	GSM2730984
				CIN1/2	CIN1/2	Endocervix	GSM2730985
				CIN1/2	CIN1/2	Outer cervix	GSM2730986
				CIN1/2	CIN1/2	Endocervix	GSM2730987
				CIN1/2	CIN1/2	Endocervix	GSM2730988
FVB	Stroma	No estrogen	5	No disease	Normal	Endocervix	GSM2730989
				No disease	Normal	Outer cervix	GSM2730990
				No disease	Normal	Endocervix	GSM2730991
				No disease	Normal	Endocervix	GSM2730992
				No disease	Normal	Outer cervix	GSM2730993
K14E6/E7	Stroma	No estrogen	7	CIN2	CIN1/2	Endocervix	GSM2730994
				CIN2	CIN1/2	Outer cervix	GSM2730995
				No disease	Normal	Endocervix	GSM2730996
				CIN1/2	CIN1/2	Endocervix	GSM2730997
				CIN1/2	CIN1/2	Outer cervix	GSM2730998
				CIN1/2	CIN1/2	Outer cervix	GSM2730999
				CIN1/2	CIN1/2	Endocervix	GSM2731000
K14E6/E7	Stroma	Estrogen (1 mo)	5	Cancer	Proximal to cancer	Endocervix	GSM2731001
				Cancer	Proximal to cancer	Endocervix	GSM2731002
				Cancer	Proximal to cancer	Endocervix	GSM2731003
				Cancer	Proximal to cancer	Endocervix	GSM2731004
				Cancer	Proximal to cancer	Endocervix	GSM2731005
FVB	Stroma	Estrogen (6 mo)	3	Hyperplasia	Normal	Endocervix	GSM2731006
				Hyperplasia	Normal	Endocervix	GSM2731007
				Hyperplasia	Normal	Endocervix	GSM2731008

^*Grade of worst disease determined by disease present in epithelial compartment.

Epithelial Expression of HPV Oncogenes Dramatically Alters Host Gene Expression in Adjacent Cervical Stroma.

We first compared gene expression in the cervical epithelium and stroma of E6/E7 and WT mice. HPV E6 and E7 significantly altered gene expression in the epithelium: 207 unique annotated genes were up-regulated, and 853 genes were down-regulated, in the K14E6/E7 cervical epithelium (Fig. 1A and Dataset S1). Strikingly, the HPV oncogenes had substantial effects on gene expression in the surrounding stroma: 411 genes were up-regulated, and 532 genes were down-regulated, in the cervical stroma adjacent to K14E6/E7 epithelia (Fig. 1A and Dataset S2 A and B). Differentially expressed genes in the K14E6/E7 epithelium and its adjacent stroma were largely exclusive to each compartment, suggesting that the HPV oncogenes cause gene-expression changes in the stromal microenvironment distinct from those in the epithelium.

Fig. 1.

Epithelial expression of HPV oncogenes alters host gene expression in the stroma and affects the transcriptional response to estrogen. (A) Venn diagrams showing comparative analyses of differentially expressed genes up-regulated (Upper) and down-regulated (Lower) in the K14E6/E7 cervical epithelium (red) and stroma (white) compared with nontransgenic (E6/E7 vs. WT). Genes shared between the epithelium and stroma are shown in pink. (B, Left) Three-way Venn diagram analysis of differentially expressed genes in the stroma of E6/E7 (red), WT+E2 (blue), and E6/E7+E2 (light gray) compared with WT mice. The total number of differentially expressed genes for each given comparison is indicated in parentheses. See Dataset S2 G and H for detailed lists of genes in this comparative analysis. (Right) Pie charts showing the percentage of total genes up-regulated (n = 466) (Upper) and down-regulated (n = 279) Lower) in E6/E7+E2 versus WT stroma that are also differentially expressed in E6/E7 (HPV, red segment), WT+E2 (Estrogen, blue segment), all three groups (HPV and Estrogen, black segment), or are unique to the E6/E7+E2 stroma (Unique, gray segment). (C) The 50 genes with the highest increased (Left) and decreased (Right) fold changes between E6/E7+E2 and WT cervical stroma are shown in black. Gene names are given on the x axis, and fold change is indicated on the y axis. For each gene, the fold changes measured for the WT+E2 versus WT comparison (blue bars) and E6/E7 versus WT comparison (red bars) are superimposed. Genes unique to the E6/E7+E2 cervical stroma lack superimposed blue or red bars.

Gene ontology (GO) analysis (Fig. 2A and Table S2) showed that many genes up-regulated in the K14E6/E7 stroma are involved in epithelial processes such as keratinization, keratinocyte differentiation, and epidermis development as well as intermediate filament organization, cell-cycle, metabolic, and apoptotic processes. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways associated with genes up-regulated in the transgenic stroma included the p53 signaling pathway, cell cycle, and DNA replication. Ontology associated with down-regulated genes in the K14E6/E7 stroma highlighted dysregulation of genes involved in stromal tissue architecture, such as cell adhesion, extracellular matrix (ECM) organization, and collagen fibril organization. KEGG pathway analysis reinforced this observation, indicating significant association with focal adhesion, ECM–receptor interaction, and gap junction. We conclude that HPV oncogene expression in the epithelial compartment changes the expression of genes in the microenvironment that are largely involved in both epithelial and stromal tissue dynamics.

Fig. 2.

GO analysis of differentially expressed genes. (A) GO of up-regulated (Upper) and down-regulated (Lower) genes differentially expressed in K14E6/E7 versus nontransgenic stroma. The top five most significant GO groups are shown (for P values, see Table S2) along with GO identifiers in parentheses. The five genes classified in each GO group with the highest fold-change values measured between E6/E7 and WT groups are shown in the bar graphs. (B) GO comparisons of genes differentially expressed in WT+E2 versus WT stroma (blue bars) and E6/E7+E2 versus WT stroma (gray bars). The top 10 most significant GO groups are shown (for more details, see Tables S3 and S4) for the up-regulated genes (Upper) and down-regulated genes (Lower). The number of genes in each GO group that are differentially expressed in each comparison is indicated in the bar graphs.

Table S2.

GO analysis of genes differentially expressed in K14E6/E7 (E6/E7) stroma versus nontransgenic (WT) stroma

GO identifier	GO description	Count*	Percentage†	P value‡
Up-regulated
Biological processes
GO:0031424	Keratinization	12	42.86	2.24 E -14
GO:0030216	Keratinocyte differentiation	11	23.91	4.27 E -10
GO:0008544	Epidermis development	8	24.24	3.49 E -07
GO:0045109	Intermediate filament organization	6	42.86	6.81 E -07
GO:0007049	Cell cycle	23	4.32	1.26 E -06
GO:0008152	Metabolic process	43	2.58	2.63 E -06
GO:0006915	Apoptotic process	20	3.97	3.40 E -05
GO:0006810	Transport	38	2.35	1.12 E -04
GO:0018149	Peptide cross-linking	5	26.32	1.13 E -04
GO:0010466	Negative regulation of peptidase activity	9	8.57	1.19 E -04
KEGG pathways
KEGG:04115	p53-signaling pathway	9	13.43	3.51 E -06
KEGG:04110	Cell cycle	9	7.32	3.31 E -04
KEGG:03030	DNA replication	5	14.29	1.24 E -03
KEGG:00480	Glutathione metabolism	5	10.00	5.36 E -03
KEGG:05200	Pathways in cancer	10	3.12	4.00 E -02
Down-regulated
Biological processes
GO:0007155	Cell adhesion	38	8.44	3.74 E -17
GO:0010811	Positive regulation of cell-substrate adhesion	11	34.38	8.16 E -11
GO:0030198	ECM organization	15	17.65	9.42 E -11
GO:0007507	Heart development	17	10.83	6.54 E -09
GO:0030199	Collagen fibril organization	9	32.14	1.23 E -08
GO:0006355	Regulation of transcription, DNA-dependent	53	3.05	5.35 E -07
GO:0030335	Positive regulation of cell migration	13	11.30	6.10 E -07
GO:0070208	Protein heterotrimerization	5	71.43	1.20 E -06
GO:0006351	Transcription, DNA-dependent	49	3.04	2.02 E -06
GO:0045893	Positive regulation of transcription, DNA-dependent	24	4.71	6.04 E -06
KEGG pathways
KEGG:04510	Focal adhesion	33	16.58	5.54 E -25
KEGG:04512	ECM–receptor interaction	23	26.74	1.65 E -22
KEGG:04974	Protein digestion and absorption	14	17.95	4.59 E -11
KEGG:05146	Amoebiasis	14	12.28	5.59 E -09
KEGG:05200	Pathways in cancer	22	6.85	5.84 E -09
KEGG:04270	Vascular smooth muscle contraction	13	11.30	5.89 E -08
KEGG:05412	Arrhythmogenic right ventricular cardiomyopathy	9	12.16	8.12 E -06
KEGG:04540	Gap junction	9	10.47	2.59 E -05
KEGG:05215	Prostate cancer	9	10.11	2.77 E -05
KEGG:05414	Dilated cardiomyopathy	9	10.23	2.80 E -05

Items listed are restricted to the top 10 most significant GO terms where applicable.

^*Number of annotated genes in input gene list included in GO term.

^†Percentage of genes within GO term associated with input gene list.

^‡Corrected hypergeometric P value.

Estrogen Affects Stromal Gene Expression Differently in K14E6/E7 Versus Nontransgenic Mice.

To examine how estrogen affects stromal gene expression, differentially expressed genes were determined by comparing WT+E2 and E6/E7+E2 mice with WT mice. All E6/E7+E2 mice developed squamous cell carcinomas, and these stroma samples therefore represented the tumor microenvironment.

Estrogen treatment up-regulated 461 genes in the WT+E2 stroma and 466 genes in the E6/E7+E2 cervical stroma (Dataset S2 C and E). E6/E7+E2 stroma and WT+E2 stroma shared some Gene Ontology (GO) associations, but a large proportion in the E6/E7+E2 stroma was distinct (Fig. 2B and Tables S3 and S4). For instance, inflammatory response, keratinization, antigen processing and presentation, neutrophil chemotaxis, cellular response to IFN-γ, and NOD-like receptor signaling were significantly associated with up-regulated genes in E6/E7+E2 stroma but not in WT+E2 stroma. Estrogen treatment down-regulated 140 genes in WT+E2 stroma and 279 genes in E6/E7+E2 stroma (Dataset S2 D and F). A large overlap in processes negatively affected by estrogen in the WT+E2 and E6/E7+E2 cervical stroma included GO groups such as focal adhesion, ECM–receptor interaction, collagen fibril organization, and cell adhesion. However, these processes were more significantly associated with down-regulated genes in E6/E7+E2 mice than in the WT+E2 mice (compare Tables S3 and S4). Other GO associations were unique to genes down-regulated in E6/E7+E2 stroma, including cell migration, negative regulation of epithelial cell proliferation, and the TGF-β–signaling pathway. Collectively, the gene-expression patterns indicated that estrogen treatment affected distinct biological processes in the cervical stroma of K14E6/E7 versus nontransgenic mice.

Table S3.

GO analysis of genes differentially expressed in estrogen-treated nontransgenic mice (WT+E2) stroma versus nontransgenic (WT) stroma

GO identifier	GO description	Count*	Percentage†	P value‡
Up-regulated
Biological processes
GO:0006810	Transport	61	3.77	1.90 E-13
GO:0008152	Metabolic process	54	3.25	2.19 E-09
GO:0007155	Cell adhesion	26	5.78	5.84 E-09
GO:0006811	Ion transport	27	4.88	7.77 E-08
GO:0006915	Apoptotic process	25	4.96	2.09 E-07
GO:0006821	Chloride transport	11	13.41	4.26 E-07
GO:0055085	Transmembrane transport	21	4.73	7.55 E-06
GO:0045087	Innate immune response	11	7.75	9.94 E-05
GO:0006508	Proteolysis	20	3.58	0.0008
GO:0034220	Ion transmembrane transport	15	4.53	0.0008
KEGG pathways
KEGG:04514	Cell adhesion molecules (CAMs)	14	9.79	1.50 E-07
KEGG:04610	Complement and coagulation cascades	9	11.84	1.44 E-05
KEGG:04530	Tight junction	10	7.58	9.98 E-05
KEGG:04972	Pancreatic secretion	9	8.91	0.0001
KEGG:04670	Leukocyte transendothelial migration	9	7.69	0.0002
KEGG:04978	Mineral absorption	6	13.33	0.0003
KEGG:04970	Salivary secretion	7	9.46	0.0004
KEGG:04142	Lysosome	8	6.56	0.0013
KEGG:05150	Staphylococcus aureus infection	5	10.00	0.0040
KEGG:00330	Arginine and proline metabolism	5	9.26	0.0051
Down-regulated
Biological processes
GO:0030199	Collagen fibril organization	6	21.43	3.72 E-07
GO:0007155	Cell adhesion	13	2.89	2.29 E-06
GO:0001568	Blood vessel development	6	10.17	1.36 E-05
GO:0071230	Cellular response to amino acid stimulus	6	5.66	3.37 E-04
GO:0008284	Positive regulation of cellular proliferation	9	2.59	0.0005
GO:0006813	Potassium ion transport	5	3.55	0.0078
GO:0045893	Positive regulation of transcription, DNA-dependent	8	1.57	0.0115
GO:0043066	Negative regulation of apoptotic process	6	1.76	0.0195
GO:0055085	Transmembrane transport	6	1.35	0.0311
GO:0006811	Ion transport	7	1.27	0.0424
KEGG pathways
KEGG:04510	Focal adhesion	16	8.04	1.24 E-15
KEGG:04512	ECM–receptor interaction	11	12.79	4.41 E-13
KEGG:05146	Amoebiasis	9	7.89	6.87 E-09
KEGG:04974	Protein digestion and absorption	7	8.97	2.16 E-07
KEGG:04270	Vascular smooth muscle contraction	7	6.09	2.58 E-06
KEGG:05200	Pathways in cancer	7	2.18	1.77 E-03
KEGG:04080	Neuroactive ligand–receptor interaction	5	1.81	2.00 E-02

Items listed are restricted to the top 10 most significant GO terms where applicable.

^*Number of annotated genes in input gene list included in GO term.

^†Percentage of genes within GO term associated with input gene list.

^‡Corrected hypergeometric P value.

Table S4.

GO analysis of genes differentially expressed in estrogen-treated K14E6/E7 mice (E6/E7+E2) stroma versus nontransgenic (WT) stroma

GO identifier	GO description	Count*	Percentage†	P value‡
Up-regulated
Biological processes
GO:0006954	Inflammatory response	20	11.17	2.16 E -11
GO:0045087	Innate immune response	17	11.97	3.30 E -10
GO:0006955	Immune response	16	11.35	2.47 E -09
GO:0031424	Keratinization	9	32.14	4.60 E -09
GO:0006915	Apoptotic process	25	4.96	1.80 E -07
GO:0008152	Metabolic process	47	2.82	1.24 E -06
GO:0019882	Antigen processing and presentation	7	26.92	2.12 E -06
GO:0030593	Neutrophil chemotaxis	7	25.93	2.20 E -06
GO:0071346	Cellular response to IFN-γ	6	37.50	2.25 E -06
GO:0042832	Defense response to protozoan	6	35.29	2.51 E -06
KEGG pathways
KEGG:05150	S. aureus infection	12	24.00	3.77 E -11
KEGG:04514	Cell adhesion molecules (CAMs)	16	11.19	5.12 E -10
KEGG:05323	Rheumatoid arthritis	12	14.81	5.37 E -09
KEGG:05140	Leishmaniasis	10	15.63	8.61 E -08
KEGG:05152	Tuberculosis	14	8.09	3.06 E -07
KEGG:04670	Leukocyte transendothelial migration	11	9.40	1.82 E -06
KEGG:04380	Osteoclast differentiation	11	9.48	1.94 E -06
KEGG:04621	NOD-like receptor-signaling pathway	8	14.29	3.25 E -06
KEGG:04620	Toll-like receptor-signaling pathway	10	10.00	3.45 E -06
KEGG:05160	Hepatitis C	11	8.09	5.92 E -06
Down-regulated
Biological processes
GO:0030199	Collagen fibril organization	9	32.14	1.03 E -10
GO:0007155	Cell adhesion	20	4.44	8.43 E -09
GO:0007507	Heart development	12	7.64	1.70 E -07
GO:0001568	Blood vessel development	8	13.56	9.04 E -07
GO:0030335	Positive regulation of cell migration	10	8.70	1.06 E -06
GO:0043588	Skin development	7	16.67	1.62 E -06
GO:0071230	Cellular response to amino acid stimulus	9	8.49	3.99 E -06
GO:0016477	Cell migration	9	7.76	6.37 E -06
GO:0001501	Skeletal system development	8	9.76	6.40 E -06
GO:0050680	Negative regulation of epithelial cell proliferation	6	13.33	3.57 E -05
KEGG pathways
KEGG:04510	Focal adhesion	27	13.57	8.40 E -26
KEGG:04512	ECM–receptor interaction	20	23.26	3.81 E -24
KEGG:04974	Protein digestion and absorption	14	17.95	3.29 E -15
KEGG:05146	Amoebiasis	13	11.40	1.41 E -11
KEGG:05200	Pathways in cancer	15	4.67	6.44 E -08
KEGG:04270	Vascular smooth muscle contraction	9	7.83	1.08 E -06
KEGG:04060	Cytokine–cytokine receptor interaction	11	4.51	8.81 E -06
KEGG:05222	Small cell lung cancer	7	8.24	1.76 E -05
KEGG:04360	Axon guidance	8	6.15	2.50 E -05
KEGG:04350	TGF-β–signaling pathway	6	7.14	1.89 E -04

Items are listed restricted to the top 10 most significant GO terms where applicable.

^*Number of annotated genes in input gene list included in GO term.

^†Percentage of genes within GO term associated with input gene list.

^‡Corrected hypergeometric P value.

Interestingly, 40% (184 of 466) of the up-regulated genes and 22% (61 of 279) of the down-regulated genes in the E6/E7+E2 cervical stroma were unique, meaning their differential expression required both E6/E7 and estrogen (Fig. 1B and Dataset S2 G and H). Of the down-regulated genes, 52% (145 of 279) and 5% (13 of 279) were shared exclusively with E6/E7 and WT+E2 conditions, respectively, suggesting that HPV oncogenes drive much of the gene down-regulation in the E6/E7+E2 stroma. We conclude that the HPV oncogenes influence how estrogen affects gene expression in the microenvironment, resulting in a subset of genes responding differently to estrogen in K14E6/E7 versus nontransgenic mice.

To determine the relative magnitude of the effect of the HPV oncogenes or estrogen on gene dysregulation in the E6/E7+E2 cervical stroma, the 50 highest-fold gene-expression changes between the E6/E7+E2 and WT cervical stroma (Fig. 1C, black) were compared with their respective fold changes in E6/E7 (red) and WT+E2 (blue) stroma. These fold changes in the E6/E7+E2 cervical stroma reflected either their unique nature, indicated by genes shown in black that have no overlap with genes in red or blue (e.g., CXCL1, GBP2, HDC) or strong synergy compared with either E6/E7 or estrogen alone (e.g., CXCL5, SPRR2D, IGJ). Conversely, nearly all the top 50 genes most down-regulated in the E6/E7+E2 stroma showed overlap with both factors, particularly with E6/E7. We conclude that the HPV oncogenes and estrogen not only function individually but also cooperate to alter stromal gene expression.

Candidate Paracrine Factors in the Cervical Microenvironment.

We hypothesized that estrogen induces paracrine factors in the K14E6/E7 stroma to promote cervical carcinogenesis in the adjacent HPV⁺ epithelia. Therefore, we compared differentially expressed stromal genes ontologically annotated as components of the extracellular space, extracellular region, ECM, and/or proteinaceous ECM (Fig. 3). In E6/E7+E2 cervical stroma, 87 extracellular genes were up-regulated, and 61 genes were down-regulated. Consistent with the unique effects of estrogen on gene expression in the K14E6/E7 stroma, more than 25% (25 of 87) of the up-regulated extracellular genes were unique to the E6/E7+E2 condition. With this list of “extracellular” genes, we thus identified a collection of candidate paracrine factors in the cervical microenvironment regulated by estrogen and/or epithelial HPV oncogene expression.

Fig. 3.

Identification of candidate paracrine factors in the cervical stroma. Overview of analysis of differentially expressed extracellular genes. Significantly differentially expressed genes determined previously for E6/E7, WT+E2, and E6/E7+E2 compared with WT were used as input for cellular component GO analysis. Results that were classified as extracellular (GO:0005578, GO:0005576, GO:0031012, and GO:0005615) were pooled to generate a consolidated list of extracellular genes. The lists of up-regulated (A) and down-regulated (B) extracellular genes were compared using Venn diagrams. The total numbers of significantly differentially expressed extracellular genes are shown in parentheses under each comparison label. The diagrams on the right contain gene symbols corresponding to the Venn diagrams.

Genes Involved in Inflammatory Response Are Increased by Epithelial E6/E7 Expression and Further Enhanced by Estrogen.

We narrowed our focus to the extracellular genes up-regulated in the E6/E7+E2 stroma, as they could function as positive-acting factors on the nearby epithelia. GO of these genes identified a strong inflammatory and immune response signature (Fig. 4A). The top five most significant GO classes included cytokine activity, chemokine activity, inflammatory response, immune response, and cytokine–cytokine receptor interaction. A core group of genes driving these signatures were proinflammatory cytokines, including CCL3, CCL6, CCL28, CXCL1, CXCL2, CXCL3, CXCL5, IL1A, and IL1B, as well as the S100A8 and S100A9 genes that have well-characterized roles in inflammation and carcinogenesis (Fig. 4A) (35). CXCL1, CXCL2, CXCL3, and CXCL5 were among the genes most up-regulated in the E6/E7+E2 stroma (Dataset S2E). These members of the ELR⁺ CXC family of chemokines are CXCR2 receptor ligands and comprise a signaling axis involved in several facets of tumorigenesis (36). Their identification as some of the most highly up-regulated genes in the E6/E7+E2 cervical stroma makes them prime candidates for paracrine factors in epithelial–stroma communication.

Fig. 4.

Inflammation-associated gene expression is increased by epithelial E6/E7 expression and is exacerbated by chronic estrogen treatment. (A) GO was performed on the 87 extracellular genes up-regulated in the E6/E7+E2 stroma identified in Fig. 3A. The most significant GO groups associated with these genes are shown. Individual genes driving the significant GO associations are shown; a plus sign denotes that the gene is classified in the corresponding GO group. The fold change of each gene in the E6/E7+E2 stroma compared with WT stroma is shown in the bar graph. Error bars indicate SD. (B) Fold-change values are shown for a representative group of core genes (x axis) in the inflammatory signature identified in A. The fold change of each gene in the stroma of WT+E2 (blue bars), E6/E7 (red bars), and E6/E7+E2 (purple bars) compared with WT stroma is shown. Error bars indicate SD. (C) RT-PCR validation of selected inflammation-associated genes. RNA from laser-captured stroma was used to measure transcript levels of the indicated genes using RT-PCR. Statistical analysis was performed using a Wilcoxon rank-sum test. ^#P ≤ 0.01, *P ≤ 0.05 compared with WT. Error bars indicate SD. (D) Comparison of fold-change values for selected inflammation-associated genes in the epithelium (red) and stroma (gray) in WT+E2 (rearmost values), E6/E7 (middle values), and E6/E7+E2 (foremost values) compared with WT stroma.

Interestingly, many of the inflammation-associated genes were highly up-regulated in the E6/E7 stroma even in the absence of estrogen, and their expression was further increased by estrogen treatment. Estrogen treatment alone in nontransgenic mice had little, if any, effect on inflammation-associated gene expression in the stroma, although CXCL5 and S100A9 were increased slightly over WT stroma (Fig. 4B). Real-time qPCR specific to several of the inflammation-associated genes performed on independently laser-captured cervical stroma validated these observations (Fig. 4C).

We next compared the effects of HPV oncogenes and estrogen on inflammation-associated gene expression in the cervical epithelium and stroma (Fig. 4D). In WT mice, estrogen treatment had minimal effect on inflammation-associated gene expression in the stroma but significantly up-regulated expression in the epithelia. In E6/E7 mice, inflammation-associated gene expression increased in both the cervical epithelia and surrounding stroma. Expression of inflammation-associated genes in the epithelia of E6/E7+E2 mice increased modestly compared with E6/E7 cervical epithelia but increased dramatically in the stromal compartment. The level of inflammation-associated gene expression was much higher in the E6/E7+E2 stroma than in the adjacent E6/E7+E2 epithelia or in either WT+E2 or E6/E7 stroma. Thus, HPV oncogenes induced the expression of proinflammatory genes in both the epithelial and stromal compartments of the cervix. However, estrogen appeared to function synergistically with epithelial HPV oncogene expression to specifically enhance stromal proinflammatory gene expression.

Chemokine Expression Increases upon Coculture of Primary Human Cervical Cancer Cells with Cervical Fibroblasts.

Since stromal chemokine expression in E6/E7+E2 mice increased significantly (Fig. 4), we evaluated whether it was likewise affected in human cervical cancer cells. Fresh normal human cervix and cervical cancers were evaluated for tissue histopathology and immunohistochemistry for p16, a well-validated marker of HPV infection (Fig. 5A). The cancers generally showed histopathological markers of HPV⁺ cervical carcinomas, such as keratin pearls and p16⁺ cells, whereas normal cervical samples did not. From these same tissue samples, primary epithelial and fibroblast cell cultures were established (Fig. 5B). HPV-specific PCR and immunoblots verified that normal samples were HPV⁻ and that cancer samples were positive for high-risk HPV16 (cancer tissue no. 1) or HPV18 (cancer tissue no. 2) (Fig. 5B).

Fig. 5.

Chemokine expression increases upon coculture of primary human cervical cancer cells with cervical fibroblasts. (A) Tissue sections of representative primary human normal cervical and cervical cancer samples (normal tissue no. 1 and cancer tissue no. 1) were stained with H&E (Upper) or were used for immunohistochemistry for the human p16 protein (Lower). (Scale bars, 100 μM.) (B) Characterization of primary human cervical cell strains. (Upper) Representative brightfield images of epithelial (Left) and fibroblast (Right) cell strains isolated from primary human cervical samples. (Lower Left) HPV16-, HPV18-, and HPV31-specific PCR analysis of total DNA isolated from primary epithelial cell strains. Negative controls were DNA from untransfected normal cervical epithelial cells. Positive controls were DNA from normal cervical epithelial cells transfected with HPV16, HPV18, or HPV31 genomes. N#1, normal cervical epithelial cells, sample WICVX-1; N#2, normal cervical epithelial cells, sample WICVX-3; C#1, cervical cancer epithelial cells, sample WICVX-2; C#2, cervical cancer epithelial cells, sample no. 6204. (Lower Right) Immunoblot analysis of protein lysates harvested from normal (N#2) and cervical cancer epithelial (C#1) cells probed with an antibody specific to HPV16 E7 protein. β-Actin is shown as a loading control. (C) In vitro coculture experiments and analysis of secreted chemokines. The average concentration (pg/mL) of CXCL5 (Upper) and CXCL1 (Lower) in the CM of monocultured and cocultured cervical normal (normal tissue no. 2) and cancer (cancer tissue no. 2) epithelial cells and fibroblasts measured using a multiplexed bead-based screening assay. Shown are composite data from normal fibroblasts and cancer-associated fibroblasts, as no significant difference was observed between the different sources in either monoculture or coculture. Results represent three independent experiments. Epithelial-only and fibroblast-only concentration values were compared with coculture values using a multiple-experiment statistical test to preserve each experiment structure. Normal coculture and cancer coculture values were compared using a two-sided Wilcoxon rank-sum test. Error bars indicate SD.

We hypothesized that interaction between HPV⁺ cervical epithelia and the surrounding microenvironment contributes to the inflammatory signature we observed in mice. To test this, 3D cultures were generated from primary epithelial cells only, primary cervical fibroblasts only, or cocultures of both cell types. After 72 h of culture, levels of two representative chemokines, CXCL1 and CXCL5, were measured in conditioned medium (CM). Results from one representative normal cervical sample and one cervical tumor (normal tissue no. 2 and HPV18⁺ cancer tissue no. 2 in Fig. 5B) are shown in Fig. 5C. The CXCL5 level in CM from cocultures was significantly higher than in CM from monocultured normal epithelial cells (P = 0.041) and cancer epithelial cells (P = 0.018). Likewise, CXCL5 levels were higher in cocultures containing both normal (P = 0.001) and tumor (P = 9.1 × 10⁻⁵) epithelial cells and fibroblasts than in cultures of fibroblasts alone. The CXCL5 concentration was also significantly higher in CM from cocultured cancer epithelial cells and fibroblasts than in CM from cocultured normal epithelial cells and fibroblasts (P = 0.039). An additional comparison of an early-passage normal and a cervical cancer sample (normal tissue no. 1 and cancer tissue no. 1 in Fig. 5B and Fig. S2A) showed similar results. CXCL5 secretion was also increased when cervical cancer epithelial cells were cocultured with human dermal lymphatic endothelial cells (HDLECs) (Fig. S2B), suggesting that, in addition to fibroblasts, other cell types may contribute to the inflammatory signature. The amount of CXCL1 produced from cocultured normal epithelial cells and fibroblasts (P = 0.005) and cancer epithelial cells and fibroblasts (P = 0.0007) was significantly higher than that produced from fibroblast monocultures. The level of CXCL1 produced from cocultured cells was also increased over that from epithelial cell monocultures, but this difference did not reach statistical significance (normal epithelial cell coculture, P = 0.095; cancer cell epithelial cell coculture, P = 0.188) (Fig. 5C). These results indicated that the expression of two representative inflammation-associated chemokines, CXCL5 and CXCL1, increases when early-passage human cervical epithelial cells and fibroblasts are cocultured. Interestingly, levels of CXCL1 and CXCL5 secreted from cocultured cells were much higher when early-passage cancer epithelial cells were cultured with their autologous fibroblasts (Fig. S2C), suggesting an inherent relationship between an individual tumor and its microenvironment that promotes high chemokine secretion. The analysis of primary human cervical cells showed that HPV⁺ human cervical tumors interact with their surrounding microenvironment to increase inflammation-associated gene expression, as is consistent with the observations in the cervix of K14E6/E7 mice.

Fig. S2.

Chemokine analysis in cocultured primary human cervical cells. (A) Average CXCL5 concentration (pg/mL) measured in the CM of monocultured and cocultured cervical normal (normal tissue no. 1) (Left) and cancer (cancer tissue no.1) (Center) epithelial cells and fibroblasts measured using a multiplexed bead-based screening assay. The sample size of monocultured epithelial cells was insufficient for statistical analysis (n = 1 sample each). A two-sided Wilcoxon rank-sum test was performed to compare the monocultured fibroblasts with cocultured cervical normal (normal E+F; n = 5 replicates each) and cancer epithelial cells (cancer E+F; n = 4 replicates each). (Right) A comparison of two normal and two cervical cancer epithelial samples (normal tissue no. 1 and cancer tissue no. 1 are shown in gray; normal tissue no. 2 and cancer tissue no. 2 are shown in black and are the same as those shown in Fig. 5C). A Wilcoxon rank-sum test was used to compare normal cocultured values with cancer cocultured values (at least n = 4 replicates for each). (B) Average CXCL5 concentration (pg/mL) measured in the CM of monocultured and cocultured cervical normal (normal tissue no. 2) and cancer (cancer tissue no. 2) epithelial cells and lymphatic endothelial cells (LEC) measured using a multiplexed bead-based screening assay. Sample sizes were insufficient for statistical analysis. (C) Analysis of patient-matched epithelial cells and fibroblasts. Normal tissue no. 2 and cancer tissue no. 2 epithelial cells and fibroblasts were monocultured or were cocultured with only patient-matched counterpart cells. CXCL1 and CXCL5 average concentrations were measured in the CM as described previously. Sample size was insufficient for statistical analysis. All error bars in A–C represent SDs for samples in which two or more samples were tested. Where error bars are absent, only one sample was tested. Where P values are given, the sample size in each group was at least four.

Discussion

The activities of the multifunctional HPV oncogenes E6 and E7 (37) are not sufficient for cervical carcinogenesis (11). Previous work identified estrogen as a cocarcinogen that appears to function through signaling in the stroma (9, 10, 21, 27–29). In this study, we show that epithelial expression of the HPV oncogenes dramatically affects gene expression in the cervical stroma of K14E6/E7 transgenic mice and influences the effects of estrogen on gene expression in the microenvironment (Figs. 1 and 2). Thus, HPV oncogenes and estrogen, individually and synergistically, affect gene expression in the murine cervical stroma. We reasoned that HPV-dependent, estrogen-induced paracrine factors in the stroma contribute to cervical carcinogenesis, and we identified differential expression of several candidates in the cervical stroma (Fig. 3). The stroma of estrogen-treated K14E6/E7 mice had a strong inflammatory and immune-related gene-expression signature, specifically for chemokine and cytokine activity (Fig. 4). Primary human cervical cancer epithelial cells and fibroblasts similarly showed increased chemokine expression, but only when cocultured (Fig. 5). Therefore, inflammation-associated chemokines in K14E6/E7 cervical stroma are candidate genes that may function in estrogen-driven epithelial–stromal crosstalk during carcinogenesis. While several studies have analyzed gene expression in cervical cancer epithelial cells (28, 38–40), few have performed such analyses on the cervical stroma (41–43). Our results define potentially important mechanisms through which HPV-infected epithelia and the microenvironment cooperate during cervical cancer development.

The primary objective of our study was to determine mechanisms of estrogen-dependent stroma-to-epithelium signaling. Surprisingly, we discovered additional epithelium-to-stroma signaling. Because we detected (i) low K14E6/E7 transgene transcript levels in the stroma (Fig. S1C), (ii) little overlap in differentially expressed genes between stroma and epithelium (Fig. 1A), and (iii) strong opposing trends in gene-expression patterns, such as for cytokines, between the epithelium and stroma (Fig. 4D), the profound changes in the stroma of K14E6/E7 mice cannot be due to epithelial contamination of laser-captured stroma but rather is due to epithelium-to-stroma signaling between the HPV⁺ epithelia and the surrounding microenvironment.

At least two nonexclusive hypotheses could explain this paracrine effect. First, the HPV oncogenes might alter the epithelial secretion repertoire in vivo as they do in keratinocytes cultured in organotypic rafts (44), leading to stromal changes. Second, changes in the number and/or content of extracellular vesicles secreted by HPV⁺ epithelial cells and delivered to stromal cells might alter stromal gene expression. Exosomes, for instance, deliver proteins, RNA, miRNA, and/or DNA to recipient cells to modulate gene expression (45, 46). Virus infections can alter vesicle profiles (47). For example, Epstein–Barr virus–positive cancer cells release exosomes containing viral proteins, miRNAs, and signal transduction molecules that can activate protumorigenic signaling pathways in recipient cells (48, 49). Exosomes from HPV⁺ keratinocytes contain several antiapoptotic proteins (50), unique miRNAs (51, 52), and even E6 and E7 transcripts (52), suggesting transferable oncogenic potential. Thus, HPV-infected epithelial cells could use exosomes to alter gene expression in the microenvironment to support tumorigenesis. It will be important to determine not only the extent to which HPV⁺ epithelial cells alter gene expression in a paracrine fashion but also the functions of individual HPV oncogenes in this process and whether these effects are reversible, as has been done previously in epithelial cells (53). Likewise, establishing whether exosomes are involved and which stromal cell types contribute to the differentially expressed gene patterns in the K14E6/E7 cervical stroma will provide further mechanistic insight.

The striking increase in markers for epithelium-associated processes in the K14E6/E7 stroma indicated that HPV16 E6 and E7 induce an epithelium-like stromal signature (Fig. 2A and Table S2). While surprising, epithelium-specific gene expression has been reported in cervical cancer stroma. Gius et al. (42) observed an increase in the transcription of DSG3-encoding desmoglein 3, involved in epithelial cell–cell contacts, in the stromal compartment of human cervical cancers. We did not detect an increase in DSG3, but DSG1 and DSG2 were increased significantly in the K14E6/E7 stroma compared with nontransgenic stroma. Explanations for this epithelial signature in the stroma include the possibility that the HPV oncogenes elicit mesenchymal-to-epithelial transition or epithelial–mesenchymal plasticity in the stroma (54, 55). Another possibility is that HPV oncogenes increase the abundance of normally infrequent cervical mesenchymal stem cells expressing cytokeratins (56). The HPV oncogenes may induce such changes in the stroma to promote tissue regeneration through reepithelialization (57), a process that can occur by reprogramming stromal cells to become epithelial cells (58). This could also reflect a stromal response to the metabolic demands of HPV⁺ epithelia, as suggested by signatures observed in a previous report (42). Future studies will investigate this epithelial signature and determine whether the HPV oncogenes induce cellular plasticity in the stroma to facilitate neoplastic progression.

The ontology of down-regulated genes in K14E6/E7 stroma was associated with ECM organization and stromal architecture (Fig. 2A and Table S2), driven by fibronectin (FN1), more than 15 collagen genes, tenascin C (TNC), thrombospondin 2 (THBS2), and cysteine-rich protein 61 (CYR61). These data are consistent with the reported association of precancerous lesions in K14HPV16 mice with stromal collagen fibril degradation and extensive ECM remodeling (59). HPV16 E7 inhibits the fibronectin promoter (60), and the HPV oncogenes reduce THBS1 gene expression in human keratinocytes (61). Interestingly, the potent angiogenesis inhibitors THBS1, THBS2, and THBS3 thrombospondins (62) were all down-regulated in K14E6/E7 stroma, suggesting a proangiogenic effect of HPV oncogenes on the microenvironment. However, the proangiogenic CYR61 gene was among the most down-regulated genes in the K14E6/E7 cervical stroma, in contrast to its previously reported increased expression in cultured fibroblasts isolated from the stroma of K14HPV16 transgenic mice and human cervical cancers (29, 41). We postulate that HPV oncogenes reorganize the underlying stroma to accommodate processes such as vasculature reorganization, immune cell recruitment, and epithelial cell invasion. Our results indicate that, in addition to altering the local microenvironment through enzymatic means (63), the HPV oncogenes may also regulate ECM dynamics through transcriptional regulation.

In addition, our results show that the HPV oncogenes condition the stroma to respond differently to estrogen (Figs. 1 B and C and 2). Many differentially expressed genes, particularly those up-regulated, were uniquely expressed in the estrogen-treated K14E6/E7 stroma (Fig. 1B). Moreover, the HPV oncogenes and estrogen synergistically increased the expression of several genes, including proinflammatory chemokines (Figs. 1C and 4), and may cooperate to dysregulate genes involved in collagen/ECM dynamics and focal adhesion in the cervical microenvironment of E6/E7+E2 mice (compare Tables S3 and S4). Therefore, while broadly affecting some pathways, estrogen distinctly affects biological processes in the cervical stroma of K14E6/E7 mice compared with nontransgenic mice (Fig. 2B). We expect that among these HPV-dependent, estrogen-induced stromal genes will be microenvironmental factors critical for cervical carcinogenesis.

Many of the up-regulated extracellular genes in the estrogen-treated K14E6/E7 cervical stroma encode proinflammatory cytokine/chemokines (Figs. 3 and 4), including CCL3, CCL6, CCL28, CXCL1, CXCL2, CXCL3, CXCL5, IL1A, IL1B, S100A8, and S100A9. We confirmed that CXCL1 and CXCL5 are similarly increased in CM of cocultured early-passage human cervical cancer cells and fibroblasts (Fig. 5), suggesting that cell-to-cell signaling elevates these chemokines in human cervical cancer. Cytokines have well-characterized roles in immune cell recruitment, tumor cell proliferation, angiogenesis, and modulation of the tumor microenvironment (36, 64), in large part through their effects on immune and endothelial cells. The cluster of CXCL1, CXCL2, CXCL3, and CXCL5 genes in the inflammation signature are members of the ELR⁺ C-X-C family of chemokines that bind the CXCR2 receptor. Despite the increase in expression of several CXC ligands in the E6/E7+E2 stroma, we did not observe significant variation in CXCR2 receptor gene expression among various conditions. The CXCL/CXCR2 signaling axis is considered proangiogenic and is implicated in a variety of human cancers (65–70). Strikingly similar to our report here, proinflammatory signatures have been observed in stromal fibroblasts adjacent to prostate cancer (71), melanoma (72), breast cancer (73), and HPV-associated skin and cervical cancers (29, 41).

Inflammation is implicated in several facets of HPV-associated carcinogenesis (74). In K14HPV16 skin, an influx of inflammatory cells was proposed to cause stromal ECM remodeling (75), and inflammation regulates the angiogenic switch (76). Our results contrast with an earlier report by Erez et al. (41), who identified a highly similar proinflammatory signature including CXCR2 ligands in fibroblasts adjacent to dysplastic skin, but not the cervix, of HPV transgenic mice. Conversely, moderate expression of these proinflammatory genes was reported in fibroblasts isolated from human cervical cancers (29). Our results identified inflammation-associated genes as some of the most highly differentially expressed genes in the E6/E7 and E6/E7+E2 cervical stroma and revealed the origins of this response. Our tissue culture-based experiments further support the hypothesis that communication between epithelial cells and multiple stromal cell types, including fibroblasts (Fig. 5) and endothelial cells (Fig. S2B), contributes to proinflammatory gene expression. Therefore, the discordant observations between reports may be due to evaluating a single cell type rather than the microenvironment in its entirety. Nonetheless, our results show that proinflammatory cytokines and chemokines are clear candidate extracellular genes that are induced synergistically by the HPV oncogenes and estrogen in the K14E6/E7 stroma. There is evidence that prostate epithelial cell-derived IL-1 can induce the expression of several of the same CXCR2 ligands in prostate stromal cells (71). Interestingly, the IL-1 gene was significantly increased in the K14E6/E7 cervical epithelia in our study, and this factor could therefore be part of a mechanism that induces proinflammatory gene expression in the cervical stroma. Antiestrogen drugs reduced the expression of several genes in the proinflammatory signature in human cervical cancer-associated fibroblasts (29), suggesting that estrogen signaling may also contribute to chemokine expression. It will be interesting to determine the contribution of proinflammatory chemokines to cervical cancer development in estrogen-treated K14E6/E7 mice and how their stromal expression correlates with cervical cancer regression and recurrence following antiestrogen therapy (23, 77).

Additional candidate paracrine factors unique to the estrogen-treated K14E6/E7 cervical stroma were identified (Fig. 3), including IL1A and IL1B, fibroblast growth factor 9 (FGF9), and heparin-binding epidermal growth factor-like growth factor (HBEGF). IL-1α, a proinflammatory cytokine, has been suggested to provide a selective growth advantage to HPV⁺ epithelia (78). Secreted FGF9 is highly induced in gastric cancer-associated fibroblasts and promotes invasion and antiapoptotic mechanisms in gastric cancer epithelial cells (79). Moreover, estrogen and FGF9 synergize to increase the number of breast cancer stem cells (80). HBEGF, an EGF receptor (EGFR) ligand, is detected in and produced by stromal fibroblasts adjacent to cervical cancers (29, 81). Such fibroblasts increase cervical cancer epithelial cell proliferation in an HBEGF-dependent manner (81). Antiestrogen drugs reduce HBEGF gene expression in cervical cancer-associated fibroblasts, suggesting that estrogen promotes its induction (29).

In conclusion, using our in vivo mouse model of HPV-associated cervical cancer, we have shown that the HPV oncogenes and estrogen dramatically alter host gene expression in the cervical stroma. With the extensive prior results supporting stromal-to-epithelium signaling (27, 28), this reveals that the epithelial–stromal crosstalk in cervical carcinogenesis is bidirectional. These results provide a foundation for future investigations in several important facets of the progression and maintenance of cervical cancers and potentially other HPV⁺ cancers. This includes further focus on key paracrine factors and molecular pathways implicated here; their effects on angiogenesis, immune cell recruitment, and so forth; and the sequence of these events. Since our principal aim was to analyze genome-wide gene expression in the tumor microenvironment of HPV⁺, estrogen-driven cervical cancer, the current study analyzed time points soon after cancers appeared, i.e., after 6-mo estrogen treatment of K14E6/E7 mice. Our recent preliminary evidence from 1-mo estrogen treatment of K14E6/E7 mice indicates that at least some of the changes in stromal gene expression observed here are already apparent well before the appearance of cancers. Thus, in the future it will be useful to analyze intermediate estrogen treatment times to more fully define the order of stromal and epithelial changes that accompany neoplastic progression. Such studies should further illuminate how HPV oncogenes and estrogen interact across tissue compartments to drive progression and maintenance of cervical cancer and further assist the identification of therapeutic interventions.

Materials and Methods

Animals, Treatment, and Tissue Processing.

K14E6/K14E7 (referred to as K14E6/E7 herein) bitransgenic mouse lines maintained on the FVB/N genetic background have been described previously (3, 82). Mice were housed and treated in the American Association of Laboratory Animal Care-approved Wisconsin Institute for Medical Research Vivarium of the University of Wisconsin School of Medicine and Public Health (Madison, WI) according to a protocol approved by the University of Wisconsin Institutional Animal Care and Use Committee. Nontransgenic FVB/N and K14E6/E7 transgenic female mice were either untreated or treated with exogenous estrogen (17β-estradiol). Frozen reproductive tracts were cryosectioned and used for histopathological analysis to assign lesion location and grade. Regions of interest (ROIs) in the epithelial or stromal compartments were sampled using LCM.

RNA Processing and Microarray-Based Gene-Expression Analysis.

Comprehensive mRNA levels were measured essentially as described previously (40). Briefly, RNA was extracted from laser-captured material and used to generate T7 RNA polymerase promoter-linked, oligo(dT)-primed double-stranded cDNA, which was then used as a template to produce T7 transcripts complementary to all mRNAs (cRNA). These T7 transcripts were then subjected to a second round of cDNA synthesis and T7 RNA polymerase-based amplification. The resulting cRNA was hybridized on Affymetrix Mouse Genome430 2.0 microarrays.

Biostatistics and GO.

Robust microarray analysis (RMA)-normalized data were used as input, and analysis was performed using the MeV microarray analysis suite (83). Rank product analysis was performed to compare gene expression between pairs of sample groups. The resulting P values were corrected for multiple analyses using the Benjamini–Hochberg method, which sets the FDR. Statistical cutoffs were set at twofold changes in gene expression and FDR ≤0.05 unless otherwise stated. The GeneCodis program was used for ontology analysis (84–86), using significantly differentially expressed genes as input. All Venn diagrams were created using BioVenn (87).

Coculture Experiments in Early-Passage Cell Strains.

The use of human tissue samples deidentified before receipt was determined to be exempt from the need for Institutional Review Board (IRB) approval and informed consent. Primary human cervical epithelial and fibroblast cell strains were isolated from cervical normal and tumor tissues within 24–48 h of surgical resection. Epithelial cells were cultured in F medium plus 10 μM ROCK inhibitor Y-27632, a Rho kinase inhibitor that extends the life span of keratinocytes (88), with a feeder layer of mitomycin C-treated murine fibroblast J2 3T3 cells. Fibroblasts were cultured in F12 Ham’s medium supplemented with 10% FBS and 1% penicillin/streptomycin (P/S). For all in vitro experiments, derived epithelial cell strains were used before six passages, and fibroblast cell strains were used before 12 passages. hTert-immortalized HDLECs (hTert-HDLECs) have been described previously (89). 3D monocultures or cocultures were generated with fibroblasts embedded in collagen and with epithelial cells layered on top of the collagen. All cultures were maintained in F medium without ROCK inhibitor. At 72 h, CM was assayed for several analytes using a magnetic bead-based multiplex assay.

For more detailed materials and methods, please refer to SI Materials and Methods.

SI Materials and Methods

Animal Care and Estrogen Treatment.

K14E6/K14E7 (referred to as “K14E6/E7” herein) bitransgenic mouse lines maintained on the FVB/N genetic background have been described previously (3, 82). At 4–6 wk of age, nontransgenic FVB/N and K14E6/E7 female mice were anesthetized with 5% isoflurane for s.c. insertion of a continuous-release estrogen (E2) tablet (17β-estradiol; 0.05 mg/60 d) (Innovative Research of America) into the shoulder fat pads. New tablets were inserted every 2 mo as needed. Nontransgenic mice were treated for 1 mo with estrogen to induce hyperplasia, and K14E6/E7 mice were treated for an additional 5 mo to induce cervical cancer (10, 90). Mice were housed and treated in the American Association of Laboratory Animal Care-approved Wisconsin Institute for Medical Research Vivarium of the University of Wisconsin School of Medicine and Public Health (Madison, WI), according to a protocol approved by the University of Wisconsin Institutional Animal Care and Use Committee.

Cryosectioning, Histology, and LCM.

Frozen reproductive tracts embedded in Tissue-Tek optimum cutting temperature (O.C.T.) compound (Sakura) were cryosectioned to provide a series of 7-µm and 14-µm sections. One 7-µm section was H&E stained and used for histopathological analysis to assign lesion location and grade. Adjacent sections were lightly hematoxylin-stained and dehydrated through increasing ethanol concentrations. A PixCell II Laser Capture Microscope (Applied Biosystems/Arcturus) was then used to sample epithelial or stromal compartments.

RNA Extraction, Amplification, Labeling, and Microarray-Based Gene-Expression Analysis.

Comprehensive mRNA levels were measured essentially as described previously (40). Briefly, RNA was extracted using TRIzol (Invitrogen) according to the manufacturer’s instructions and was used to generate T7 RNA polymerase promoter-linked, oligo(dT)-primed double-stranded cDNA, which was then used as a template to produce T7 cRNA transcripts. These T7 transcripts were then subjected to a second round of cDNA synthesis and T7 RNA polymerase-based amplification. The resulting cRNA was hybridized on Affymetrix MouseGenome430 2.0 microarrays (Affymetrix).

Biostatistics.

RMA was performed on the resulting Affymetrix data. RMA-normalized data were used as input, and analysis was performed using MeV (mev.tm4.org/). Rank product analysis was performed to compare gene expression between pairs of sample groups. The resulting P values were corrected for multiple analyses using the Benjamini–Hochberg method, which sets the FDR. Statistical cutoffs were set at twofold changes in gene expression and a FDR ≤0.05 unless otherwise stated.

Data Access.

A complete set of Affymetrix CEL files and GC-RMA–normalized gene-expression values for specimens was deposited in the National Center for Biotechnology Information Gene Expression Omnibus database under accession no. GSE102232. Individual sample accession numbers are listed in Table S1.

GO.

The GeneCodis program was used for ontology analysis (84–86). Significantly differentially expressed genes (fold change >2, corrected P value ≤ 0.05) as determined by rank product analysis were used as input, and up-regulated and down-regulated genes were analyzed separately to gauge directional GO associations. Results were parsed to include only GO terms with at least five genes and P value ≤ 0.05.

Gene-Expression Measurement by Real-Time qPCR.

Additional cryosections were used to independently validate the Affymetrix microarray-based mRNA-profiling data by real-time qPCR for a small subset of genes. Using the criteria and procedures describe above, RNA was extracted from laser-captured cells and used for cDNA synthesis, T7 RNA polymerase-based linear amplification, and subsequent second-round cDNA synthesis. Using TaqMan technology-based real-time qPCR with a Bio-Rad CFX96 real-time qPCR thermocycler and SsoFast Probes Supermix reagents according to the manufacturer’s instructions, we used 1-µL aliquots of 20-µL final cDNA preparations to determine the cDNA levels of 10 mRNAs: KRT14, CXCL1, CXCL2, CXCL3, CXCL5, IL1A, IL1B, S100A8, S100A9, and the HPVE6/E7/KRT14 virus-mouse chimeric transcript. Levels of ACTB and GAPDH mRNAs were measured for normalization. Reference measurements to generate standard curves were obtained using serial dilutions of synthetic DNA fragments (gBLOCKs; Integrated DNA Technologies) containing primer/probe set target sequences for each of the mRNAs. The following primers were used in analysis: K14E6/E7 transcript (F: 5′-ACCTGTTAATGGGCACACTAG-3′ and R: 5′-TCCAGCTGTGAAGTGCTTG-3′); CXCL1 (F: 5′-GAGACCACTAAGTGTCAACCAC-3′ and R: 5′-CACACATGTCCTCACCCTAATAC-3′); CXCL2 (F: 5′-GGTTTCAGT GTTTGTAAACTGTATG-3′ and R: 5′-GTGCCTTACGAGGAAGACATAA-3′); CXCL3 (F: 5′-TTTGGTGGCAGCTGTGATAG-3′ and R: 5′-TGTGACACCGTAAGACCATTTC-3′); CXCL5 (F: 5′-CCTGGGCTGGCATATAACTT-3′ and R: 5′-AGATGAGAACACATTGATCCTTCTA-3′); IL1A (F: 5′-CCACGAAGCTCTCTGTACATTC-3′ and R: 5′-GCTTTAAGGATGGGAGGGAAA-3′); IL1B (F: 5′-GGTACATCAGCACCTCACAA-3′ and R: 5′-TTAGAAACAGTCCAGCCCATAC-3′); S100A8 (F: 5′-GTGACAATGCCGTCTGAACT-3′ and R: 5′-GGGCATGGTGATTTCCTTGTA-3′); and S100A9 (F: 5′-GAGGAGTGTATGATGCTGATGG-3′ and R: 5′-GTCACATGGCTGACCTCTTAAT-3′).

Early-Passage Cell Strains and Cell Culture.

Primary human cervical epithelial and fibroblast cell strains were isolated from normal tissue no. 1 (WICVX-1), normal tissue no. 2 (WICVX-3), and cancer tissue no. 1 (HPV16⁺ WICVX-2). Human cervical tissue samples were kindly provided by J.S.R. Cancer tissue no. 2 (no. 6204; HPV18⁺ cervical adenocarcinoma) and samples from which three primary fibroblast strains (no. 5559, no. 5602, and no. 6204) were derived were procured from the National Disease Research Interchange. Within 24–48 h of surgical resection, tissues were incubated overnight at 4 °C with P/S, gentamycin, fungizone, and 1 unit/mL dispase. Epithelial layers were peeled from digested tissue, and the remaining tissue was used for fibroblast isolation. Both cell preparations were minced; then the epithelial portion was treated with 0.25% trypsin at 37 °C, and the stromal portion was treated with 0.35% collagenase at 37 °C. The epithelial cell preparation was cultured in F medium + ROCK inhibitor, composed of three parts F12 Ham’s medium to one part DMEM supplemented with 5% FBS, 24 μg/mL adenine, 8.4 ng/mL cholera toxin, 10 ng/mL epidermal growth factor, 2.4 μg/mL hydrocortisone, 5 μg/mL insulin, P/S, and 10 μM ROCK inhibitor Y-27632, a Rho kinase inhibitor that extends the life span of keratinocytes (no. S1049; Selleck Chemicals) (88). The fibroblast cell preparation was resuspended in F12 Ham’s medium supplemented with 10% FBS and 1% P/S. Epithelial cell colonies were pooled and cultured with a feeder layer of mitomycin C-treated murine fibroblast J2 3T3 cells. For all in vitro experiments, derived epithelial cell strains were used before six passages, and fibroblast cell strains were used before 12 passages. hTert-HDLECs have been described previously (89).

Immunohistochemistry.

Portions of primary human tissue samples were fixed in 4% paraformaldehyde and were paraffin-embedded, and 5-μM sections were cut from the tissue blocks. Sections were deparaffinized and rehydrated with xylenes and graded ethanol, respectively. Endogenous peroxidase activity was quenched with 3% H₂O₂ in methanol followed by heat-induced antigen retrieval in 10 mM citrate buffer (pH 6.0). Immunohistochemistry used anti-p16^INK4A (p16) mouse monoclonal antibody (CINtec p16, clone E6H4; Ventana Medical Systems, Inc.), and biotinylated horse anti-mouse IgG (Vector Laboratories) as a secondary antibody. Proteins were visualized using 3,3′-diaminobenzidine (Vector Laboratories), and tissues were counterstained with hematoxylin. Images were acquired on a Zeiss Axio Imager M2 microscope using AxioVision software version 4.8.2.

Immunoblotting.

Protein concentrations of lysates from primary cell strains were determined using Bio-Rad Protein Assay reagent (Bio-Rad). Equivalent amounts of protein were resolved in precast Mini-PROTEAN TGX Any kD or Mini-PROTEAN TGX 4–20% gradient gels (Bio-Rad) and were transferred to a 0.45-μM nitrocellulose membrane (Whatman Protran BA85; GE Healthcare). Following transfer, membranes were blocked with 5% nonfat dry milk in TBS-BGT (Tris-buffered saline containing BSA and glycine supplemented with 0.1% Tween-20). Proteins were detected using anti–β-actin (1:5,000 dilution) (Sigma) and anti-HPV16 E7 (1:500 dilution) (Santa Cruz Biotechnology). Horseradish peroxidase-conjugated secondary antibodies (1:10,000) (Jackson ImmunoResearch) and chemiluminescent substrates (Clarity ECL Substrates; Bio-Rad) were used to visualize immune complexes on a Bio-Rad ChemiDoc Imaging System.

HPV Genotyping.

DNA was isolated using a Qiagen DNeasy Blood and Tissue kit (Qiagen). Equivalent amounts of DNA were analyzed by PCR for high-risk mucosotropic HPV types using primers specific for HPV16 (709-1: 5′-CCCGGATCCTACCTGCAGGATCAGCCATG-3′ and 709-4: 5′-GGCGGATCCTTTTATGCACCAAAAGAGAACTG-3′), HPV18 (MG#3: 5′-CGGTTGCAGCACGAATGGC-3′ and HPV18E6F: 5′-GGTCGGGACCGAAAACGG-3′), and HPV31 (HPV31E6F: 5′-CTGCCAAGGTTGTGTCATGC-3′ and HPV31E6R: 5′-CCTCCTCATCTGAGCTGTCG-3′). PCR products were resolved on a 1% agarose gel and visualized with ethidium bromide.

Analyte Measurements in Coculture CM.

For monocultures, fibroblasts were embedded in 3 mg/mL rat-tail collagen (Wako Chemicals), and epithelial cells were cultured on top of collagen. For cocultures, cells were cultured at a fibroblast:epithelial cell ratio of 4,000:20,000 with fibroblasts embedded in collagen and epithelial cells layered on top of the collagen. All cultures were maintained in F medium without ROCK inhibitor. At 72 h, CM from three replicate cultures were pooled. Undiluted, clarified CM (50 μL) was used to assay several analytes using a magnetic bead-based multiplex assay (R&D Systems) following the manufacturer’s protocol. Prepared samples were analyzed in a 96-well plate using a MagPix instrument (Luminex Corporation) and xPONENT software (Luminex Corporation). To compare analyte concentrations in epithelial or fibroblast monocultures with those in cocultured epithelial cells and fibroblasts, at least three independent biological replicates were compared using a two-sided multiple-experiment permutation test. To compare chemokine concentrations in normal epithelial cells cocultured with fibroblasts versus tumor epithelial cells cocultured with fibroblasts, a two-sided Wilcoxon rank-sum test was performed. Results were considered significant when P values were ≤0.05. Statistical analyses used the MSTAT statistical software version 6.3.1 (https://mcardle.wisc.edu/mstat/; last accessed August 11, 2017).