Mucosal Alpha-Papillomaviruses are not associated with Esophageal Squamous Cell Carcinomas: Lack of Mechanistic Evidence from South Africa, China and Iran and from a World-Wide Meta-Analysis

Abstract

Epidemiological and mechanistic evidence on the causative role of human papillomaviruses (HPV) in esophageal squamous cell carcinoma (ESCC) is unclear. We retrieved alcohol- and formalin-fixed paraffin-embedded ESCC tissues from 133 patients seropositive for antibodies against HPV early proteins, from high-incidence ESCC regions; South Africa, China, and Iran. With rigorous care to prevent nucleic acid contamination, we analyzed these tissues for the presence of 51 mucosotropic human alpha-papillomaviruses by two sensitive, broad-spectrum genotyping methods, and for the markers of HPV-transformed phenotype: (i) HPV16/18 viral loads by quantitative real-time PCR, (ii) type-specific viral mRNA by E6*I/E6 full-length RT-PCR assays and (iii) expression of cellular protein p16^INK4a. Of 118 analyzable ESCC tissues, 10 (8%) were positive for DNA of HPV types: 16 (four tumors); 33, 35, 45 (one tumor each); 11 (two tumors); and 16, 70 double infection (one tumor). Inconsistent HPV DNA+ findings by two genotyping methods and negativity in qPCR indicated very low viral loads. A single HPV16 DNA+ tumor additionally harbored HPV16 E6*I mRNA but was p16^INK4a negative (HPV16 E1 seropositive patient). Another HPV16 DNA+ tumor from an HPV16 E6 seropositive patient showed p16^INK4a up-regulation but no HPV16 mRNA. In the tumor tissues of these serologically preselected ESCC patients, we did not find consistent presence of HPV DNA, HPV mRNA or p16^INK4a up-regulation. These results were supported by a meta-analysis of 14 other similar studies regarding HPV-transformation of ESCC. Our study does not support the etiological role of the 51 analyzed mucosotropic HPV types in the ESCC carcinogenesis.

INTRODUCTION

Esophageal cancer is the eighth most common malignancy worldwide and the sixth most common cause of death from cancer¹. Of various histological subtypes esophageal squamous cell carcinoma (ESCC) is the dominant histological subtype worldwide².

ESCC incidence varies over 100-fold between low-incidence regions such as western Africa and eastern China, and high-incidence regions such as southern Africa, northern China, and the Caspian littoral of Iran¹. While tobacco smoking and alcohol consumption are the main risk factors for ESCC in many countries^3–5, additional factors play a role such as consumption of hot tea in South America and Iran, diet deficiencies in parts of South Africa and China, and the consumption of opiates in Iran (⁶ and references therein).

A potential association of mucosotropic human papillomavirus (HPV) types of the alpha-papillomavirus genus with ESCC, was first proposed in 1982 based on the histological identification of condylomatous lesions in 38% of ESCC^{7, 8}. Since then, more than 100 epidemiological and laboratory studies have reported HPV DNA presence in ESCC, with positivity varying from 0 – 100%^{9, 10}, even among studies conducted within the same geographical region^11–13. Some of the highest HPV DNA prevalences have been continuously reported from the high-incidence regions in South Africa (39%)^{9, 10, 14}, China (44%)¹⁰, and Iran (37%)¹⁰. However, in the three most recent studies that documented significant efforts to prevent potential nucleic acid cross-contamination during esophageal tissue sectioning and HPV DNA analysis, these high prevalence values for South Africa, China and Iran could not be reproduced (prevalence of HPV DNA positivity was <1%)^15–17.

While epidemiological data are an important starting point in ascertaining causation, mechanistic data are considered essential to establish a biological causal link, especially for infection-cancer associations. Cervical squamous cell carcinoma (CSCC) is the best-understood model for HPV-transformation by one of the 12 high-risk (HR-)HPV types that are considered carcinogenic to humans by the International Agency for Research on Cancer (IARC)^{3, 18–22}. In addition to HPV DNA presence, CSCC is characterized by: (i) at least 1 viral genome copy present in each tumor cell^{20, 23}, (ii) expression of viral oncogenes E6 and E7^{19, 24}, and (iii) alteration of steady state levels of cellular proteins, most consistently up-regulation of p16^{INK4a 24, 25}. Such biological evidence has been provided for the 12 HR-HPV types, but not for the low risk (LR-)HPV types in CSCC³. Studies on head-and-neck cancers have provided further evidence that the same criteria are valid for a subset of malignant head-and-neck lesions^26–29. Most importantly, these studies demonstrated that the presence of HPV DNA alone in invasive tumor tissues is insufficient proof of viral causality and could result in misclassification of malignant lesions. For HPV DNA-positive ESCC, very limited direct evidence for viral transforming activity exists: just one study demonstrated in vitro transformation of esophageal epithelium by HPV18³⁰, while two other reports that analyzed the presence of HPV genomes in esophageal cell lines had contradictory outcomes^{31, 32}. Scarce evidence has been reported on HPV viral load^33–36 or expression of viral transcripts³⁷. And while 14 studies have reported on expression of p16^INK4a protein in ESCC tissues they used varying cut-offs to define p16^INK4a positivity and reported contradicting results^{15, 34, 38–49}. As a result of inconsistent HPV DNA data and very limited direct functional evidence provided in the last 30 years of research, the worldwide expert panel of the 100^th IARC Monograph concluded that the evidence for a causal association between HPV and ESCC remains inconclusive³.

In addition to HPV functional markers that can be assessed in tumor tissues, antibodies to HPV early proteins, especially E6 and E7, have been demonstrated to be markers for HPV-driven SCC of the cervix^{50, 51}, penis⁵², and oropharynx^{51, 53}. Recently InterSCOPE, the largest sero-epidemiological study on HPV in ESCC, compared (in a blinded fashion) 1,561 ESCC case, with 2,502 control subjects from six geographical areas, for antibodies against sixteen early proteins of the eight HR-HPV types most prevalent in HPV-driven cancers, and LR-HPV types 6 and 11 which are also known to infect the upper aerodigestive tract. HPV early protein antibodies were rare; the highest prevalence in cases was 2.6% for HPV6 E6. Only two significant but weak ESCC associations were found for antibodies to E6 proteins; these were for HPV16 (OR=1.89, 95%CI=1.09–3.29, p=.023) and HPV6 (OR=2.53, 95%CI=1.51–4.25, p<.001)⁶.

In order to identify ESCC cases which were most likely to be HPV-driven, we selected patients who were seropositive for HPV early proteins in the InterSCOPE study and analyzed their tumor tissues for the presence of HPV DNA and a combination of HPV functional markers (viral load, HPV mRNA, p16^INK4a up-regulation) known to characterize HPV-driven cervical and oropharyngeal cancers, and used state-of-the-art methods, and stringent conditions, to prevent nucleic acid cross-contamination.

MATERIALS AND METHODS

Ethical clearance

All analyses were approved by the appropriate national or institutional ethics committees or review boards. Written or witnessed oral informed consent was obtained from study participants.

Study population

Sera from 1,811 ESCC patients were analyzed in the sero-epidemiological InterSCOPE study, of which 1,561 had sufficient covariate data to be included in the case-control analysis⁶. Of these 1,561 ESCC patients, 357 were seropositive to at least one of the sixteen HPV early proteins from the eight most prevalent HR-HPV types: HPV16 (E1, E2, E6, E7); HPV18 (E6, E7); HPV31, 33, 35, 45, 52, and 58 (E6); or LR-HPV types 6 and 11 (E6, E7). These 357 patients originated from six world regions with widely varied ESCC incidence: South Africa, Northern China, Brazil, Central and Eastern Europe, Australia, and Iran⁶. Alcohol-fixed or formalin-fixed, paraffin-embedded (AFPE/FFPE) tumor tissue blocks could be retrieved for HPV molecular analysis from 133 patients originating from three regions with high ESCC incidence (South Africa: 58 FFPE, Northern China: 35 AFPE, and Iran: 40 FFPE).

Preparation of tissue sections and nucleic acid extraction

Tissue blocks (40 FFPE from high-incidence ESCC region in Iran and 58 FFPE from South Africa) or sections (corresponding to 35 AFPE from Shanxi Province in Northern China) were sent to Heidelberg, Germany for molecular analysis. Sections were cut for HPV DNA and HPV RNA extractions (5 µm each), and p16^INK4a immunohistochemical staining (IHC) (4 µm), following established protocols²⁴. Rigorous care was applied to control for and prevent potential nucleic acid cross-contamination during: (i) tissue sectioning, (ii) DNA and RNA extractions, (iii) PCR/Reverse-Transcription (RT)PCR/quantitative (q)PCR analysis, and (iv) hybridization of PCR products. For each patient’s specimen, a new cutting blade was used and the sectioning area was cleaned with acetone and 70% alcohol. A sectioning control (HPV DNA-free FFPE mouse liver) was cut after each 10^th patient specimen and included in DNA/RNA extractions. DNA/RNA extraction controls (one lysis buffer included after each 11 patient samples), (RT-)PCR controls (one water aliquot and one PCR master-mix aliquot after each 14 patient samples on the 96 well-plate), and hybridization controls (two hybridization buffer samples per 96 well-plate) were included. All controls yielded HPV DNA−/RNA− results. For each 11 ESCC cases, tissue sections of pretested HPV16 DNA+/RNA+ CSCC were included in DNA/RNA extraction, (RT-)PCR runs and on hybridization plate as a positive control and to control for data reproducibility. These controls showed reproducible HPV16 DNA+ and RNA+ results in all assay runs. Detailed data of individual molecular assays per patient sample, in combination with HPV serology data from patient blood, are summarized in Supplementary Table S1.

Additional sections above and below the DNA/RNA/IHC sections were examined and verified by a pathologist (CF) for the presence of: (i) squamous cell carcinoma, (ii) ≥25% non-necrotic tumor cells (23/118 ESCC tissues had 25 – 50% of tumor cells in the biopsies, and 95/118 had >50% tumor cells), (iii) muscular wall (in surgical specimens only) to confirm esophageal origin, and (iv) for the degree of keratinization. DNA was extracted from sections by overnight Proteinase K digestion at 56oC as described²⁴. RNA was extracted from sections using the PureLink FFPE Total RNA Isolation Kit (Invitrogen) with overnight incubation at 56oC and QIAGEN DNase digestion as described²⁴.

HPV genotyping, viral load and mRNA analysis

For genotyping, 5 µl of DNA extract was used from each of the 133 specimens. Two sensitive genotyping assays targeting HPV L1 and E7 gene sequences were applied. The broad spectrum BSGP5+6+-PCR/Multiplex Papillomavirus Genotyping (BSGP5+6+-PCR/MPG) assay homogenously amplifies a ~150 bp fragment from the L1 region of 51 defined mucosotropic human alpha-papillomavirus types including all HPV types classified by IARC/WHO as carcinogenic, probably carcinogenic, or possibly carcinogenic³. These 51 HPV types hereafter are called mucosal HPV types. The assay further amplifies a 208 bp cellular β-globin sequence. The detection limits per reaction are between 10 to 1,000 copies for the viral genomes and 300 copies for β-globin²³.

The type-specific E7-PCR/MPG assay (TS-E7-PCR/MPG) utilizes HPV type-specific primer pairs targeting the E7 region of 21 genital HPV types plus primers for the amplification of a β-globin sequence^{54, 55}. The cycling conditions and the sequences of the primers have been previously described⁵⁶. Here, a modified protocol for the amplification of shorter (~100 bp) fragments for ten HPV types: HPV16, 18, 31, 33, 35, 52, 56, 66; 6 and 11, and 117 bp for β-globin, was applied. Modified, shorter primer sequences are listed in Supplementary Table S2.

To measure viral load, a multiplex HPV16/18 quantitative real-time PCR (qPCR) with very short amplicons was developed to best suit the analysis of DNA extracted from fixed tissues (Schmitt et al., in preparation). The multiplex HPV16/18 qPCR amplifies 104 bp of HPV16 E6, 110 bp of HPV18 E7, and 110 bp of β-globin sequence with a detection limit of 10 HPV plasmid and 10 β-globin copies per reaction. One µl of DNA extract was used for viral load measurements. The primer and probes sequences are listed in Supplementary Table S3.

The HPV type-specific E6*I mRNA assays developed for 20 HR/pHR-HPV types²⁴ and the E6 full length (fl) mRNA assay developed for HPV11, were applied for detection of viral transcripts. These assays amplify 65 – 75 bp HPV and 81 bp ubiquitin C (ubC) cDNA and were extensively validated on cervical and head-and-neck SCC FFPE samples, deep fresh frozen specimens (DFT) and exfoliated cells^{24, 29}. Analytical sensitivity of each assay is 10 to 100 copies per reaction for 19 HPV types and for ubC, 1,000 copies for HPV67 and 10,000 copies for HPV70²⁴. All 133 ESCC patients’ tissues were analyzed for the presence of: (i) HPV16 E6*I mRNA, (ii) ubC mRNA as a cellular mRNA positive control, and (iii) mRNA of the non-HPV16 types determined by genotyping and/or serological assays. We did not test for mRNA of LR-HPV6 since the mRNA assay for HPV6 could not be thoroughly validated on HPV6 DNA+ tissues.

Each ESCC specimen that had ≥25% non-necrotic tumor cells prior and after sectioning for HPV DNA, RNA and IHC, and yielded HPV DNA and/or β-globin DNA-positive (DNA+) signal in at least one of the three methods used in the DNA analysis, was considered DNA valid. Specimens that were HPV and/or ubC mRNA-positive (RNA+) in RNA analysis were considered RNA valid.

P16^INK4a immunohistochemistry

Expression of p16^INK4a was assessed in all 133 samples using the primary p16^INK4a antibody (clone No. E6H4, Roche MTM Laboratories, Heidelberg, Germany) in a manual IHC procedure as previously described^{24, 29, 57}. Each staining batch included tissue sections for HPV16 DNA+/RNA+ cervical cancer, and HPV DNA-negative (HPV DNA−) normal oral epithelium, which served to control for intra- and inter-day staining reproducibility and protocol performance. Expression of p16^INK4a was evaluated separately by well-defined criteria for protein down- and up-regulation by two experienced investigators blinded of HPV results (GH and DH). Evaluation involved semi-quantitative scoring of the staining intensity (0=no expression, 1 or 2=low intensity, and 3=high intensity), estimation of the percentage of stained tumor cells (<10%, 11–25%, 26–50%, 51–75% and >75%) and of the staining pattern (focal or diffuse). This thorough evaluation criterion for p16^INK4a up-regulation in relation to HPV was defined on 321 HPV DNA+/RNA+ cervical cancer specimens in collaboration with two experienced pathologists as described²². For final protein expression only two categories were applied: (i) up-regulation (diffuse p16^INK4a expression in >25% of tumor cells with intensity 3+), or (ii) down-regulation (focal or patchy p16^INK4a expression in ≤25% of tumor cells with intensity ≤3+)²⁹. Evaluation was discordant in 1 case for which a new tissue section was cut and stained for p16^INK4a, re-evaluated by both investigators (GH and DH) and consensus was reached.

Overview of total number of ESCC sera and ESCC tissues analyzed in the InterSCOPE study, and overview of the molecular methods applied for tissue analysis, is depicted in Figure 1.

Figure 1. Overview of serological and tissue analyses in InterSCOPE study.

^a 1,565 of the 1,811 sera had sufficient covariate data that were included in a previous case-control analysis (Sitas et al., 2012). ^b 118 ESCC tissues were analyzed using five molecular methods to assess: (i) presence of HPV DNA (BSGP5+6+-PCR/MPG and TS-E7-PCR/MPG), (ii) viral load (HPV16/18 E6 qRT-PCR), (iii) HPV mRNA (E6*I/E6 fl RNA RT-PCR), and (iv) expression of p16^INK4a protein as a surrogate marker of HPV-transformed phenotype (p16^INK4a IHC). ^c Of total 118 ESCC tissues, 10 (8%) were HPV DNA+ when results of both genotyping methods were combined, none of the HPV16 DNA+ tissues showed high viral load, and only a single HPV16 DNA+ case expressed viral mRNA (HPV16 DNA+/mRNA+/p16^INK4a−). Seven ESCC tissues showed up-regulation of p16^INK4a and only one tissue with p16^INK4a up-regulation was HPV16 DNA+ but showed no positivity for HPV16 mRNA (HPV16 DNA+/mRNA−/p16^INK4a +).

Meta-analysis of studies with HPV DNA and p16^INK4a data

A literature search of the PUBMED and MEDLINE databases was performed to identify studies in English language published in peer-reviewed journals by December 2014, which addressed association of HPV and esophageal cancer by including both HPV DNA analyses and analyses of p16^INK4a expression in ESCC tissues. We did not include studies that addressed presence of HPV DNA and expression of p16^INK4a in Barrett’s esophagus, esophageal adenocarcinoma or esophageal papilloma.

Fourteen previous studies which reported data on both HPV DNA testing and expression of p16^INK4a as an HPV functional marker were identified for possible inclusion in the meta-analysis^{15, 34, 38–49}. Of these fourteen studies, five were excluded for the following reasons: Antonsson and colleagues³⁸, and Koshiol and colleagues¹⁵, assessed p16^INK4a in HPV DNA+ tumors only; Malik and colleagues⁴⁷ tested only p16^INK4a positive (p16^INK4a +) tumors for HPV DNA expression; Bellizzi and colleagues identified no HPV DNA+ cases in their ESCC series and hence the study provides no information about relative probabilities of HPV DNA+ and p16^INK4a+³⁹ ; and Vaiphei and colleagues⁴⁹ did not report specific frequencies of p16^INK4a + tumors. The remaining nine studies, and our own findings (i.e. 10 studies in total), were included in the meta-analysis. Study-specific relative risks (RR) were calculated as the proportion of HPV DNA+ cases that were p16^INK4a + relative to the proportion of HPV DNA− cases that were p16^INK4a +. The pooled RR was estimated using the DerSimonian and Laird random effects model⁵⁸ and heterogeneity was assessed using the residual heterogeneity statistics from the inverse-variance fixed-effect model⁵⁹. A continuity correction of 0.2 was added to the single zero cell in Teng et al., 2014⁴⁸. A funnel plot and Egger’s test⁶⁰ were used to assess the likelihood of publication bias. In additional analysis, for the 8 out of 10 studies that provided a p16^INK4a + cut-off, a meta-regression was performed with study-specific (log) RR regressed against p16^INK4a + cut-off values.

RESULTS

Patient samples

Of the 133 ESCC tissues analyzed, 118 contained sufficient non-necrotic tumor cells (at least 25%) in sections above and below the DNA/RNA/IHC sections. Validity of DNA in DNA extracts varied between PCR methods and increased with shorter β-globin DNA amplicon size (Table 1). BSGP5+6+-PCR/MPG yielded 74/118 (63%), TS-E7-PCR/MPG 115/118 (97%), and HPV16/18 qRT-PCR 106/118 (90%) DNA valid samples (Table 1, Supplementary Table S1). All 118 tissues were positive for ubiquitin C (100%) and thus valid in RNA analysis. Two DNA extracts were invalid in all three HPV DNA assays (HPV DNA− and β-globin DNA−) but valid in RNA analysis and therefore not excluded.

Table 1.

Characteristics of ESCC patients and tissues stratified by HPV status

Parameter		All analyzed (N=133)a		Valid (N=118)b		HPV DNA− (N=108)c		HPV DNA+ (N=10)d		p-valuee
		N	(%)	N	(%)	N	(%)	N	(%)
Gender
Male		82	(65)	71	(64)	68	(67)	3	(30)	0.034
Female		44	(35)	40	(36)	33	(33)	7	(70)
No data		7		7		7		0
Age (years)
Median		62		61		61		61
Alcohol use
Ever drinker		48	(38)	42	(38)	37	(37)	5	(50)	0.502
Never		77	(62)	68	(62)	63	(63)	5	(50)
No data		8		8		8		0
Tobacco use
Currentf		58	(46)	53	(48)	48	(48)	5	(50)	1.000
Former g		19	(15)	15	(14)	14	(14)	1	(10)
Never		48	(39)	42	(38)	38	(38)	4	(40)
No data		8		8		8		0
Other: hot teag
Warm or lukewarm		19	(48)	17	(49)	16	(48)	1	(50)	1.000
Hot		15	(38)	12	(34)	11	(33)	1	(50)
Very hot		6	(15)	6	(17)	6	(18)	0	(0)
No data		93		83		75		8
Histology
Keratinizing		121	(97)	114	(97)	104	(96)	10	(100)	1.000
Basaloid		3	(2)	3	(3)	3	(3)	0	(0)
Mixed form		1	(1)	1	(1)	1	(1)	0	(0)
NA		8		0		0		0
p16INK4a status (IHC)
Negative		111	(94)	111	(94)	102	(94)	9	(90)	0.471
Positive		7	(6)	7	(6)	6	(6)	1	(10)
NA		15		0		0		0

^aAll tumor tissues analyzed. Tumor tissues that are:

^bvalid in molecular/immunohistochemical analyses,

^cHPV DNA−/RNA−,

^dHPV DNA+ only (N=9) or DNA+/RNA+ (N=1).

^eFisher’s exact test for differences in proportions of HPV DNA− and HPV DNA+ cases.

^fSmoking within 5 years of the interview.

^gQuit smoking prior to 5 years before the interview.

^hInformation about hot tea consummation was available for ESCC patients from Iran only.

NA - not analyzable

HPV DNA, viral load and mRNA expression in ESCC tissues

Of the 118 tissues, 10 (8%) were HPV type DNA+ but always by a single HPV DNA assay; four (3%) by BSGP5+6+-PCR/MPG, seven (6%) by TS-E7-PCR/MPG, and none by HPV16/18 qPCR, indicating low viral loads (Table 2). HPV types identified were: HPV16 (four tumors); HPV33, 35, 45 (one tumor each), HPV11 (two tumors); and HPV16, 70 double infection (one tumor) (Table 2).

Table 2.

Molecular characteristics of HPV DNA+ ESCC tissues

Patient Number	Patient Seropositivity	HPV type DNA			HPV type RNAd	p16INK4a IHC
		BSGP5+6+- PCR/MPGa	TS-E7- PCR/MPGb	HPV16/18 qPCRc
1	16 E6	−	16	−	−	+
2	6 E6	−	16	−	16	−
3	45 E6	−	16	−	−	−
4	16 E2	−	16	−	−	−
5	18 E6	33	−	invalid	−	−
6	16 E6	−	35	−	−	−
7	16 E1	45	−	−	−	−
8	18 E6	−	11	−	−	−
9	16 E1	11	−	−	−	−
10	16 E1	70	16	invalid	−	−

“−“, negative finding; “+“, positive finding; invalid, negative for HPV and β-globin DNA.

^aAvailable for all currently defined 51 mucosal HPV types (12 HR-, 8 pHR- and 31 LR-HPV); amplicon size: 150 bp HPV, 208 bp β-globin (total 51×118 = 6,018 reactions).

^bAvailable for 21 mucosal HPV types (12 HR-, 7 pHR- and 2 LR-HPV); amplicon size: ~100 bp HPV (for HPV6, 11, 16, 18, 31, 33, 35, 52, 56 and 66), 117 bp β-globin (total 21×118 = 2,478 reactions). For other HPV types the amplicon size was as previously published by Gheit and colleagues.

^cAvailable for HPV16 and 18; amplicon size: 104 bp HPV16, 110 bp HPV18, 110 bp β-globin (total 2×118 = 236 reactions).

^dAvailable for 21 mucosotropic HPV types (12 HR-, 8 pHR- and LR-HPV11); amplicon size: 65–75 bp HPV, 81 bp ubiquitin C. All samples were tested for the presence of HPV16 E6*I mRNA and for the presence of E6*I mRNA of the HPV type defined by genotyping assays and/or serology (total 170 reactions). E.g. for Patient Number 5, the RNA sample was analyzed for the presence of HPV16, HPV18 and HPV33 E6*I mRNA, respectively. No mRNA assay was available for LR-HPV type 6.

Of the 118 tissues analyzed in a total of 170 RNA reactions (118 reactions for HPV16 and ubC and 52 additional reactions performed for a non-HPV16 type determined by genotyping or serological assays), a single tissue positive for HPV16 DNA exclusively by TS-E7-PCR/MPG also expressed HPV16 E6*I mRNA (Table 2). None of the other HPV DNA+ or HPV DNA− tissue samples contained viral mRNA.

Expression of p16^INK4a in ESCC tissues

All 118 tissue samples were valid in immunohistochemical analysis. Up-regulation of p16^INK4a was found in 7/118 (6%) tissues (Supplementary Table S1). Only one of the 7 tissues with up-regulated p16^INK4a was HPV16 DNA+ by TS-E7-PCR/MPG only, but it had a low viral load and lacked HPV16 transcripts (Table 2).

Meta-analysis of studies with HPV DNA and p16^INK4a data

We identified 14 studies in the current literature that evaluated both tumor tissue HPV DNA positivity and expression of p16^INK4a as a functional marker in their ESCC patient series (Table 3, Figure 2), and performed a meta-analysis including our own findings. Ten of the total 15 studies were valid for the meta-analysis (see Material and Methods for details on study inclusion). Significant heterogeneity was observed among the 10 study-specific relative risks (RR) (p for heterogeneity<0.001) (Table 3, Figure 2), but removal of the outlier RR estimate from Cao and colleagues⁴⁰ left a considerably more homogenous set of estimates (p for heterogeneity=0.17). In two of the 10 studies^{40, 43}, HPV DNA+ cases were significantly more likely to be p16^INK4a + than HPV DNA− cases. No significant associations between HPV DNA and p16^INK4a were found in any of the remaining eight studies or overall (pooled RR=1.32, 95%CI 0.75 to 2.31). Neither the funnel plot nor Egger’s test suggested evidence of publication bias (p=0.19). Meta-regression suggested no evidence of a relationship between study-specific (log) RR and p16^INK4a + cut-off values (p=0.65).

Table 3.

Overview of HPV DNA and p16^INK4a studies in ESCC tissues

Nr studies	Author	Country	Years of diagnosis	Biopsy type	Genotyping method	N (ESCC patients enrolled)	N (ESCC valid DNA)a	N (HPV DNA+ ESCC)(%)	HPV types identified	VL [viral copies per cell]	N (HPV RNA+ among HPV DNA+)	N (p16INK4a+ among HPV DNA+)	N (p16INK4a+ among HPV DNA−)	Cut off for p16INK4a positivity	Cross- contamination controls
	InterSCOPE study					133	118	10 (8%)	16, 33, 35, 45, 70; 11	0.00	1 (10%)	1 (10%)	6 (6%)
		South Africa	1995–2000	AFPE	BSGP5+6+ PCR & TS-E7-PCR	58	48	7 (6%)	16, 35, 45, 70; 11	0.00	1 (10%)b	1 (10%)c	2 (2%)	≥25%	yes (A, B, C)d
		North China (Shanxi)	1997–2005	FFPE		35	35	1 (1%)	16	0.00	0 (0%)	0 (0%)	3 (3%)
		Iran	2005–2007	FFPE		40	35	2 (2%)	16, 33	0.00	0 (0%)	0 (0%)	1 (1%)
1	*Antonsson et al., 2011	Australia	2002–2005	FFPE	GP5+6+ PCR + seq.	222	222	8 (4%)	16, 35	nt	nt	4 (50%)e	nte	ns	yes (A, B, C)
2	*Bellizzi et al., 2009	USA	1994–2007	FFPE	ISH	31	29	8 (4%)	none	nt	nt	0 (0%)	7 (24%)f	≥25%	yes (C)
3	Cao et al., 2014	North China (Shandong)	2006–2008	FFPE	ISH	105	105	29 (28%)	16	nt	nt	25 (86%)	14 (18%)	≥50%	nsg
4	Castillo et al., 2011					166	166	31 (19%)	16, 18, 35, 45, 51, 68; 6	0.121 (0.05 – 0.27)h		16 (53%)i	42 (33%)i
		Pakistan	1996–2002	FFPE	SPF1/2 InnoLiPA	42	42	11 (26%)	16, 18; 6	0.124h	nt	ns	ns	≥10%	nsj
		Columbia	1996–2001			49	49	9 (18%)	16, 18, 35, 45; 6	0.251h		ns	ns
		Japan	1987–2005			75	75	11 (15%)	16, 18, 51, 68	0.072h		ns	ns
5	Castillo et al., 2006	South America				73	73	21 (29%)	16, 18			12 (57%)	25 (49%)k
		Columbia	1996–2001	FFPE	GP5+6+ PCR & seq. & Southern blot	47	47	16 (34%)	16, 18	nt	nt	ns	ns	≥10%	ns
		Chile	1996–2000			26	26	5 (19%)	18			ns	ns
6	Ding et al., 2010	North China (Henan)	2005–2008	FFPE	HPV16 E6 PCR	17	17	8 (47%)	16	nt	nt	2 (25%)	5 (56%)	≥10%	ns
7	Doxatader et al., 2011	USA	ns	FFPE	ISH	20	20	1 (17%)l	ns	nt	nt	1 (17%)l	5 (83%)l	≥50%	ns
8	Herbster et al., 2012	Brazil	2000–2009	FFPE & DFT	GP5+6+ PCR & ISH & seq.	264	264	34 (13%)	16, 18, 66	nt	nt	7 (21%)	3 (10%)	ns	yes (A, B, C)
9	*Koshiol et al., 2009	North China (Linxian)	2006–2007	FFPE & DFT	PCR (PGMY) & SPF10 LiPA	272	267	3 (1%)	16, 31; 89	nt	nt	0 (0%)	ntm	ns	yes (A, B, C)
10	Lofdahl et al., 2012	Sweden	1999–2006	FFPE	GP5+6+ PCR & Luminex	204	204	20 (10%)	16, 33, 45, 51, 52, 73, 82, 66; 42	nt	nt	4 (24%)n	18 (16%)o	nsp	yes (C)
11	*Malik et al., 2011	USA	2001–2009	FFPE	ISH	25	25	0 (0%)q	none	nt	nt	0 (0%)q	ntq	nsr	yes (C)
12	Shuyama et al., 2007	China				59	59	19 (32%)	16, 18, 51; 6	<0.00001 – 1.7		1 (5%)	10 (25%)
		Gansus	1994–2005	FFPE	SPF10 InnoLiPA & GP5+6+ PCR & Southern blot	26	26	17 (65%)	16	<0.00001	nt	nsu	3 (12%)	≥80%	ns
		Shandogs				33	33	2 (6%)	16, 18	<0.0001 – 1.7		nsu	7 (21%)
13	Teng et al., 2014	East China (Shanghai)	1999–2011	FFPE	PCR + RDB	177	177	6 (3%)	16, 35; 11	nt	nt	0 (0%)	5 (3%)	≥80%	yes (B, C)
14	*Vaiphei et al., 2013	India	ns	DFT	PCR	23	23	20 (87%)	16, 18, 39, 52, 59; 6, 11 + 11 more	nt	nt	nsv	nsv	≥40%	yes (Bw, C)

^*Studies that could not be included in meta-analyses for p16^INK4a due to limited tissue analyses (see specific footnotes).

^aNumber of valid ESCC biopsy equals number of patients enrolled in each of the studies.

^bthe single HPV16 mRNA+ case was HPV16 DNA+ by TS-E7-PCR only, did not show upregulation of p16^INK4a (<10% p16^INK4a+ tumor cells), and was not seropositive for HPV16 antibodies.

^cA single HPV DNA+/p16^INK4a+ case (>75% p16^INK4a+ cells) was HPV16 mRNA-. HPV DNA was identified by TS-E7-PCR only; patient was seropositive for HPV16 E6 antibodies.

^dUse of a new blade for sectioning of each tissue specimen (A), different laboratories for DNA extraction, PCR set up and PCR products analyses (B), use of (pretested) HPV DNA+ and HPV DNA− tissues as a control (C).

^eOnly HPV DNA+ tumors (N=8) were tested for p16^INK4a expression by IHC.

^f6/7 tumors were scored as 25–50%, and 1/7 tumors was scored as 50–75% p16^INK4a+ tumor cells.

^gCSCC was used as a positive control during ISH.

^hGeometric mean of HPV genome copies per cell across HPV16 DNA+ tumors analyzed.

ⁱ158 of 166 samples had enough material for p16^INK4a analysis (30/31 HPV DNA+ and 128/135 HPV DNA− cases).

^jFor 11 ESCC cases from Japan authors additionally analyzed tissue biopsies of tumor adjacent epithelia of which none showed HPV DNA+.

^k72/73 samples had enough material for p16^INK4a analysis (21/21 HPV DNA+ and 51/52 HPV DNA− cases).

^lonly p16^INK4a+ cases (N=6), and the equal number of p16^INK4a− cases, were tested for HPV by ISH. Only 1/6 p16^INK4a+ cases was HPV DNA+.

^mOnly HPV DNA+ ESCC were tested for p16^INK4a.

ⁿ17/20 HPV DNA+ ESCC had enough material for p16^INK4a evaluation.

^op16^INK4a was evaluated in 113/184 HPV DNA− cases (6.5 HPV DNA− controls per HPV DNA+ case).

^pSamples were considered positive if there was homogeneous p16^INK4a staining of a clear majority of tumor cells.

^qOnly p16^INK4a+ tumors (N=11) were tested for HPV DNA by ISH.

^rPercentage of p16^INK4a+ cells was not specified. Authors stated that only biopsies with nuclear or nuclear and cytoplasmic, diffuse and strong staining were considered p16^INK4a+.

^sGansu is a high-incidence ESCC region in NW China and Shandong is a low-incidence ESCC region in E China.

^tHPV51 and HPV6 were present only as a co-infection with HPV16 in two respective cases for which region of origin was not specified.

^uAuthors did not specify if the one p16^INK4a+ case was from Gansu or from Shandong.

^vNumber of p16^INK4a+ tumors was not specified (12 tumors were positive for p16 and p63). Authors concluded that p16^INK4a+ and p63 positivity did not correlate with HPV status.

^wAuthors report using sterile forceps and blades but do not specify changing them for sectioning of each tissue sample.

Abbreviations: AFPE - alcohol-fixed paraffin-embedded, FFPE - formalin-fixed paraffin-embedded, DFT - deep frozen tissue, ISH - in situ hybridization, RDB - reverse dot blot, CSCC – cervical squamous cell carcinoma, seq. – sequencing, nt – not tested, ns - not specified

Figure 2. Study-specific and pooled relative risks (RR) corresponding to the proportion of HPV DNA+ cases that were p16^INK4a + relative to the proportion of HPV DNA− cases that were p16^INK4a +.

Clinical characteristics of ESCC patients and histopathological tumor patterns according to the HPV status

ESCC patients with HPV DNA− tumors (N=108) did not differ statistically from the patients with HPV DNA+ tumors (N=10) with respect to age, alcohol consumption, smoking habits, or degree of tumor keratinization (Table 1). Interestingly, only 33% of patients with HPV DNA− tumors were female, in contrast to the 70% females among patients with HPV DNA+ tumors (p=0.019).

Pathology review of the 118 tumors revealed 114 (97%) keratinizing, 3 (3%) basaloid, and one (1%) tumor of mixed histology (keratinizing/basaloid). All HPV DNA+ (N=9) and HPV DNA+/RNA+ (N=1) tumors were keratinizing.

DISCUSSION

In order to comprehensively address the question of causal HPV involvement in the pathogenesis of esophageal squamous cell carcinoma, we used advanced molecular methods on ESCC tissue sections and searched for key molecular markers indicative of HPV-association (HPV DNA with viral load, HPV RNA and a surrogate cellular protein p16^INK4a) in a large series of 118 ESCC tissues. These analyzed 118 tumor tissues were enriched for potentially HPV-driven cancers by pre-selection among 1,561 ESCC patients who were positive for HPV early protein antibodies in their blood. Using stringent precautions to prevent nucleic acid contamination, we found HPV DNA positivity in 10 (8%) of these cases, using a combination of two highly sensitive HPV genotyping assays. One of these sensitive assays targeted the L1 region of all defined 51 mucosal HPV types, while the other evaluated the E7 region of 21 mucosal HPV types (12 HR-, 7 possibly (p)HR- and 2 LR-HPV types). All HPV DNA+ findings were inconsistent between the two genotyping assays and all samples were negative by HPV16/18 qPCR, suggesting that viral loads in HPV DNA+ ESCC tissues were extremely low and/or close to the lower limits of detection. All ESCC tissues were also analyzed for HPV16 E6*I mRNA but only one (an HPV16 DNA+ tumor) showed this marker. None of the tumors that were positive for other HPV DNA types (HPV33, 35, 45, 70, or 11) showed type-concordant E6*I or E6 fl mRNA. Of seven tumors with p16^INK4a up-regulation, only one contained HPV DNA (type 16), and all seven were negative for HPV16 mRNA. Thus, among the 118 ESCC tissues analyzed for three HPV functional markers in addition to HPV DNA (i.e. (i) viral load 1 copy per cell, (ii) presence of HPV type-concordant E6*I mRNA, and (iii) up-regulation of the cellular surrogate marker p16^INK4a), none was positive for all three or even two functional markers. Of the two HPV16 DNA+ tissues that displayed at least one functional marker, one harbored HPV16 mRNA but without p16^INK4a up-regulation and the other showed up-regulated p16^INK4a in the absence of HPV16 transcription. Based on the absence of HPV functional markers in our evaluation here, we conclude that ESCC does not follow the HPV-transformation pathogenetic model of cervical cancer, and that there is no evidence for a role of mucosal alpha-papillomavirus types investigated here, in the etiology of ESCC in the geographic high-incidence regions studied.

In line with other studies^{15, 38, 44, 48} we identified HPV DNA of mucosal alpha-papillomaviruses in a few ESCC tissues, albeit at very low copy numbers and almost always with none or only a single marker of HPV-transformed phenotype. Our findings suggest four possible sources for HPV DNA positivity in ESCC: (i) HPV DNA in the oral cavity, which could pass into the esophagus after swallowing HPV-containing saliva (it has been shown that oral rinses can contain HPV DNA)⁶¹, (ii) non-transforming HPV infection in neighboring normal esophageal tissue or even parts of the tumor (where HPV acts as a passenger and not as a transforming agent)³⁷ ; (iii) cross-contamination from other HPV-infected tissues in routine pathology tissue processing⁶² ; or (iv) cross-contamination from HPV PCR products or HPV plasmids in the laboratory analysis⁶³. Furthermore, Schaffer and colleagues have demonstrated increased uptake of HPV virions under certain conditions: HPV18 pseudovirion uptake by esophageal cells can be increased in vitro when the cells are treated with benzo-α-pyrene, an abundant carcinogen in tobacco smoke⁶⁴. In line with these functional data, epidemiological studies showed that tobacco users have higher prevalence of HPV16 DNA in their oral cavity⁶⁵, and clinical studies demonstrated that OPSCC patients who smoke have worse survival irrespective of HPV tumor status^66–68. Also, several studies suggested that there is a high risk of HPV vaginal contamination during use of routine endocavity vaginal ultrasound probes, an underestimated route of nosocomial infection^69–71. Similarly, contamination with DNA of HPV types from oral cavity could occur during esophagoscopy or esophageal tissue sampling.

In support of above described data, Gillison and colleagues demonstrated that the prevalence of oral infection with mucosal HPV types from the five HR-species among healthy men and women aged 14 to 69 years was 6.9% in the United States⁶¹. Similar data were shown in Iran where 6.1% of 114 healthy individuals age 16 – 66 years, had oral HPV DNA⁷². Bottalico and colleagues showed that 37% of 117 immunocompetent men, harbored HPV DNA in oral washes of which 28% were alpha-, 64% were beta- and 8% were gamma-papillomavirus infections⁷³. The authors have concluded that the oral cavity contains a wide spectrum of HPV types predominantly from the beta- and gamma-papillomavirus genera, which were previously considered to be cutaneous types. Similar results were found by a recent study for alpha- (72%), beta- (27%), and gamma-papillomaviruses (27%) in the anal canal⁷⁴. The presence of beta- and gamma-papillomaviruses at mucosal sites leaves an open question of their tissue tropism and their biological role e.g. biological role of beta- and gamma-papillomaviruses in ESCC, which was not examined here. However, none of these questions can be answered based on HPV DNA data only. Beta- and gamma-papillomaviruses are frequently found both in skin lesions and in healthy skin³. Further, the functional data on transforming activity for beta-papillomaviruses are scarce and limited to HPV5 and HPV38^75–80. Therefore, to substantiate a biological contribution of HPV types (alpha-, beta-, or gamma-) to the transformed phenotype of tumor cells, viral RNA or protein, representing viral activity, and expression levels of cellular proteins altered by viral oncoproteins, need to be demonstrated in addition to viral DNA.

It should be noted that in contrast to high prevalences for mucosal alpha-papillomaviruses ranging from 37% to 44% previously reported for Africa, China, and Iran¹⁰, the most recent studies that also documented significant efforts to prevent potential nucleic acid cross-contamination during tissue sectioning and HPV DNA analysis found less than one percent of HPV DNA+ ESCC cases^15–17. In agreement with these latest studies, our combination of two highly sensitive HPV genotyping methods, capable of identifying the broadest spectrum of mucosal HPV types analyzed so far, detected traces of HPV DNA in 3% to 15% of ESCC from the same geographic regions.

Our observation of low viral load in the few ESCC cases that were HPV DNA+ is in agreement with Guo and colleagues⁸¹ who in 29 Chinese ESCC tissues found a median of 0.04 HPV16 copies per cell, and only a single case with close to one copy per cell. In contrast, four other studies identified cases with HPV loads >1 genome copy per cell^33–36 with only one study³⁴ analyzing additional HPV functional markers and demonstrating p16^INK4a up-regulation (≥80% tumor cells p16^INK4a positive) in one of 19 HPV16 DNA+ cases.

The type-specific HPV RNA assays applied here have been extensively validated on FFPE tissues and shown to reliably and sensitively detect HPV E6*I mRNA of 20 mucosal alpha-papillomaviruses in cervical cancer tissues^{22, 24}. We found a single HPV16 DNA+ ESCC tumor that expressed HPV16 mRNA, however, without p16^INK4a up-regulation. HPV16 E6*I transcripts can be abundantly expressed in the cervix in the absence of malignant transformation, as well as in some non-HPV-driven oropharyngeal and laryngeal cancers^{27, 29}. The E6*I transcripts are, therefore, markers of active viral infection but are not transformation-specific; thus we did not consider this single low viral load HPV DNA+/RNA+ ESCC an HPV-driven tumor. Our RNA data disagree with one HPV16 transcription study which used in situ hybridization to identify HPV16 RNA in 19 of 31 Chinese ESCC and six of 23 normal adjacent mucosa biopsies³⁷.

Though we tested for RNA of all HPV types identified by two HPV DNA assays and also, all HPV types for which HPV antibodies were detected in patients’ sera (see Supplementary Table S1), we did not test for RNA of LR-HPV6. The HPV6 RNA assay could not be validated on HPV6 DNA+ tumor tissues in contrast to the HPV11 RNA assay^{29, 82}. Among the 118 patients with the analyzable ESCC tissues, 20 HPV6, and 127 non-HPV6 type-specific antibody responses were detected (see Supplementary Table S1). In only 3/127 (2%) of these HPV type-specific antibody responses (in three separate cases) could we also detect HPV DNA in the corresponding ESCC tissue (three of the five HPV16 DNA+ cases), and in none of them could we detect both HPV DNA and RNA, or HPV RNA only. Therefore, it is unlikely that testing for HPV6 RNA would yield crucial functional data in any of these 20 HPV6 DNA−/p16^INK4a - ESCC tissues of patients seropositive for HPV6.

In a setting of HPV-infection, p16^INK4a up-regulation is considered the best biomarker to define tumors with HPV-transformed phenotype both in research and in clinical studies^{27, 57, 83–85}. P16^INK4a up-regulation in HPV-infected cells is a result of a cellular defense mechanism referred to as oncogene-induced senescence induced by expression of the viral E7 oncoprotein^{86, 87}. Lack of CDKN2a mutations in combination with p16^INK4a up-regulation was demonstrated for HPV RNA+ cervical cancers and cervical cancer cell lines^88–91, as well as for the HPV RNA+ tumors of the head-and-neck including oral, oropharyngeal and laryngeal tumors⁹². Furthermore, immunohistochemical studies demonstrated that p16^INK4a expression gradually increases from 5% in normal cervical epithelium, to 10% in low grade squamous intraepithelial lesions (LSIL), 25 – 80% in high (H)SIL and 90 – 100% in cervical cancers^93–95. Together, these studies have established p16^INK4a up-regulation as an excellent biomarker to define HR-HPV-associated lesions and cancers in addition to HPV DNA and/or HPV RNA positivity.

With a single tumor among 10 HPV DNA+ ESCC, and six tumors among 108 HPV DNA− ESCC, that showed p16^INK4a up-regulation, we found no significant association between p16^INK4a up-regulation and HPV DNA positivity (p=0.57). For the one case with HPV16 DNA+ ESCC with p16^INK4a up-regulation in our study, the low viral load and absence of HPV16 E6*I mRNA do not support classification of this case as HPV16-driven. Nine other studies have also addressed question of p16^INK4a expression in HPV DNA+ and HPV DNA− tumors of which seven found no significant associations^{34, 41, 42, 44–46, 48}. Two studies found p16^INK4a positivity statistically significantly associated with HPV DNA+ ESCC^{40, 43}. We do not find an explanation for the high detection of p16^INK4a up-regulation in 86% of HPV DNA+ and 18% of HPV DNA− tissues by Cao and colleagues that applied a cut-off as high as 50% to define p16^INK4a positivity⁴⁰. Castillo and colleagues used 10% as a cut-off to define p16^INK4a positivity and reported high p16^INK4a positivity in both HPV DNA+ (56%) and HPV DNA− tissues (33%)⁴³. In summary, seven of nine studies specified cut-offs for definition of p16^INK4a positivity/up-regulation, however this cut-off varied from ≥10% (three studies), ≥50% (two studies) to ≥80% (two studies) indicating that evaluation of p16^INK4a in tumor tissues in relation to HPV is still a non-standardized method. Our statement is supported by, and in agreement with, a recent study that reviewed p16^INK4a and HPV DNA studies in ESCC⁹⁶. The use of p16^INK4a expression as an indicator of virally induced deregulation of the cell cycle⁹⁷ in HPV DNA+(/RNA+) tissues, requires standardization of methods and cut-off values in different anatomical sites. Only then can it be reliably and consistently used to define HPV-driven cancers.

Our study is limited by the relatively small number of tumors analyzed from each of the three high-incidence regions. Other limitations include the absence of full functional data on all samples tested, and the fact that our samples all came from high-incidence ESCC regions and may not be representative of other geographic regions. Also, if HPV was supposed to contribute to the cancer development by a never fully convincingly demonstrated “hit and run mechanism”⁹⁸, we were not able to detect this ill-defined possibility by our approach here. Further, we did not test in the tissues for DNA of human beta- and gamma-papillomavirus types whose DNA occasionally has been isolated from mucosal sites.

The major strengths of our study include the application of multiple sensitive and established technologies for the analysis of HPV DNA, the use of an array of functional markers for active HPV infection, and the use of stringent precautions to prevent and monitor laboratory contaminations. A further strength, counteracting the overall small number of tumors analyzed from each of the three individual high-incidence regions, is the unique preselection of potentially HPV-associated tumors by HPV serology. Our study also demonstrates how detailed knowledge about HPV-transformation in cervical cancer can be used to clarify the potential transforming role of HPV in other tumor sites in which HPV DNA is found.

Recently, HPV serology has been shown to be highly specific for the definition of potentially HPV-associated oropharyngeal squamous cell carcinomas (OPSCC) in which HPV16 E6 antibodies can be detected as early as 13 years prior to cancer diagnosis⁵³. More precisely, HPV16 E6 seropositivity was present in prediagnostic samples of 35% of OPSCC patients and 0.6% controls (OR, 274; 95% CI 110 to 681) but was not associated with cancer at other sites including esophagus⁵³. Here we confirm that positive HPV serology does not define an HPV-driven ESCC, and we base this conclusion not only on serology but also on a broad set of tissue biomarkers, the combination of which have been analyzed for the first time in ESCC. The only previous study that has provided data on such a broad biomarker panel was a recent study by Anantharaman and colleagues who similarly demonstrated no association of HPV with cancer of the lung despite the occurrence of HPV seropositivity in some of the lung cancer patients⁸².

In conclusion, the InterSCOPE study provides strong biological data against a transforming role of mucosal alpha-papillomavirus types in the pathogenesis of esophageal squamous cell carcinoma for the three high-incidence regions; China, South-Africa, and Iran. Further, our meta-analysis performed on published data on co-occurrence of HPV DNA and p16^INK4a up-regulation also showed no evidence for a significant association. Together with the largely negative serological findings documented previously in the sero-epidemiological InterSCOPE study for a large population of ESCC patients from all over the world, our data provide compelling evidence that mucosal alpha-papillomaviruses have little or no role in the etiology of ESCC according to the HPV-transformation model described for cervical, other anogenital and a subset of head-and-neck cancers.

The novelty and impact of the work.

South Africa, China and Iran are countries with the highest incidences and mortalities of esophageal squamous cell carcinoma (ESCC). Epidemiological studies have suggested a role for HR-HPV types in ESCC occurring in these countries, but consistent biological evidence of viral transformation, important for vaccination program planning, is still missing. Using state-of-the-art technology, we provide comprehensive data on absence of viral biological activity for 51 mucosotropic HPV types from alpha-papillomavirus genus in ESCC from high-incidence regions.