Human Papillomavirus Load in Eyebrow Hair Follicles and Risk of Cutaneous Squamous Cell Carcinoma

Keratinocytic cancers, predominantly basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), are the most frequently occurring cancers in Caucasians. The primary environmental cause of both BCCs and SCCs is exposure of the skin to solar ultraviolet radiation. However evidence is accumulating to suggest that infection with human papillomaviruses (HPV), particularly those of the beta genus (betaPV), may play a synergistic role (1). This is particularly apparent in patients with a rare genetic disorder that predisposes to high viral load betaPV infection, epidermodysplasia verruciformis, where 2 beta types, HPV5 and 8, have been classified by the International Agency for Research on Cancer as possibly carcinogenic (2). Patients who are immunosuppressed following organ transplantation are also at substantially increased risk of keratinocyte cancers, particularly SCCs (3), suggesting the involvement of an infectious agent in people without epidermodysplasia verruciformis.

The bulge area of the outer root sheath of the hair follicle has been suggested as a reservoir for cutaneous HPV, and comparative studies have shown reasonable concordance in prevalence and type multiplicity between eyebrow hair follicles and normal skin, hair follicles from other body sites, perilesional skin, and actual skin lesions (4, 5). Viral DNA in hair follicles has thus been used as an indicator of cutaneous infection in a number of epidemiologic studies. In our multicentre case–control study in immunocompetent populations, the presence of betaPV DNA in hair follicles was associated with a significant 3-fold increased risk of SCCs in the Netherlands but no significant overall association in Australia or Italy (6). In our study, in people who had been organ transplanted, there was a nonsignificant 2-fold increase in risk in people with evidence of betaPV infection in hair follicles (7). Earlier studies using older laboratory techniques were largely supportive of an association, although the data were somewhat inconsistent (8–10).

The challenge in identifying causal associations between HPV and skin disease is that it appears to be part of the commensal flora of the skin. In our recent studies more than 90% of both organ transplant recipients (OTR) and immunocompetent people were betaPV-positive (6, 7, 11). Finding markers of increased viral activity may help to elucidate the role of HPV in cutaneous carcinogenesis. HPV antibodies, present in about half of those who have evidence of eyebrow hair infection, have been associated with SCC risk in several case–control studies (12–16), possibly indicating the presence of more active virus. However, it is also possible that presence of skin cancer causes the generation of HPV antibodies and that reverse causality is responsible for this finding (17). An attempt to highlight biologically relevant betaPV infections by looking for betaPV DNA and antibody directed against the same type identified a significant association between SCCs and the concordant detection of antibodies and DNA for at least one betaPV type but no association with antibodies without concordant DNA (7). Similarly, a recent study found several type-specific associations between antibodies and SCCs in people who were DNA-positive for the same HPV type in tumor tissue but not in those whose tumor was DNA-negative (18).

Another possible indicator of increased viral activity is the betaPV DNA load. Exceptionally high loads of up to more than 400 viral DNA copies per cell were detected in eyebrow hair follicles of patients with epidermodysplasia verruciformis, who have a high risk of SCC development (19). The viral loads in eyebrow hair follicles from the general population and from OTR span 7 orders of magnitude and are mostly extremely low but OTR are more likely to have higher loads (20). This is consistent with a hypothesis that higher viral DNA loads may indicate an enhanced activity of betaPV, contributing to the strongly enhanced risk of SCC development seen in this patient group. However, the association between viral load and skin cancer has not been formally evaluated.

To explore the role of viral load in skin cancer development, we used a case–control study design to determine the association between the load of 8 different betaPV types and risk of SCCs in immunocompetent and immunosuppressed people from different geographical locations. The 8 types were selected because they are relatively prevalent in hair bulbs (11) and have been implicated in skin malignancies in patients with epidermodysplasia verruciformis (HPV5, 8, 20; ref. 21) in some previous epidemiology studies (HPV8, 36, 38; refs. 7, 13) and in studies of transgenic mice (HPV8, 38; refs. 22, 23)

Selection and characteristics of participants have been reported previously (6). The immunocompetent participants (ICP) included in this study were recruited from Queensland, Australia and Rome, Italy. In Australia, cases were recruited from primary care skin cancer clinics and controls from several sources (150 from the federal electoral roll on which enrolment is compulsory; 75 from primary care skin screening clinics or dermatology clinics and 85 from community-based volunteer groups). In Italy, both cases and controls were recruited from the outpatient dermatology department of a major hospital in Rome. However, only 27 controls had a dermatological condition (actinic keratoses, psoriasis, seborrheic warts) with the potential to influence betaPV positivity or load, with the majority having conditions such as lipoma, cysts, and other problems requiring minor surgical procedures. In both countries, cases had a histologically-confirmed incident cutaneous SCC and controls, who were matched to the cases by sex and age in 5-year age groups, had no history of squamous cell carcinoma.

The selection processes have been reported previously (7). Cases had a history of histologically-confirmed posttransplant primary cutaneous SCCs (excluding in situ carcinoma) and were recruited from outpatient dermatology and transplant clinics in the Netherlands, the United Kingdom, France and Italy. Controls had no history of skin cancer and were selected from the same clinics. We attempted to match controls to cases by sex, age, and time since transplantation (2–7, 8–12, 13–17, 18–22, 23, and more years) in a 2:1 ratio. Patients with brown or black skin (Fitzpatrick skin type V and VI) were excluded, as were patients with only liver, lung, and pancreas transplants.

All studies had local human research ethics approval.

All participants completed a questionnaire in which information was elicited about skin type, hair and eye color, sun exposure on weekends and weekdays, sunburns, smoking, education, and alcohol consumption. Medical charts of OTR were reviewed to document the timing and number of transplants and immunosuppressive medications used.

Eight to 10 eyebrow hairs were plucked from each participant and snap-frozen in liquid nitrogen. They were stored at −80°C until analysis.

DNA from plucked eyebrow hair samples was isolated with the Qiamp DNA Mini Kit (Qiagen; ref. 24). The presence or absence of individual betaPV types was determined with the skin (beta) PV prototype research assay (Diassay BV), which combines PCR detection and a reverse hybridization assay for typing of 25 genotypes (i.e., HPV types 5, 8, 9, 12, 14, 15, 17, 19, 20, 21, 22, 23, 24, 25, 36, 37, 38, 47, 49, 75, 76, 80, 92, 93, and 96) as described previously (25). Samples from ICP from Queensland, Australia and OTR from Netherlands were analyzed at the Department of Medical Microbiology of the Leiden University Medical Center (Leiden, The Netherlands). Samples from ICP from Rome, Italy, and those from OTR from the United Kingdom, France, and Italy were analyzed at the Institute of Virology of the University of Cologne (Cologne, Germany). DNA isolation and betaPV type determination were conducted at the end of the collection phase using different sets of reagents with appropriate negative and positive controls to ensure comparability of results. Interlaboratory reproducibility between both testing sites was extensively tested with bridging panels and agreement was shown to be high (25).

The amount of the single-copy gene beta-globin in each sample and the number of DNA copies of HPV types 5, 8, 15, 20, 23, 24, 36, and 38 was determined by quantitative PCR with type-specific primers and probes in the hybridization probe format using the “LightCycler-Control Kit DNA” (Roche) and protocols described before. The HPV plasmid dilution series were standardized relative to a fixed standard (24, 26). Quantitative PCR was only carried out on samples that were positive for at least one of the 8 types, and all analyses were conducted at the Institute of Virology of the University of Cologne.

We calculated viral load by dividing the number of copies of viral DNA detected by the total number of cells (estimated by the number of beta-globin copies divided by 2) in the sample. Viral load is expressed as one viral copy per X cells. Cumulative load was determined by summing the DNA copy numbers of the 8 individual HPV types. We excluded any participants who had less than 400 amplifiable beta-globin copies per 2 μL solution of extracted DNA as we have shown that HPV detection is limited by lower cellular input (27).

We classified participants first according to the presence or absence of DNA of each of the 8 specific HPV types tested and for positivity to at least one of these types. For each type and for cumulative viral load, we then divided those positive into tertiles, using the viral load distribution of cases and controls combined for each study group to identify the tertile cutoff points. For each participant, we calculated the number of types, where the viral load was in the highest tertile.

We compared the continuous viral load distributions of cases and controls for those HPV DNA-positive using Wilcoxon 2-sample tests. Participants who were completely negative for HPV DNA were not included in these analyses. For each type less than 1% of people who were positive using the Diassay BV were negative on quantitative PCR. These were also excluded from these continuous analyses. Cumulatively, 0.2% of people who were positive for at least one of the 8 types were load-negative.

Logistic regression analysis was used to estimate associations between each of the exposure measures (overall positivity, type-specific positivity, type-specific load tertiles, cumulative load tertiles) and SCCs. For type-specific viral load analyses, the reference group consisted of those who were negative for that type including those who were qualitatively positive but load negative. The number of participants negative to all 8 types was low, so for cumulative load analyses, we combined the lowest HPV load tertile and HPV DNA-negative groups. All analyses were adjusted for age and sex. Further adjustment for skin type, education, or measures of time outdoors did not alter the ORs so these variables were not included. In the OTR population, we also tested the potential confounding effect of years since transplantation and country of recruitment and found them to have negligible impact on estimates of effect.

We recruited 1,705 participants and included a total of 1,394 participants (82%) who had adequate cellular input in this analysis, 627 cases (232 ICP from Australia, 216 ICP from Italy, and 179 OTR) and 782 controls (241 ICP from Australia, 223 ICP from Italy, and 318 OTR). The proportion of men was 4% higher in both cases and controls in the included sample than in the entire study population, but there was no difference in the mean age.

Demographic characteristics of cases and controls, stratified by study group, are shown in Table 1. In all 3 groups, cases were more likely to have fair skin than controls. The matching by age and sex was successful in the Australian population, with only small differences in the distribution of cases and controls. In the Italian and the OTR datasets, the cases were somewhat older than controls (although this was statistically nonsignificant in Italy) and had a higher proportion of men (0.05 < P < 0.1).

The proportion of participants positive for at least 1 of the 8 HPV types tested for viral load ranged from 83% in Australian controls to 93% in Italian and OTR cases (Table 2). In all 3 groups, the proportion positive was higher in cases than in controls, although none of these differences was statistically significant. The prevalence of each individual HPV type is shown in Table 2. Although we observed few significant differences between cases and controls, in most cases, the prevalence was slightly higher in cases than in controls. We found a positive association between the number of HPV types positive and SCCs in Australia and OTR, although in OTR, those positive to 4 types did not follow the trend of increasing risk with increasing number of types (Table 3). For each additional type, there was approximately a 13% increased risk of SCCs (P < 0.05 for both groups). There was no apparent trend in Italy. The associations identified were almost identical when we included participants with low cellular input in the analyses (data not shown).

The median viral load of the selected 8 HPV types, cumulatively and for individual HPV types, was substantially less than one copy per cell (Fig. 1). In Australia, the median load was higher in cases than in controls for most types, but with P values greater than 0.05 (Fig. 1A). The median cumulative load was significantly higher in cases than in controls (Wilcoxon 2-sample test, P = 0.008). A similar pattern was observed for OTR, with significant differences in the distribution for HPV8 (P = 0.02) and cumulative load (P = 0.005; Fig. 1C). There were no significant differences between cases and controls for any type or for cumulative load in Italian participants (Fig. 1B).

The analyses of viral load tertiles are displayed in Fig. 2. We have also displayed ORs for type-specific HPV positivity for comparison purposes. We did not observe consistently increasing risk with increasing viral load in any group. However, in Australia, those in the highest load tertile for types HPV23 and 36 were at significantly higher risk than participants negative for each type, respectively, and there were marginally significant associations for HPV5 and 38. There was a significant positive association for those in medium and high cumulative load tertiles (compared with those in the low tertile combined with those negative as the reference group). The associations in OTR were similar to those in Australia, with significant associations for the highest load tertiles of HPV8, 38 and cumulative load and marginally significant associations for HPV20 and 36 (Fig. 2C). In Italian participants, there was a significantly increased risk in people in the low load category for HPV36, and similarly to the Australian and OTR participants, a marginally significant increased risk among those in the high load category for HPV38 (Fig. 2B). For those types where load was associated with SCCs, the risk was higher than that observed for simply being positive. For cumulative load, the patterns observed were similar when we stratified by skin color (Supplementary Table). In OTR, the effect seemed to be stronger in people who had been transplanted less than 13 years ago than in those who had been transplanted for longer, but this difference was not significant (Supplementary Table).

Only a small proportion of participants had more than 4 types in the highest load tertile, but Australian and OTR participants in this category were at approximately 3-fold increased risk of SCCs (Table 3). There was a trend of increasing risk with increasing number of types with high load, which was particularly noticeable in OTR.

This case–control study focused on 8 betaPV types for which there is evidence of potential involvement in the development of skin malignancies. In OTR and Australian ICPs, the presence of high cumulative viral load was associated with increased risk of SCCs.

We have previously reported the prevalence of positivity to at least one of 25 betaPV types (11). In all participants, including those with beta-globin input of less than 400 copies/2 μL DNA, the proportion positive ranged from 90% to 94% across the 3 groups. Restricting the population to those with higher cellular input, the proportion positive to 1 of the 25 types ranged from 89% to 94%. As the 8 types selected for viral load analysis are the most frequently occurring, this was only slightly higher than overall positivity to one of these 8 types, which ranged from 83% to 90%.

The overall prevalence was slightly but not significantly higher in cases than in controls in all groups. This was also true for most individual types, but no one type was consistently significantly associated with increased risk in all groups, suggesting that overall betaPV burden may be more important than infection with any specific type. In contrast to our earlier studies of 25 betaPV types, we found a significantly positive association between the number of selected types of betaPV detected and SCC risk in Australia and OTR but again not in Italy.

As observed before in controls (20), the viral loads in hair bulbs of cases spanned 7 orders of magnitude. The median loads for all types were below one viral DNA copy per 10 cells and usually did not differ significantly between cases and controls. The median cumulative load was significantly higher in cases than in controls in Australia and OTR but not in Italy. However, a comparison based on median load may not reflect the risk in the small number of people who have a much higher viral load. Low DNA loads probably reflect the commensal betaPV flora of the skin and have a negligible impact on skin cancer risk. When we grouped the viral loads into tertiles, we found a significantly higher risk of SCCs for those in the high cumulative load tertile both in ICPs from Australia and in immunosuppressed OTR. Australian and OTR participants with more than 4 HPV types in the high load tertile were at about a 3-fold increased risk of SCCs. Unfortunately, the sample size was too small to enable an analysis of the number of types in the highest load tertile independently of the number of types positive, but these results do suggest that the burden of HPV, possibly as a combination of the number of types positive and the load, is associated with risk of SCCs.

In analyses of type-specific viral loads, no particular type stood out as being consistently associated with risk of SCCs. In Australia, being in the highest load tertile for types HPV23 and 36 was significantly associated with risk of SCCs, with borderline associations for HPV5 and 38. In OTR, HPV8 and 38 were significantly associated and HPV20 and 36 were borderline. However, the high rate of co-infection makes it difficult to assess independent effects, and comparative laboratory studies are likely to be more informative. Viral DNA loads within the top decile are comparable with loads in patients with epidermodysplasia verruciformis highly predisposed to SCC development (19). An analysis of the proportion of cases and controls in the top load decile might reveal stronger associations, but the numbers are too small to enable a robust analysis.

We previously reported associations between antibody positivity and SCCs for Australia (6), and for OTR in the presence of concordant DNA positivity (7) but no association for Italy (6). A type-specific evaluation of these data shows significant associations for all 8 types except HPV23 in Australia and for HPV5 and 38 in OTR, with ORs between 1.5 and 2.0. Thus, there is little overlap in the types where associations were seen for antibodies and the types with associations for viral load. This may be due to the fact that it is total HPV burden rather than infection with a specific type that is more important or because a detectable seroresponse is not only dependent on the viral load but also on additional signals such as inflammation.

While the results from Australia and the OTR provide evidence for an association between viral load and cutaneous SCCs, results from Italy were largely null. We are unsure of the reasons for this. The source of controls varied according to centre, with all Italian controls being hospital-based and only a proportion of Australian controls being clinic-based. However, this is unlikely to explain the difference. We compared the overall prevalence and viral load of the Australian clinic-based controls with that of the non–clinic-based controls and found no difference (both P > 0.6). Similarly, excluding the 27 Italian controls who presented specifically with dermatological conditions did not alter the results. While there were differences in the study groups according to sun exposure and skin type, adjustment for these factors made no difference to the results and stratifying by skin type did not reveal significantly different patterns. OTR controls were significantly younger than cases, which may have introduced bias despite adjustment in the analysis. However, we have previously shown no association between age and viral load in OTR, and the OTR results are supported by the Australian ICP results where age matching was successful.

Assuming that the findings from Australia and OTR are not due to chance or bias, the following explanations for the association should be considered. It is possible that skin cancer and high viral load independently have a common etiology, possibly UVR-induced immune suppression. However, the association may also point to an effect of viral load on cellular transformation. This explanation is supported by data from HPV8 transgenic mice. Homozygous HPV8-E2 transgenic animals all spontaneously developed flat or polypous skin tumors during their first year of life in contrast to heterozygous mice (with half the viral DNA load and half viral mRNA levels), where only about 60% developed skin tumors in their second year of life (28).

This is the first time that the association between viral load and risk of cutaneous SCCs has been described. As no one particular type seems to be particularly oncogenic, it may be that total HPV burden is the most important etiologic factor. Our findings lend weight to the hypothesis that HPV plays a role in cellular transformation of human skin.

No potential conflicts of interest were disclosed.

Conception and design: R.E. Neale, S. Weissenborn, D. Abeni, J.N.B. Bavinck, M.C.W. Feltkamp, C. Harwood, I. Nindl, M. Pawlita, W.G. Quint, H. Pfister

Development of methodology: R.E. Neale, S. Weissenborn, A.C. Green, M. de Koning, T. Waterboer, U. Wieland

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.E. Neale, S. Weissenborn, D. Abeni, J.N.B. Bavinck, S. Euvrad, C. Harwood, M. de Koning, L. Naldi, C. Proby, W.G. Quint, T. Waterboer, U. Wieland, H. Pfister

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.E. Neale, S. Weissenborn, D. Abeni, J.N.B. Bavinck, S. Euvrad, A.C. Green, I. Nindl, M. Pawlita, H. Pfister

Writing, review, and/or revision of the manuscript: R.E. Neale, S. Weissenborn, D. Abeni, J.N.B. Bavinck, S. Euvrad, M.C.W. Feltkamp, A.C. Green, C. Harwood, M. de Koning, L. Naldi, I. Nindl, M. Pawlita, C. Proby, W.G. Quint, T. Waterboer, U. Wieland, H. Pfister

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.N.B. Bavinck, M. de Koning, U. Wieland

Study supervision: D. Abeni, J.N.B. Bavinck, L. Naldi, C. Proby, W.G. Quint, H. Pfister

The authors thank J. Lindeman and Labo Bio-Medical Products B.V. for providing the RHA strips and Monika Junk for excellent technical assistance and the members of the EPI-HPV-UV-CA group.

EPI-HPV_UV_CA group members: J.N. Bouwes Bavinck, H.C. Wisgerhof, K.D. Quint, E.I. Plasmeijer.

Department of Dermatology, Leiden University Medical Center: M.C.W. Feltkamp.

Department of Medical Microbiology, Leiden University Medical Center: S. Euvrard, J. Kanitakis.

Department of Dermatology, Hospices Civils de Lyon, Lyon, France: I. Nindl, E. Stockfleth, F. Chatzinasiou.

Department of Dermatology, University Hospital Charite, Skin Cancer Center Charite, Berlin, Germany: L.Naldi, A. Pizzagalli, F. Sassi.

Department of Dermatology and GISED Study Center, Bergamo, Italy: G. Tessari.

Department of Biomedical and Surgical Sciences, Section of Dermatology, University of Verona, Verona, Italy: C.A. Harwood, C.M. Proby, L. Mitchell, K. Purdie.

Centre for Cutaneous Research, Blizard Institute, Barts and the London School of Medicine and Dentistry, Queen Mary University of London, London, United Kingdom: H. Pfister, U. Wieland, S. Weissenborn.

Institute of Virology, University of Cologne: M. Pawlita, T. Waterboer.

German Cancer Research Center (DKFZ), Heidelberg, Germany: W.G.V. Quint, M.N.C. de Koning, J. ter Schegget, B. Kleter, L.J. van Doorn.

DDL Diagnostic Laboratory, Voorburg, the Netherlands: D. Abeni, F. Sampogna, A. Paradisi, S. Simoni, G.P. Petasecca Donati, C. Masini.

Instituto Dermopatico dell'Immacolata, IDI-IRCCS, Rome, Italy: A.C. Green, R.E. Neale.

Queensland Institute of Medical Research, Brisbane, Australia.

This work was funded by an EC grant (QLK2-CT-2002-01179), with the Australian data collection funded by the National Health and Medical Research Council (Australia). M.C.W. Feltkamp was supported by a Clinical Fellowship from the Netherlands Organization for Health Research and Development (grant 907-00-150). R.E. Neale is funded by an NHMRC (Australia) Career Development Award. T. Waterboer was supported by the Peter und Traudl Engelhorn-Stiftung zur F€orderung der Biotechnologie und Gentechnik.