Type-specific associations of human papillomavirus load with risk of developing cervical carcinoma in situ
We have previously shown that high human papillomavirus (HPV) 16 load in Papanicolaou smears negative for dysplasia is strongly associated with risk for carcinoma in situ (CIS) of the cervix. Here we study the amount of HPV DNA for some of the most frequent high-risk HPV types as determinants of progression to cervical CIS. Real-time PCR is used to estimate the normalized viral load of HPV 16, 18, 31, 33, 35, 39, 45, 52, 58 and 67 in 457 cases of cervical CIS and 552 matched population controls. A total of 2,747 archival Pap smears from gynecologic health examinations, collected over a period of up to 26 years, were analyzed to assess viral load during the infection history. Cervical smear samples differ widely in amount of DNA, underscoring the need for normalization of HPV load to number of cells in the sample. The risk of developing cervical CIS increases with higher viral load for most of the HPV types studied. The range of copy numbers per cell does not differ between HPV types but the odds ratio for CIS in the percentile with highest viral load is substantially higher for HPV 16 (OR = 36.9; 95% CI = 8.9–153.2) than for HPV 31 (OR = 3.2; 95% CI = 1.1–9.1) or HPV 18/45 (OR = 2.6; 95% CI = 1.0–6.4). Therefore, HPV viral load may be predictive of future risk of cervical CIS at a stage when smears are negative for squamous abnormalities, but differences between HPV types need closer attention. © 2004 Wiley-Liss, Inc.
Infections by certain types of human papillomavirus (HPV) increase the risk for cervical cancer.1, 2, 3 However, only a small fraction of those testing positive for high-risk HPV types at screening develop cervical cancer. In the 1990s, HPV load was discussed as a possible means for discriminating infections with underlying cervical disease.4, 5, 6, 7, 8, 9, 10, 11 These early studies used semiquantitative approaches to estimate viral load. Several studies have used the Hybrid Capture 2 (HC2) assay, which detects 13 HPV types, when estimating viral load. While some found that viral load increases with disease severity,12, 13, 14, 15 others did not and judged HC2 viral load to lack the specificity required for use in screening.16, 17, 18, 19 Investigators using more quantitative techniques for estimating HPV load have shown a positive correlation between viral load and prevalent grade of cervical disease20, 21 as well as risk of incident disease.22, 23, 24, 25 However, Schlecht et al.25 found that the prediction of high-grade intraepithelial lesions (HSILs) is weaker than the one for low-grade intraepithelial lesions (LSILs).
Viral load estimates based on summary measurements of HPV, such as those generated by the HC2 assay, may be blurred by not distinguishing HPV types.16, 17 Several studies have noted differences in viral load pattern between HPV 16 and other HPV types.20, 21 Gravitt et al.26 showed an increased odds ratio (OR) of prevalent HSIL/cancer for HPV 16 load, but did not find a similar trend for HPV 18. These observations led us to investigate the risk of cervical disease associated with the individual high-risk HPV types.
To study the relationship between HPV load and development of cervical squamous cell carcinoma in situ (roughly corresponding to cervical intraepithelial neoplasia 3 [CIN3]) for a range of HPV types commonly found in cervical tumors, we analyzed a large material of archival cervical smears collected during routine gynecologic health controls and employed our newly developed assay for estimating the normalized HPV load.27 A main asset in this study is the retrospective design that makes it possible to examine the infection history of women over a period of up to 26 years.
MATERIAL AND METHODS
The cohort has been described previously.22, 23, 28 Women included in the case-control study were selected from a cohort comprising women resident in Uppsala County, Sweden, between 1969 and 1995. Eligible cases of cervical carcinoma in situ were identified by merging information from the organized screening program between 1969 and 1995 with the National Cancer Registry. As entry criteria, the woman had to be born in Sweden, be less than 50 years of age at entry and have their first smear classified as normal by cytology on squamous cell (PAP = 1). For each case, 5 separate controls, matched on date of entry to the cohort (± 90 days) and year of birth, were randomly selected from the Uppsala County cohort. The women included as controls had to be alive without having developed cervical carcinoma in situ or invasive cancer prior to the diagnosis date of their corresponding matched case. Similar to the cases, controls were also obliged to have their first smear classified as negative for squamous cell abnormalities. To verify the diagnosis of squamous cell carcinoma in situ, an experienced pathologist (Professor Jan Pontén) reviewed histologic samples from cases. First smears were reviewed by a cytotechnician blinded to case-control status. The study was performed with approval from local ethics committees.
DNA was purified from archival Papanicolaou-stained smears using published procedures.29 Amount of HPV DNA was estimated using real-time PCR.30, 31, 32 The assay employed yields individual estimates for the following types: HPV 16, 31, 35 and 39.27 In addition, estimates are obtained for the groups of viral types: HPV 18/45 and HPV 33/52/58/67. In the case of these groups, the viral burden can be due to either a single or a combination of viral types within the group. In total, up to 6 different signals may be obtained from a single sample. This design was chosen in order to be able to screen as many HPV types as possible, while still using a small number of parallel reactions.27 PCR is performed in a 25 μL volume containing 1 × Buffer A (Applied Biosystems, Foster City, CA), 3.5 mM MgCl2, 200 nM each of dATP, dCTP, dGTP and 400 nM dUTP (Pharmacia Biotech, Uppsala, Sweden), 0.625 U AmpliTaq Gold (Applied Biosystems), 3.1 μg BSA (Sigma, St. Louis, MO) and 200 nM of each primer and probe.27 Each sample is analyzed in 3 different PCR reactions. Reaction 1 quantifies HPV types 16, 31, 18 and 45 (HPV 18 and 45 are indistinguishable from each other); reaction 2 quantifies HPV types 33, 35, 39, 52, 58 and 67 (HPV 33, 52, 58 and 67 are indistinguishable from each other); reaction 3 quantifies the amount of the human nuclear gene hydroxymethylbilane synthase (HMBS). HMBS is located at 11q23.3 and encodes 2 forms of enzymes involved in heme production. To our knowledge, no polymorphisms have been reported in the amplified sequence. Amplification and detection is performed using a 7700 Sequence Detection System (Applied Biosystems). All PCR products are less than 200 base pairs of length to enable amplification of the partially degraded DNA from archival smears.33 The amplification ramp includes an initial hold program of 10 min at 95°C followed by a 2-step cycle consisting of 15 sec at 95°C and 1 min at 57°C, repeated 50 times. No template controls, i.e., reactions where the clinical sample is substituted with water, were included in each real-time PCR run to screen for possible contamination. Thresholds for positivity of 10 HPV copies per PCR and 10 copies of the human gene per PCR for the human gene assay are used to reduce risk of false positives due to contamination and avoid the stochastic variation associated with low amounts of target DNA in reactions.
Odds ratios are estimated with 95% confidence intervals using logistic regression (Logistic procedure, SAS software package version 6.12; SAS Institute, Cary, NC). Trends were examined using double-sided Cochran-Armitage test (Freq procedure, SAS version 6.12) and t-tests were performed allowing for unequal variances when comparing mean viral load in HPV-positive smears (Ttest procedure, SAS version 6.12).
The nested case-control study initially included 504 cases and 662 controls. As some controls only had a single available smear during the follow-up period, a second control was therefore randomly chosen from the original pool of controls to balance the number of smears between cases and controls. In total, 158 additional controls were included. After subsequent cytologic and histologic review, the study included 499 cases and 657 controls. Controls that had undergone hysterectomy prior to the diagnosis of their corresponding case were then excluded, leaving 495 cases and 649 controls. We then removed smears taken later than 1 year prior to diagnosis, since additional smears were often taken close to diagnosis among the cases. This exclusion left 471 cases with a total of 1,656 smears and 605 controls with 1,654 smears.
Due to the poor DNA quality of archival Pap smears, some samples contained very low amounts or were negative for human DNA. These smears were considered uninformative as they would not allow for normalization and were consequently removed from the study. After elimination of smears below the threshold of 10 human genome equivalents per PCR reaction, the study includes 457 cases with cervical carcinoma in situ (1,342 smears) and 552 controls (1,405 smears), each having at least 1 smear positive for human DNA (Table I). The mean number of smears per woman is fairly similar between cases (3.1) and controls (2.4), whereas the medians are equal with a somewhat wider range for the cases (Table I).
Table I. Characteristics of the Case-Control Material
|Year of smear collection|| || |
| Earliest smear||1969||1969|
| Last collected||1993||1993|
|Human-DNA|| || |
| Number of positive smears||1342||1405|
| Median number of smears per woman (range)||2 (1–21)||2 (1–13)|
| Human genome equivalents/reaction|| || |
|HPV 16|| || |
| Number of positive smears||653||86|
| Number of women with no positive smears||206||490|
| Number of women with ≥ 2 positive smears||154||11|
| HPV 16 genomes/human genome equivalent|| || |
|HPV 31|| || |
| Number of positive smears||147||47|
| Number of women with no positive smears||389||514|
| Number of women with ≥ 2 positive smears||33||7|
| HPV 31 genomes/human genome equivalent|| || |
|HPV 18/453|| || |
| Number of positive smears||171||85|
| Number of women with no positive smears||357||490|
| Number of women with > 2 positive smears||32||14|
| HPV 18/45 genomes/human genome equivalent|| || |
|HPV 33 group4|| || |
| Number of positive smears||159||42|
| Number of women with no positive smears||368||517|
| Number of women with ≥ 2 positive smears||33||7|
| HPV 33 goup genomes/human genome equivalent|| || |
|HPV 35|| || |
| Number of positive smears||16||8|
| Number of women with no positive smears||445||545|
| Number of women with ≥ 2 positive smears||3||1|
| HPV 35 genomes/human genome equivalent|| || |
|HPV 39|| || |
| Number of positive smears||12||5|
| Number of women with no positive smears||447||548|
| Number of women with ≥ 2 positive smears||2||1|
| HPV 39 genomes/human genome equivalent|| || |
DNA amount in samples
Obtained amounts of human DNA per reaction are similar to that expected from the yield of other extraction methods.34 The estimated number of human genome equivalents (hge) varies substantially between samples (Table I). The median number is slightly higher in the cases (170.2 hge) than in the controls (130.8 hge).
HPV amount and cancer risk
Two strategies were used in selecting smears. The first smear classified as normal by cytology was collected from each woman to represent baseline HPV content in relation to subsequent tumor development. To increase the chance of including HPV infections occurring at a later stage, a mean HPV load estimate was also calculated based on all HPV-positive smears from each woman.
A total of 653 smears (49%) from cases and 86 smears (6%) from controls type positive for HPV 16 (Table I). The median and maximum HPV 16 copy number per human genome equivalent are higher for cases than for the controls (Table I). The mean HPV 16 load is higher in the case smears (91.0 copies/hge) versus control smears (6.0 copies/hge; p = 0.0001, t-test). Cases and controls differ somewhat in the number of smears per woman (Table I), but when introduced in the logistic regression model, this variable shows only minor effects on the OR. Therefore, the ORs are presented unadjusted. When stratifying women in percentiles according to mean HPV 16 copy number per human genome equivalent, the ORs estimated by logistic regression increase sharply with higher viral loads and reaches OR = 185 for the percentile with the highest HPV viral load (Table II). The OR is significant for all percentiles and the trend of increasing ratio of cases to controls with higher viral load is significant (p = 0.001 for trend, Cochran-Armitage).
Table II. Odds Ratios for Cervical Carcinoma in situ in Relation to the Mean Normalized Viral Load of Individual HPV Types or Groups of Closely Related HPV Types
|HPV 16|| || || || |
| 0 < mean viral load3 ≤ 1.2||46/33||3.3||2.1–5.3||0.0001|
| 1.2 < mean viral load3 ≤ 9.4||58/20||6.9||4.0–11.8||0.0001|
| 9.4 < mean viral load3 ≤ 70.1||69/8||20.5||9.7–43.4||0.0001|
| Mean viral load3 > 70.1||78/1||185.5||25.6–999||0.0001|
|HPV 31|| || || || |
| 0 < mean viral load3 ≤ 0.7||14/12||1.5||0.7–3.4||0.2782|
| 0.7 < mean viral load3 ≤ 4.9||16/11||1.9||0.9–4.2||0.1002|
| 4.9 < mean viral load3 ≤ 35.0||19/8||3.1||1.4–7.2||0.0074|
| Mean viral load3 > 35.0||19/7||3.6||1.5–8.6||0.0043|
|HPV 18/454|| || || || |
| 0 < mean viral load3 ≤0.7||18/23||1.1||0.6–2.0||0.8243|
| 0.7 < mean viral load3 ≤ 2.3||22/18||1.7||0.9–3.2||0.1118|
| 2.3 < mean viral load3 ≤ 13.8||30/10||4.1||2.0–8.5||0.0001|
| Mean viral load3 > 13.8||30/11||3.7||1.9–7.6||0.0002|
|HPV 33 group5|| || || || |
| 0 < mean viral load3 ≤ 3.0||25/6||5.9||2.4–14.4||0.0001|
| 3.0 < mean viral load3 ≤24.2||24/7||4.8||2.1–11.3||0.0003|
| 24.2 < mean viral load3 ≤202.1||26/5||7.3||2.8–19.2||0.0001|
| Mean viral load3 > 202.1||14/17||1.2||0.6–2.4||0.6914|
We then included only the first smear from each woman in the logistic regression analysis. These smears were collected on average 9 years before diagnosis and are classified as normal by cytology on squamous cells. The OR based on HPV 16 copy number per human genome equivalent in the first smears is significant for all percentiles and the ratio of cases to controls increases with viral load (p = 0.001, Cochran-Armitage; Table III). Since the controls are matched to cases with respect to time of first smear, the results cannot be due to asymmetric sampling of cases and controls. In the analysis of the first smear, we initially adjusted for age at sample collection and time span between the sampling and diagnosis of cervical carcinoma in situ. None of these adjustments affected the OR substantially and the values used are therefore from the unadjusted analysis.
Table III. Odds Ratios for Cervical Carcinoma in situ in Relation to the Normalized Viral Load in the First Smear (Negative for Malignant Cells) from Each Individual
|HPV 16|| || || || |
| 0 < viral load3 ≤ 0.5||30/15||3.4||1.8–6.3||0.0002|
| 0.5 < viral load3 ≤ 2.6||37/9||6.9||3.3–14.5||0.0001|
| 2.6 < viral load3 ≤ 19||37/7||8.9||3.9–20.1||0.0001|
| Viral loadC > 19||44/2||36.9||8.9–153.2||0.0001|
|HPV 31|| || || || |
| 0 < viral load3 ≤ 0.4||10/8||1.5||0.6–3.9||0.3678|
| 0.4 < viral load3 ≤ 2.8||7/9||1||0.4–2.6||0.9331|
| 2.8 < viral load3 ≤ 24||14/3||5.7||1.6–20||0.0063|
| Viral loadC > 24||13/5||3.2||1.1–9.1||0.0283|
|HPV 18/454|| || || || |
| 0 < viral load3 ≤ 0.4||10/12||1||0.4–2.4||0.9835|
| 0.4 < viral load3 ≤ 2.0||10/12||1||0.4–2.4||0.9835|
| 2.0 < viral load3 ≤ 19||14/8||2.1||0.9–5.1||0.0943|
| Viral loadC > 19||15/7||2.6||1.0–6.4||0.0395|
|HPV 33 group5|| || || || |
| 0 < viral load3 ≤ 2.2||7/2||4.2||0.9–20||0.0759|
| 2.2 < viral load3 ≤ 22||7/3||2.8||0.7–11||0.1402|
| 22 < viral load3 ≤ 296||6/3||2.4||0.6–9.6||0.2214|
| Viral loadC ≥ 296||3/7||0.5||0.1–2.0||0.3327|
Other HPV types
A total of 147 smears (10%) from cases and 47 smears (4%) from controls typed positive for HPV 31 (Table I). The median and maximum HPV 31 copy number per human genome equivalent is higher for cases than for controls (Table I). The mean HPV 31 load does not differ between case (44.5 copies/hge) and control smears (21.3 copies/hge; p = 0.1637, t-test). The OR for mean HPV 31 viral load is significantly elevated in the 2 highest percentiles (Table II), but no continuous trend in the ratio of cases to controls with increasing viral load is seen. In the first smears, the HPV 31 viral load is not significantly different between cases and controls. The OR for the first smear is significant for the 2 highest percentiles, but the ratio of cases to controls shows no trend with increasing viral load (Table III).
The typing system does not distinguish between HPV 18 and HPV 45. A total of 171 smears (12%) from cases and 85 smears (6%) from controls are positive for HPV 18/45 (Table I). The median and maximum HPV 18/45 copy number per human genome equivalent is higher for cases than for controls (Table I), but the mean HPV 18/45 load does not differ between case (127.1 copies/hge) and control smears (36.0 copies/hge; p = 0.0770, t-test). The OR for mean HPV 18/45 copies/hge is significantly elevated for the higher percentiles (Table II) and the ratio of cases to controls increases with viral load (p = 0.002, Cochran-Armitage). In the analysis of the first smear, the OR is significant for the highest percentile, but the ratio of cases to controls does not show a continuous increase with viral load (Table III). Also, for the first smear, we do not see significant differences in viral load between HPV 18/45-positive cases and controls.
For HPV 35 (in total 24 positive smears), the median copy number per human genome equivalent is slightly higher in the cases, while the maximum copy number is higher in the controls (Table I). For HPV 39 (in total 17 positive smears), the median copy number per cell is higher in the controls and maximum copy number is higher in the cases (Table I). However, the frequency of these 2 HPV types is too low to examine in detail the relationship between viral load and risk of cervical carcinoma in situ.
The typing system combines the viral load of HPV 33, 52, 58 and 67 into one estimate. A total of 159 smears (11%) from cases and 42 smears (3%) from controls are positive for the HPV 33 group (Table I). The maximum copy number per human genome equivalent for this group is higher than for the other HPV types. The median and maximum copy numbers per human genome equivalent are higher in the controls, but the mean viral load does not differ between case (908.9 copies/hge) and control smears (182.0 copies/hge; p = 0.0770, t-test). The OR for mean HPV 33 group copy number per human genome equivalent is statistically significant for 3 of the percentiles (Table II), but not for the analysis of the first smear (Table III).
We examined the relationship between viral load and risk of developing cervical carcinoma in situ for a series of high-risk HPV types. The number of viral copies per sample in our study for the most frequent HPV types is similar to estimates reported by other studies.20, 21, 24, 25 Our results show that the risk of cervical carcinoma in situ increases with higher amounts of HPV 16 per nuclear genome equivalent, both when based on samples taken at a time when the cervical smears are normal by cytology and when based on the mean viral load for a series of smears selected at random from those available in the cytology registry from a woman. This is consistent with our previous results22, 23 as well as those of others.24 For both HPV 31 and HPV 18/45, the viral load in cervical smears is higher in the cases than in the controls, although the difference is not statistically significant. The ORs reached in the percentile with highest viral load are significant for HPV 31 and HPV 18/45, but are low compared to the corresponding HPV 16 ORs. Thus, these HPV types do not show strong positive relationships between viral load and risk for cervical carcinoma in situ. Given the comparatively low OR for HPV 18/45, it is not surprising that other studies with a smaller sample size have been unable to find a relationship between viral load and risk of cancer development.26 The combined viral load for the group of HPV 33, 52, 58 and 67 has a different relationship between viral load and risk of cervical carcinoma in situ from that of the other HPV types studied. Thus, while HPV 16 load shows a clear relationship with cancer risk, the other HPV types studied appear to have weaker relationships between viral load and risk of cervical carcinoma in situ, in line with the findings of studies based on a different design.26 These differences between viral types may be a cohort effect or may reflect biologic properties of the individual viral types. For instance, HPV 16 has been shown to be able to induce malignant transformation without integrating into the cell genome in contrast to HPV 18, HPV 31 and HPV 35, which always seem to be present in an integrated physical state in malignant lesions.35, 36, 37
Given the difference in risk trends between HPV types, studies using methods based on summary measurements of viral load without identification of the underlying HPV type are likely to generate erroneous risk estimates. To assess the effect of using a typing system that does not identify individual HPV types, we estimated the OR for total HPV load from our data by summarizing the copy numbers for the individual HPV types or groups of types. The OR for total HPV load is markedly lower than when based on the type-specific estimate for HPV 16, in particular for the higher viral load percentiles. For example, when only using the first smear in the analysis, a type-specific assay gives an OR = 36.9 for the highest percentile of HPV 16-infected women, while a measurement of total HPV load only would yield an OR = 4.9 for the highest percentile. Thus, in studies of the relationship between HPV load and cytologic or histologic outcomes, valuable information may be gained by applying a method that is able to distinguish the viral load of some of the most prevalent high-risk types, and minimally to distinguish the HPV 16 load from that of other high-risk types.
Our study was designed to address the relationship between viral load and risk of cancer in situ development. Nevertheless, the clinical utility of HPV viral load has been discussed.15, 24 Presently in cervical cancer screening, qualitative HPV typing is recommended for certain outcomes,38 but it has been suggested that HPV load may be used to identify clinically relevant infections.15, 24 Inclusion of an HPV load estimate in clinical screening has also been proposed to reduce referral rates for women with atypical squamous cells of undetermined significance (ASCUS) smears, although at a questionable expense of sensitivity.19 However, a number of issues remain to be addressed in order to determine the clinical value of an HPV load assay. HPV load has been estimated using a variety of laboratory methods with very different dynamic range. Given the wide variation in number of HPV genomes seen between clinical samples, a method with an appropriate dynamic range needs to be chosen. Clinical samples, even those taken using very standardized routine procedures such as the Papanicolaou smear, contain very different amounts of cellular material. We have used normalization of HPV copy number to the number of a human nuclear gene as a feasible means of obtaining comparable estimates between samples and studies. However, good correlation has been reported between HPV 16 copy numbers with and without such normalization for total cellular DNA.26 Consistent with this, we have found similar ORs of cancer in situ for normalized and original HPV 16 copy numbers (data not shown). However, the human nuclear gene assay provides additional information that justifies its inclusion in the analysis, such as an indication of PCR inhibition or the presence of very limited number of cells. We used a threshold of 10 nuclear gene copies per reaction for inclusion in the present analysis in order to reduce the stochastic variation expected at very low starting copy numbers in the PCR and to minimize the effect of potential crosscontamination between samples. However, normalization does not adjust for underlying heterogeneity of the cellular population sampled and this may contribute to variation in the estimated load for individual woman.39 Such heterogeneity may result either from intrinsic characteristics of the epithelium sampled or from heterogeneity in the severity and extent of lesions.39 Lesions often appear to contain simultaneously mixed CIN grades and high viral load can consequently be associated with widespread areas of CIN1 rather than with more severe dysplasia as CIN3 or cancer per se.39 Such heterogeneity clearly illustrates the complex nature of the cervical epithelial during the transformation process and complicates the use of HPV load as a means of confirming a histologic diagnosis. Nonetheless, HPV viral load estimates at an earlier stage of the transformation process may still provide information on risk of future progression of an infection, as indicated by our data. For samples at the early stages, it would be of value to study the extent of short-term variation in viral load for individual women. Such variation may either be caused by sampling variation (more of less representative sampling of the epithelial cell population) or actual changes in viral load, and the extent and cause of this variation will affect the predictive value of an individual test. The archival samples used on our present study were derived from regular screening and therefore taken at time intervals too long to be suitable for such a study. Given the need for additional biomarkers that distinguishes transient infections from those that impose a risk of progressing into cervical cancer, it appears appropriate to evaluate further the potential of using HPV load as such a variable.