The current status of cell-free human papillomavirus DNA as a biomarker in cervical cancer and other HPV-associated tumors: A review

Tumor cells release fragments of their DNA into the circulation, so called cell-free tumor DNA (ctDNA), allowing for analysis of tumor DNA in a simple blood test, that is, liquid biopsy. Cervical cancer is one of the most common malignancies among women worldwide and high-risk human papillomavirus (HR-HPV) is the cause of the majority of cases. HR-HPV integrates into the host genome and is often present in multiple copies per cell and should thus also be released as ctDNA. Such ctHPV DNA is therefore a possible biomarker in cervical cancer. In this review, we first give a background on ctDNA in general and then a comprehensive review of studies on ctHPV DNA in cervical cancer and pre-malignant lesions that may develop in cervical cancer. Furthermore, studies on ctHPV DNA in other HPV related malignancies (eg, head-and-neck and anogenital cancers) are briefly reviewed. We conclude that detection of ctHPV DNA in plasma from patients with cervical cancer is feasible, although optimized protocols and ultra-sensitive techniques are required for sufficient sensitivity. Results from retrospective studies in both cervical cancer and other HPV-related malignancies suggests that ctHPV DNA is a promising prognostic biomarker, for example, for detecting relapses early. This paves the way for larger, preferably prospective studies investigating the clinical value of ctHPV DNA as a biomarker in cervical cancer. However, there are conflicting results whether ctHPV DNA can be found in blood from patients with pre-malignant lesions and further studies are needed to fully elucidate this question.


| INTRODUCTION
Cervical cancer (CC) is the fourth most common malignancy in both incidence and mortality among females worldwide, with approximately 600 000 new cases and 350 000 deaths in 2020. 90% of affected women are in low-and middle-income countries, in part due to effective screening for precursor lesions in high-income countries. 1 The most common histological type is squamous cell carcinoma (SCC) followed by adenocarcinoma. 2,3 The worldwide prevalence of human papillomavirus (HPV) in CC is 88% in SCC and 77% in adenocarcinoma. 4 The strongest prognostic factor in CC is disease stage and presence of metastases in the lymph nodes at the time of diagnosis. 5,6 Tumor size is also an independent prognostic factor. 7,8 Early stage disease (stages ≤ IB2) is generally treated surgically and has a 2-year risk of recurrence of 8% to 10%, while advanced stages are treated with chemoradiotherapy and have a 30% to 50% risk of relapse within 2 years. 9 Since local control is high, most recurrences occur at regional or distant sites with dismal prognosis.
For women that have been treated for invasive cervical cancer, routine clinical examination is performed regularly with complementary radiology such as magnetic resonance imaging (MRI) or computed tomography (CT) if needed. However, no reliable surveillance markers are currently available to predict which patients are more likely to recur after treatment for cervical cancer. Circulating cell-free tumor HPV DNA (ctHPV DNA) is one possible such biomarker that holds promise and studies investigating ctHPV DNA in cervical-and other HPV-related types of cancer are reviewed below, after an introduction to cell-free DNA.

| Introduction to cell-free DNA
Liquid biopsy is a potential cancer biomarker that has the advantage of being relatively noninvasive and can potentially represent all malignant clones. The term liquid biopsy comprises all types of circulating genetic entities: circulating tumor cells; cell-free DNA (cfDNA) and exosomes (ie, small vesicles carrying RNA and DNA) as well as circulating micro-RNA, however, this review will focus on cfDNA which is easy to isolate and thus the most studied in cervical cancer (CC).
Although the phenomenon was first described in cancer patients in the 1970s, 10 it has only become possible to perform detailed studies of liquid biopsies in cancer since the development of modern methods such as massive parallel sequencing with error suppression and ultra-sensitive targeted methods (eg, digital PCR). Studies have shown that tumor-derived cfDNA (ctDNA) can predict cancer outcome 11 ; monitor residual disease and relapse 12 and detect putative targeted therapies. 13,14 In addition, it can be used to diagnose cancers that are hard to biopsy 15 and possibly in selected situations for population screening. 16 Thus, cfDNA has the potential to be a versatile cancer biomarker for many cancer types and at diverse timepoints during cancer development, although further research is needed. The first FDA-approved cfDNA test was for EGFR mutations in plasma from patients with non-small cell lung cancer in June 2016. 17 Recently, FDA has approved two comprehensive gene panel tests for cfDNA analysis in solid advanced tumors (55 genes and > 300 genes, respectively) (www.fda.gov), constituting the first step towards clinical implementation of cfDNA analysis. into body fluids such as plasma ( Figure 1). cfDNA is later degraded via enzymes in the bloodstream or metabolized in the liver and possibly in the kidney. Larger fragments of cfDNA are released by necrosis. In addition, cells, especially immune cells, perform active export of cfDNA (and RNA) inside vesicles such as exosomes and it is probable that exosome cfDNA constitutes the majority of the total cfDNA. 18 cfDNA carries all the genetic changes found in the mother cell, which in the case of cancer includes any viral DNA that acts as a driver as well as all other single nucleotide variants, structural variants and epigenetic changes. 19 In addition, the cfDNA fragment pattern is celltype specific which can be used to identify the source of the cfDNA. 20 Another advantage is that cfDNA reflects the current genetic status of the cancer cells as the half-life is <2.5 h. 21 Three ml blood yields approximately 1 ml of plasma with on average 6 ng cfDNA. This theoretically corresponds to up to 2000 copies of total cfDNA per ml plasma. In serum, the total levels are higher due to white blood cell lysis during the clotting process, therefore plasma extracted directly after sampling or from tubes that stabilize the leukocytes is recommended. 19 85% of the cfDNA derives from blood cells, 10% from vascular endothelial cells, 1% from the liver cells, with the remaining derived from other cells including cancer cells. 22 Total levels of cfDNA can also vary between individuals and at different time points due to factors such as age, inflammation, physical exercise, treatment such as surgery or cytotoxic drugs as well as nuclease activity, and liver and renal function which affect degradation. 23,24 In cancer, levels of cfDNA increase to on average 29 ng/ml plasma in stages I-III, 25 and up to 1000 ng/ml in metastatic cancer. 26 Tumor-derived cfDNA identified by genetic aberrations (ctDNA) can be detected across a wide range of cancer types, with higher levels generally found in colorectal, small cell lung, bladder and liver cancer and lower levels in brain tumors. 14 In addition, the amount of ctDNA is often affected by the tumor size and metabolic tumor burden. [27][28][29] Some studies have also shown that ctDNA levels correlate to liver metastasis; and to a lesser extent to lymph node and peritoneal metastasis. [30][31][32] However, the kinetics of cfDNA are still not fully understood and even in advanced and/or metastasized cancer (stage III/IV), half of the cases have a ctDNA fraction of <1% to 1.5% and the median variant allele frequency detected in cfDNA assays was <4 mutant alleles/1000 total cfDNA copies. 14 Thus, ultra-sensitive detection techniques are required to find such a small number of mutant molecules. An additional way of increasing the sensitivity of a cfDNA assay is to target a sequence that is present in multiple copies in the cancer cells. For instance, amplifications (defined as > five copies) of specific genes are much easier to detect than single mutations or duplications. 33

| HPV infection and precursor lesions
Human papillomavirus (HPV) is a sexually transmitted virus with double-stranded DNA that infects the basal layer of cells in the cervix and often exists as 50 to 100 copies of HPV DNA in episomes per cancer cell. In some women, high level expression of the genes that drive viral replication is acquired through mutation and often integration into the cell genome, perturbing the normal differentiation process and creating immortal cells that drive malignancy. 34 E6 and E7 are the most oncogenic of the HPV-genes and HR-HPV E6 and E7 bind to p53 and pRb, respectively, leading to significantly lower levels of these tumor suppressor proteins in affected cells. E6 and E7 are also the targets in the vast majority of studies reviewed below, with There are hundreds of different HPV strains: the HR-HPV have been strongly associated with the risk of developing CC. 35 The most common HR-HPV types are 16 and 18 that are found in 70% of all CC 35 with an additional 10 to 13 high-risk types identified, with a few more listed as potentially carcinogenic. 4,36,37 HR-HPV is a major cancer initiator, but is neither necessary nor sufficient to trigger cancer development. Over 90% of women who get infected with HR-HPV do not develop CC, but clear the infection spontaneously within 2 years. 38 For those with persistent infection the risk of developing HSIL or cancer is increased up to 22%. 39 Primary prevention is provided by vaccination with HPV vaccines before sexual debut, reducing the risk for invasive cervical cancer by up to 88%, especially in women vaccinated before the age of 17 years. 40

| ctHPV DNA detection in cervical carcinomas
Since HPV is the cause of the majority of cervical carcinomas and exists in multiple copies per cell, it is an attractive potential biomarker.
Thus, it is no surprise that the majority of studies investigating cfDNA in cervical cancer involve the detection of ctHPV DNA, Table 1, with most studies focusing on HPV16/18. E6, E7 and L1 are specific for F I G U R E 1 Top left: A vascularized tumor with the green tumor cells in a normal stroma (blue cells). Top right: When cells die, they shed small fragments of cell-free DNA (cfDNA) into the bloodstream. The majority of these fragments derive from the blood cells (leukocytes, shown in grey). A small part derives from the liver cells (black) and a minority originates from the tumor cells (white). These cancer-derived cfDNA fragments will contain all the genetic aberrations that are present in the cancer. In the case of cancer caused by HPV, this means the the cfDNA will contain specific viral (onco)genes and can be denoted ctHPV DNA. Bottom left: When plasma is collected (yellow liquid in blood tube) and separated from the leukocytes in the buffy coat and the red blood cells, cfDNA can be extracted. 96% of this cfDNA is from the leukocytes (grey), 1% to 2% from the liver cells (black) and a small proportion derives from the cancer cells (ctHPV DNA, green). Bottom right: The cfDNA is analyzed using, for example, droplet digital PCR and the total number of copies of cfDNA can be measured using primers and probes for a reference gene. This number varies greatly between cancer patients (1000-40 000 copies/blood tube). In addition, the ctHPV DNA can be measured using specific primers and probes for an HPV-specific gene such as E6 or E7. The number of ctHPV copies per blood tube also varies greatly (from 0 to >100 000 copies). HPV, therefore by using primers and probes that specifically bind to these sequences, it is possible to design a very specific and sensitive targeted quantitative PCR-assay. Detection methods have varied from conventional PCR (available from 1985), to quantitative PCR (qPCR) (available from 1993) to in recent years droplet digital PCR (ddPCR) (first described 2011). A benefit of using ddPCR, compared with for example, next generation sequencing methods, is its capability to perform absolute quantification of ctHPV DNA copy numbers in serum or plasma, despite high background genomic DNA. Furthermore, different starting material-blood, serum or plasma, and a variety of different DNA extraction methods have been used. The cost of ddPCR varies depending on approach and supplier but we estimate that it today cost approximately 45 USD per sample if samples are analyzed in triplicates on full 96 well plates, with qPCR costing roughly the same.
The PCR methods used at the time were likely not sensitive enough for the detection of the miniscule amount of ctHPV DNA. The primers used targeted products longer than the average size of the ctDNA molecules, and today a product size of <166 bp is recommended. Furthermore, sample handling was not always optimal for ctDNA, for example, whole blood or serum used instead of plasma, delay in plasma extraction, storage at À20 C instead of À80 C. Finally, in many of the above studies, very small amounts of starting material, often only 200 to 400 μl plasma or serum, were used. Given the scarcity of ctDNA, a larger starting volume would have been desirable.
However, with the advent of ddPCR, a renewed interest in ctHPV DNA was seen. In 2016, Jeannot et al published the first study using ddPCR for detection of ctHPV DNA in plasma. In 47 patients with verified HPV16/18-positive cervical cancer, they found ctHPV DNA in pretreatment plasma of 83% of the cases using ddPCR, compared with in 69% of the cases when using qPCR. 51 This was followed by another three studies 52-54 using ddPCR, finding ctHPV DNA in 63% to 83% of pre-treatment plasma from patients with cervical carcinoma. Furthermore, Kang et al, detected ctHPV DNA in serum of 19/19 (100%) patients with HPV-positive metastatic cervical cancer. 55 Recently another two studies, stratifying between early-and advanced cervical cancer, found ctHPV DNA in the majority of advanced cases (94% and 63%, respectively), but in a minority of early-stage CC (26.7% and 10%, respectively). 56,57 T A B L E 1 Studies investigating detection of ctHPV DNA in pre-treatment liquid biopsies from patients with invasive cervical cancer cfHPV. If this is true, cfHPV would more likely be detected in younger women from regions with high prevalence of HPV. There are several studies from various parts of the world including woman without cervical pathology and most report 0% to 2% detection rate of cfHPV DNA using PCR, qPCR or ddPCR in healthy women (Table 3)  within 3 months after finished treatment suffered recurrences, involving distant metastasis in 7/8 cases. 42 Table 4 for studies on ctHPV DNA in head-and neck cancer and Table 5 for anal cancer. In addition, ctHPV DNA has shown promise as a prognostic biomarker in some of these studies.  67,68 In anal cancer however, it was noted that pre-treatment ctHPV DNA levels did not correlate to patient characteristics. 69  this. 68,72 In anal cancer it has been reported that ctHPV DNA detection after treatment was correlated to shorter disease-free survival, 70 progression-free survival and a reduction of 1-year overall survival rate. 69 Furthermore, ctHPV DNA has been used to measure response to chemoradiotherapy and surgery in OPSCC and anal cancer. 73  several studies have demonstrated that persistent end-of-treatment ctHPV DNA seems to be a negative prognostic factor in cervical cancer 54,57 as well as in other HPV-related malignancies. 70

| CONCLUSIONS
We conclude that detection of ctHPV DNA in plasma from patients with HPV-related cancer is feasible. However, very sensitive methods (eg, ddPCR) and optimized sample handling are necessary to achieve appropriate sensitivity. Importantly, recent studies in both cervicaland head and neck cancer imply that post-treatment ctHPV DNA is a promising biomarker for predicting outcome. Perhaps most importantly, ctHPV DNA is a very intriguing biomarker for surveillance and early detection of relapse. This warrants larger studies to be performed, preferably prospectively, also including more HR-HPV-types than HPV16 and 18, that hopefully will elucidate the most appropriate clinical setting for the use of ctHPV DNA. More studies using ultrasensitive techniques are also needed to fully understand how ctHPV reported in the paper has been performed by the authors, unless clearly specified in the text.

ACKNOWLEDGEMENT
Graphical abstract made in Mind the graph, Cactus Communications.