Introduction

The International Agency for Research on Cancer has classified >16% of all cancers in humans as caused by infections [1]. The virus-associated cancers have a greatly increased incidence in immunosuppressed individuals with an impaired ability to control virus infections, such as AIDS patients or solid organ transplant recipients [2–6]. The cancer that has the most highly increased incidence among the immunosuppressed is non-melanoma skin carcinoma (NMSC) [7–9], but does not have an established infectious etiology. A number of infections have been found in NMSC, with Human papillomavirus (HPV) being the most commonly found [10–15], but it is unclear whether any one of the infections detected so far are etiologically linked to the tumor.

Sequencing of all nucleic acids present in a sample is today a routine procedure [16–19]. The human sequences can be removed in the bioinformatics analysis of the sequences, leaving the metagenome (non-human genomes of e.g. bacteria and viruses) for study [20–22]. Virus detection that is dependent on prior knowledge of sequences, for example PCR with specific primer sets, can give a biased representation of which viruses that are present [23] and a simple sequencing of all nucleic acids in a sample gives a more unbiased representation of the viruses that are present.

Sequencing of the viral metagenome in skin samples reported that >95% of the virus sequences found belonged to the Papillomaviridae family, mostly to the β- and γ-genera [24], although recent improvements in bioinformatics using hidden Markov models now allow detection of a broader array of viruses in such samples [25]. A multitude of HPV types is present in skin lesions as well as on healthy skin, both among immunocompetent and immunosuppressed patients [11, 12, 26–29]. The description of which viruses that is most frequently present appears to be highly dependent on the different PCR primers used for HPV detection [12–15]. Sequencing of specimens without prior PCR is a more unbiased approach to detect viruses, even those which are not amplifiable by known PCR systems (2).

There are no fully adequate controls to use for case series of cancer tissue and it is therefore difficult to even guess if detection of a virus in a tumor suggests an etiological involvement or not. We reasoned that viruses consistently present in a series of independent tumors from the same patient might be more likely to be involved in tumorigenesis than viruses only found non-systematically in only some specimens. We therefore set out to test if there were indeed any virus types that were consistently found in such tumor series. Similar reasoning has been discussed before at a time point where HPV type-specific PCR-based approaches were used to detect HPV, however no single HPV type seemed to be consistently present in skin lesions of transplant recipients using biased PCR-based detection [30].

Materials and methods

Results

Eight different NMSC from the case and 8 NMSC from the different patients were sequenced using the Illumina technology. There were ~970 million sequencing reads, from which <0.01% belonged to viral sequences (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Taxonomic assignment of sequencing reads.

https://doi.org/10.1371/journal.pone.0172308.t001

The Blast-based bioinformatics pipeline revealed 45970 viral reads, of which 38416 (83.6% of viral reads) were detected in the 8 different NMSC tumors belonging to the case and only 38 reads were detected in the 8 NMSC specimens belonging to the different patients (Table 2). Most viral reads (26308 viral reads) were from Human Herpesvirus 7 (HHV-7), which was found in only one NMSC specimen. Human papillomavirus (HPV) was detected in 6/8 NMSC specimens retrieved from the case patient (all 6 of them had a double infection with both HPV 15 and HPV 38). Only one NMSC among the controls had HPV detected (a single HPV 38 infection). The same “Faustovirus-related” apparently environmental virus was found in both case specimens, control specimen as well as in the water control with 2348, 20 and 5562 reads, respectively (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Classification of viral reads.

https://doi.org/10.1371/journal.pone.0172308.t002

All reads that mapped to contigs that did not give any hit to genomes deposited in GenBank were classified as “unknown”, representing 4.13% of total reads (Table 1). Searching for distant homologs of already known viruses among these contigs using profile hidden Markov models revealed 23 (previously not described) viral contigs (designated as SE-sequences) related to up to 5 viral clusters (Table 2). None of the 23 SE-sequences were detected in more than 2 specimens each and only 2/5 clusters (Baculoviridae and Marseilleviridae) were not detected in the negative controls. The DNA viral contigs detected with the HHMER analysis (designatedSE881-SE902, SE871) have been deposited in GenBank (accession numbers KY006985-KY007007) and the deep sequencing data is deposited at NIH Sequence Read Archive (BioSample accessions SAMN06275823-SAMN06275840).

Discussion

We report that sequencing of tumor tissues from different cancers occurring up to 10 years apart in the same individual, to a large extent tended to detect the same viruses present in multiple different tumors. While this finding does in itself not imply causality, it does suggest that viruses that are systematically detected in multiple tumors from the same patient might be more likely to be involved in tumorigenesis than viruses only found non-systematically in only some specimens.

The study is based on only a few patients and more patients would need to be sequenced to obtain more large-scale data. Although deep sequencing is today a routine, it is still costly and more large-scale studies would probably need to await less expensive technologies to be enabled. Another limitation is that the study compared only different cases of SCC, whereas benign tissues were not analyzed. As the study design was based on specimens from pathology archives, only tumors (and not benign tissues) were available.

The majority of viral reads detected (37.8% of total viral reads), were from HHV-7, a beta-herpesvirus known to infect almost all children by the age of three years with subsequent lifelong persistence, but all the HHV-7 reads were from a single specimen. The use of sequencing for virus detection is not dependent on prior knowledge of sequences, and therefore a more unbiased representation of viruses will be detected. However, the fact that we before sequencing used whole genome amplification (WGA) to increase the amount of DNA in the sample means that the number of reads obtained is not directly correlated to the amount of viral populations present [39]. This phenomenon is well described and is probably attributable to preferential amplification of genomic regions during initial WGA priming events [16, 40, 41]. The significance of the large amount of HHV-7 reads in a single specimen is therefore unclear.

Considering presence in multiple tumors, HPV was the most common virus found. Double infection of HPV 15 and 38 was detected in 6/8 NMSC belonging to the same patient, whereas the NMSCs from 8 different patients were HPV-positive only in 1/8 tumors (a single infection with HPV38). Both HPV 15 and 38 belong to the genus Beta-papillomavirus and are commonly found in patients with epidermodysplasia verruciformis. Further studies are needed to resolve the biological significance of the finding of HPV15 and 38 in multiple tumors from the same patient, but it can be concluded that as these cancers occurred up to 10 years apart the double infection with HPV15 and 38 must have persisted for many years in this subject. Persistence of HPV over time is known to be of importance for tumor maintenance and progression.

A large part of the sequencing reads from de novo sequencing projects are classified as “Unknown” due to incompleteness of public sequence databases or drawbacks of next generation sequencing technologies such as short read lengths and sequencing errors [20, 21]. The Blast-based bioinformatics pipeline revealed ~45000 viral reads, while the number of “Unknown” reads reached ~4M. Therefore, the more sensitive HHMER algorithms were used in order to search for distant homologs of already known viruses among the contigs classified as “Unknown” [25]. HHMER-based analysis identified 23 previously unknown viral contigs with up to 26948 viral reads. Each contig within a cluster was assigned a different SE-number (e.g: SE898) if its sequence did not overlap with other contigs related to the same cluster. Although the different SE-sequences were not overlapping, it is still possible that different contigs from the same cluster may belong to the same virus.

Viral sequences belonging to Mimiviridae, Phycodnaviridae, “related to Faustovirus”, and “Unassigned” (distantly related to viruses previously described, but not given any taxonomic designation yet), were detected in not more than 2 specimens each as well as in the water and/or the blank paraffin block used as negative controls. These sequences might therefore represent environmental deposit, highlighting the importance of including negative controls whenever performing metagenomic sequencing.

Conclusions

In this study, deep sequencing was used for unbiased virus detection in NMSC specimens. Consecutive specimens from the same patient over several years and from different locations identified presence of HPV (types 15 and 38) in most of the specimens (6 out of 8), whereas only 1 out of 8 specimens from different patients was HPV-positive (type 38).