TaME-seq: An efficient sequencing approach for characterisation of HPV genomic variability and chromosomal integration

HPV genomic variability and chromosomal integration are important in the HPV-induced carcinogenic process. To uncover these genomic events in an HPV infection, we have developed an innovative and cost-effective sequencing approach named TaME-seq (tagmentation-assisted multiplex PCR enrichment sequencing). TaME-seq combines tagmentation and multiplex PCR enrichment for simultaneous analysis of HPV variation and chromosomal integration, and it can also be adapted to other viruses. For method validation, cell lines (n = 4), plasmids (n = 3), and HPV16, 18, 31, 33 and 45 positive clinical samples (n = 21) were analysed. Our results showed deep HPV genome-wide sequencing coverage. Chromosomal integration breakpoints and large deletions were identified in HPV positive cell lines and in one clinical sample. HPV genomic variability was observed in all samples allowing identification of low frequency variants. In contrast to other approaches, TaME-seq proved to be highly efficient in HPV target enrichment, leading to reduced sequencing costs. Comprehensive studies on HPV intra-host variability generated during a persistent infection will improve our understanding of viral carcinogenesis. Efficient identification of both HPV variability and integration sites will be important for the study of HPV evolution and adaptability and may be an important tool for use in cervical cancer diagnostics.

widely studied and is regarded as a determining event in cervical carcinogenesis [21][22][23] . Upon integration, disruption or complete deletion of the E1 or E2 gene is often observed, resulting in constitutive expression of the E6 and E7 oncogenes [24][25][26] , inactivation of cell cycle checkpoints and genetic instability 23 . Viral integration may also lead to modified expression of cellular genes nearby, disruption of genes, as well as genomic amplifications that may promote oncogenesis 23,27 . The finding of certain chromosomal clusters of integration in precancerous lesions and cancers 28 also suggests a selective advantage of specific HPV integrations. Still, several important questions remain for HPV integration and more comprehensive analyses of integration sites are needed in order to expand our understanding of HPV pathogenesis. The development of next generation sequencing (NGS) technologies has provided new tools for viral genomic research. During the recent years, a few studies have described different NGS based approaches to study HPV variability and integration in the human genome. The most common approaches used in HPV genomic analyses are based on target enrichment using highly multiplexed degenerate primers 29 , enrichment by multiplex PCR using HPV16 forward primers 30 , bead-based target capture [31][32][33] , and rolling circle amplification 34 followed by NGS. These methods are however designed to detect either HPV integration or HPV variability. In addition, target capture methods poorly enrich HPV and remain expensive due to high probe cost and off-target sequencing.
In order to contribute to the understanding of the role of intra-host HPV genomic variability and chromosomal integration in carcinogenesis, we have developed an innovative library preparation strategy followed by an in-house bioinformatics pipeline named TaME-seq (tagmentation-assisted multiplex PCR enrichment sequencing). TaME-seq combines tagmentation and multiplex PCR enrichment, allowing simultaneous HPV genomic variability and integration analysis (Fig. 1). TaME-seq, with highly efficient target enrichment and reduced sequencing cost, enables deep sequencing analysis in order to find low frequency variants and rare integration events. Here, we present the results of HPV integration and genomic variability analysis in HPV16, 18, 31, 33 and 45 positive clinical samples and cell lines. The method described here provides an important tool for comprehensive studies of HPV genomic variability and chromosomal integration, and it can also be adapted to studies on other viruses such as retroviruses, adeno-associated viruses and integrating human herpesviruses.

Results
Read mapping analysis and genome coverage. Table 1 summarises liquid-based cytology (LBC) samples (n = 21), cell lines (n = 4) and plasmid samples (n = 3) included in the analysis. The samples generated 154.8 million raw reads of which 72.5 million reads (47%) mapped to the target HPV reference genomes. Only a small fraction (0.08%) of the reads mapped to other HPV types than those reported positive by HPV genotyping. The mean coverage ranged from 303 to 273898, while the fraction of the genome covered by minimum 10 × ranged from 0.35 to 1, and the fraction of the genome covered by minimum 100 × ranged from 0.33 to 1 (Table 1). HPV genome sequencing coverage aligned to the target HPV genomes with the location of HPV genomic regions and primers is visualised for CaSki, HeLa, LBC34, LBC11 and MS751 (Fig. 2). Overall, the samples showed varying HPV genome coverage profiles (Supplementary Figs S1-S5). Totally, 10 HPV positive samples were excluded from further analysis due to poor sequencing coverage (Supplementary Table S1). Sequencing of the HPV negative control samples resulted in no or negligible amount (<500) of reads mapped to target HPV genomes (Supplementary Table S2). The MS751 cell line was confirmed not to contain HPV18 sequences (Supplementary Table S1) 35 .
Deletions in HPV genomes. The method enables identification of regions covered with very few or no sequencing reads, interpreted as large HPV genomic deletions. Cell lines HeLa and MS751 are known to contain partial HPV genomes due to deletions of 2.5 kb and 5 kb, respectively 35,36 , which was confirmed by our method ( Fig. 2). A large deletion of 4.8 kb was revealed in the clinical sample LBC105, indicating partial or complete deletion of HPV18 genes E1, E2, E4, E5, L1 and L2 ( Supplementary Fig. S2).

HPV-human integration sites.
A two-step strategy was applied to detect possible integration sites (Fig. 3).
A total of 27 integration sites were detected in cell lines CaSki, SiHa, HeLa and MS571 (Table 2). For CaSki, 16 previously reported integration sites 30,32,37 were confirmed. In addition, three novel sites were identified. These mapped to HPV16 E6, E2 and L1 genes. One was located in an intronic region of the gene BRSK1; two were located more than 50 kb from annotated genes ( Table 2). Three sites, including one previously reported site as a control 30,37 , were subjected to Sanger sequencing to confirm the integration sites (Supplementary Table S3). Integration sites identified in SiHa, HeLa and MS751 were consistent with previous studies 31, [35][36][37][38][39] and were not subjected to validation by Sanger sequencing. Additionally, two integration sites were detected in the clinical sample LBC105 ( Table 2). The integration breakpoints were mapped to the HPV E1 and L1 genes flanking the deleted region ( Supplementary Fig. S2) and they were located in intronic regions of the gene GTF2IRD1 (Table 2). Both integration sites were confirmed by Sanger sequencing (Supplementary Table S3).

Discussion
Here, we present a novel cost-efficient approach, TaME-seq, for the simultaneous analysis of HPV variation and chromosomal integration. Previous methods have been less effective and/or limited to either one of the two analyses [29][30][31][32][33][34] . To demonstrate the performance of TaME-seq, we employed HPV16, 18, 31, 33 and 45 positive clinical samples, HPV positive cell lines and HPV plasmids. With 47% of the total of 154.8 million raw reads mapped on the target HPV reference genomes, TaME-seq proved to be highly efficient in HPV target enrichment.
Other approaches for HPV target enrichment have reported much lower HPV mapping ratios 32,40 , requiring more sequencing and therefore at a higher sequencing cost. TaME-seq currently covers HPV16, 18, 31, 33 and 45, being the most common HPV genotypes in cervical cancer 5 . TaME-seq can be extended to cover additional HPV types, as well as other viruses, by implementing new primers to the method. The ability of TaME-seq to detect chromosomal integration sites has been shown for the HPV positive cervical cancer cell lines CaSki, SiHa, HeLa and MS751. CaSki cells contain a high copy number (~600 copies/cell) of integrated full-length HPV16 arranged in concatemers 41,42 . SiHa (1-2 HPV16 copies/cell) 39,41 and HeLa (10-50 HPV18 copies/cell) 43 cells harbour integrated HPV genomes. MS751 cells contains integrated HPV45 35 , but in contrast to the product specification sheet (ATCC, Manassas, VA) no HPV18, which was verified in our analyses. For CaSki, 16 previously reported integration sites 30,32,37 were detected by our method. In addition, three novel integration sites were identified. Known integration sites in SiHa 31,37,39 , HeLa 31,36 and MS751 35 , as well as large deletions demonstrated in HeLa 36 and MS751 35 , were confirmed by the TaME-seq method. Of the 21 LBC samples, HPV integration sites could only be detected in one sample, being in line with previous studies reporting no or few HPV integration events in LSIL/ASC-US samples 44,45 . However, other studies report integration events also in LSIL samples 32,46 . The detection of integrated forms of the virus is also dependent on the amount of episomes in the sample; low copy integration sites may remain undetected against a high background of episomal HPV.
The high sequencing coverage throughout the HPV genome enables detection of low frequency variants. Variant calling was evaluated using SiHa replicates to set the variant calling threshold. Previous studies have used variant calling thresholds of 0.5% or 1% 17,34 . With the high coverage provided by the TaME-seq method there is    potential for detecting very low frequency variation. We have therefore analysed the variation using 0.2% as the variant calling threshold. Multiple and stringent filtering steps was included to filter out non-reliable variants, as we are approaching the inherent error rate profile of the PCR amplification and Illumina sequencing 47 . However, the threshold for variant calling is dependent on experimental and analytical basis and must be set according to the study aims. The results from the SiHa analysis indicate that calling ultra-low frequency variants is dependent on the sequencing coverage. Lower sequencing coverage results in the detection of fewer variants and less concordance between sample replicates. In order to find ultra-low frequency variants, high sequencing coverage is required. Figure 4 shows that at the mean coverage of 12000×, the number of variants in SiHa-1 is approaching saturation. This indicates that more variants are not likely to be found even with higher sequencing coverage. Finally, differences in sequencing coverage affect the number of variable sites found, but also experimental approaches due to stochastic sampling and variant calling can fail to reveal low frequency variants. Overall, our results uncover low frequency variants in the samples, potentially introduced by DNA repair mechanisms and APOBEC enzyme mediated DNA editing [48][49][50] , although some bias may be introduced by PCR and sequencing. Variable sites are present in all genes of the studied HPV types. Traditionally, studies have focused on sequence variation on a viral sublineage level [13][14][15][16] or the high variability has been interpreted as HPV variant co-infections 29 . The development of NGS technologies has provided comprehensive tools for the study of HPV genomic variability. Recent studies have reported high HPV variability that may be evidence of intra-host viral evolution and adaptation generated during a chronic HPV infection [17][18][19][20] . Our study has some limitations. Firstly, TaME-seq is not intended for determining HPV genotypes and we recommend it for analyses of HPV variability and integration events in samples with known HPV status. Secondly, due to variation in amplification efficacy, an uneven coverage is seen for different genomic regions. Sudden drops in the coverage, that are not genomic deletions, may be due to suboptimal primer performance or poor alignment against the reference genomes. This issue can be solved partly by designing new primers covering these regions and optimising the primer performance. Also, the read alignment step can be further optimised. Alternatively, alignment could be performed by de novo assembly to create consensus sequences for the alignment. Thirdly, enough viral DNA and good dsDNA quality are important for achieving consistent tagmentation results in the Nextera protocol 51 . Sample preparation of the excluded LBC samples failed likely due to very low viral load in the samples, which was not quantified separately.
In summary, we have developed a NGS approach that allows the simultaneous study of HPV genomic variability and chromosomal integration. TaME-seq is applicable to large sample cohorts due to its highly efficient target enrichment, leading to less off-target sequences and therefore reduced sequencing cost. Comprehensive studies on HPV intra-host variability generated during a persistent infection will improve our understanding of viral carcinogenesis. Efficient identification of HPV genomic variability and integration sites will be important both for the study of HPV evolution, adaptability and may be a useful tool for cervical cancer diagnostics.

Methods
Samples. Anonymised LBC samples from routine cervical cancer screening were included in the study, comprising cases of atypical squamous cells of undetermined significance (ASC-US) and low-grade squamous intraepithelial lesions (LSIL). HPV positive samples with the cobas 4800 HPV test (Roche Molecular Diagnostics, Pleasanton, CA) were extracted for DNA using the automated system NucliSENS easyMAG (BioMerieux Inc., France) with off-board lysis. The samples were HPV genotyped using the modified GP5+/6+ PCR protocol (MGP) 52 , followed by HPV type-specific hybridisation using Luminex suspension array technology 53    Results from both reactions of the same sample were combined and method performance was then evaluated based on the percentage of obtained reads mapped to the HPV reference genome, mean sequencing coverage and percentage of HPV reference genome coverage for each sample. Further analysis was performed when a sample had >20000 reads mapped to the target HPV reference genome. The target HPV genomes correspond to the HPV types for which the samples were reported positive by HPV genotyping.
Detecting HPV-human integration sites. The paired-end reads that mapped (HISAT2) with one end to a human chromosome and the other end to the target HPV reference genome were identified as discordant read pairs. If a specific position had ≥2 read pairs with unique start or end coordinates, it was considered as a potential integration site. To determine the exact position of HPV-human integration breakpoints, previously unmapped reads were remapped to human and HPV reference genomes (as above) using the LAST (v876) aligner (options -M -C2) 62 . Positions covered by ≥3 junction reads, with unique start or end coordinates, were considered as potential integration breakpoints. Integration site detection was not based on reads sharing the same start and end coordinates as these reads were considered as potential PCR duplicates. Selected HPV integration breakpoints were confirmed by PCR amplification and Sanger sequencing.

Sequence variation analysis. Mapped nucleotide counts over HPV reference genomes and average map-
ping quality values of each nucleotide were retrieved from BAM files and variant calling was performed using an in-house R script. To reduce the effects of PCR amplification and sequencing artefacts in the variation analysis, filtering was applied before the variant calling. Nucleotides seen ≤2 times in each position and nucleotides with mean Phred quality score of <20 were filtered out. Nucleotide counts from both reactions of the same sample were combined and variant allele frequencies (VAF) of the three minor alleles in each position were calculated. If results from either of the reaction showed >5 times larger VAF with <20% of the total coverage, it was discarded from variant calling. Finally, variants were called if VAF was >0.2% and coverage was ≥100×. Two sequencing libraries of SiHa cell line served as technical replicates to assess the variant calling performance. The technical replicates were sequenced on the MiSeq platform or on the HiSeq 2500 platform. In addition, HiSeq raw sequencing data was downsampled randomly and defined portions (90%, 75%, 50% and 25%) of the original reads were further analysed. Reproducibility of calling variants in the replicates was assessed by calculating concordance rate. The concordance rate (R c ) between duplicates was defined as follows: where N c was the number of concordant variants between a pair of replicate samples, and N 1 and N 2 were the total number of variants detected in each of the duplicated sample.
Ethical approval. This study was approved by the regional committee for medical and health research ethics, Oslo, Norway [2017/447] and we confirm that all experiments were performed in accordance with the committee's guidelines and regulations.