Insights into the mechanisms and structure of breakage-fusion-bridge cycles in cervical cancer using long-read sequencing

Summary

Cervical cancer is caused by human papillomavirus (HPV) infection, has few approved targeted therapeutics, and is the most common cause of cancer death in low-resource countries. We characterized 19 cervical and four head and neck cancer cell lines using long-read DNA and RNA sequencing and identified the HPV types, HPV integration sites, chromosomal alterations, and cancer driver mutations. Structural variation analysis revealed telomeric deletions associated with DNA inversions resulting from breakage-fusion-bridge (BFB) cycles. BFB is a common mechanism of chromosomal alterations in cancer, and our study applies long-read sequencing to this important chromosomal rearrangement type. Analysis of the inversion sites revealed staggered ends consistent with exonuclease digestion of the DNA after breakage. Some BFB events are complex, involving inter- or intra-chromosomal insertions or rearrangements. None of the BFB breakpoints had telomere sequences added to resolve the dicentric chromosomes, and only one BFB breakpoint showed chromothripsis. Five cell lines have a chromosomal region 11q BFB event, with YAP1-BIRC3-BIRC2 amplification. Indeed, YAP1 amplification is associated with a 10-year-earlier age of diagnosis of cervical cancer and is three times more common in African American women. This suggests that individuals with cervical cancer and YAP1-BIRC3-BIRC2 amplification, especially those of African ancestry, might benefit from targeted therapy. In summary, we uncovered valuable insights into the mechanisms and consequences of BFB cycles in cervical cancer using long-read sequencing.

Graphical abstract

We used long-read whole-genome sequencing to characterize a panel of cervical cancer cell lines, identifying frequent amplification of the YAP1 oncogene through a telomere-deletion mechanism. YAP1 amplification marks a subset of cervical cancer with early diagnosis and is more common in minority populations.

Introduction

Cervical cancer ranks fourth in worldwide cancer prevalence.¹ Countries with a low human development index (HDI) have higher cervical cancer incidence and mortality.² Over 90% of all cervical cancer deaths occur in low HDI countries.^3,4 Cervical cancer is predominantly caused by oncogenic human papillomavirus (HPV) infection, detected in over 90% of individuals with cervical cancer.¹ Two main high-risk HPV (hrHPV) types, HPV16 and HPV18, are responsible for at least 70% of cervical cancers.⁵

After infection, the double-stranded circular HPV genome, which is approximately 7,900 base pairs (bp) in length, relocates to the nucleus, where it replicates as an extrachromosomal episome. The main drivers of HPV pathogenesis are the E6 and E7 viral oncoproteins. Upon expression, E6 binds to and inactivates tumor suppressor p53, and E7 suppresses the cell-cycle regulator pRB.⁶ In many HPV-driven tumors, it is common for HPV to integrate into the human genome, frequently deleting the viral E1 and E2 genes.⁷ Because E2 is a transcriptional repressor of E6 and E7, integration results in higher E6 and E7 expression, promoting tumor progression.⁸ HPV integrates into many locations in the human genome, with 37 known hotspots.⁹ Integration may delete, amplify, or activate cellular genes.^10,11,12 Phylogenetic analysis groups each HPV type into lineages and sublineages based on nucleotide variation.¹³ The distribution of lineages and sublineages varies by geographic region and genetic background.¹⁴ Cancer risk also varies by lineage and sublineage. For example, the HPV16 A4, D2, and D3 sublineages are more carcinogenic¹⁵ than the most frequent HPV16 A1 sublineage.¹⁴

Another factor that influences the risk of HPV infection and cancer progression is genetic variation of the host immune system. Thus, the immune response to viral infection is mediated by the major histocompatibility complex (MHC) proteins.¹⁶ MHC class I genes, also known as human leukocyte antigens (HLAs) A, B, and C, encode glycoproteins found on the surface of all nucleated cells. These complexes bind and present viral peptides to cytotoxic T cells, specialized immune cells that can recognize and kill virally infected cells.¹⁷ Indeed, 90% of HPV infections are cleared within 1–2 years.¹⁸ However, some HPV infections can persist and progress to cervical cancer due to various factors interfering with MHC class I-mediated immune recognition. HLA class I genes are highly polymorphic, and 90% of people are heterozygous for a given HLA gene.¹⁹ This increases the diversity and coverage of the immune response against HPV. However, many cervical cancer cells show loss of heterozygosity (LOH) for the HLA class I genes.²⁰ This reduces the number and variety of viral peptides that can be presented to cytotoxic T lymphocytes (CTLs), allowing the virus to evade immune detection. Another factor that impairs the MHC class I-mediated immune recognition is the expression of HPV proteins that modulate the cellular pathways involved in antigen processing and presentation. For example, the HPV E5 protein inhibits the transport of HLA class I protein complexes to the cell surface,²¹ and the E6 and E7 proteins prevent apoptosis, which is a process that induces the release of viral antigens and stimulates the immune response.²² Moreover, the integration of HPV DNA into the host genome disrupts the expression of viral proteins, especially E2, which regulates the transcription of other viral genes.²⁰ This reduces the availability of viral peptides for MHC class I presentation.

Cervical cancer, like all solid tumors, is characterized by specific chromosomal rearrangements that are vital somatic events during the tumorigenesis process.²³ The deletion of the telomere can lead to breakage-fusion-bridge (BFB) cycles. This mechanism was described by Barbara McClintock in maize,²⁴ demonstrating that the deletion of the telomere of a chromosome can lead to BFB cycles. Telomere loss leads to the formation of a dicentric chromosome that is pulled apart during cell division, resulting in breakage at a random site. BFB cycles can repeat indefinitely, generating complex chromosomal rearrangements and amplifications.²⁵ In cancer cells, BFB cycles can increase the copy number of segments of the unstable chromosomes that contain oncogenes, providing a selective advantage for cells carrying them.^26,27 However, the structure and origin of BFB events in cancer cells are poorly investigated due to the limitations of the conventional short-read sequencing methods.²⁸ Long-read whole-genome sequencing (WGS) is a powerful tool for studying BFB and provides an opportunity to understand this mechanism of oncogene activation.

Cervical cancer requires a better understanding of the molecular defects, such as BFB, that drive its progression and resistance. The main treatment options for cervical cancer are surgery, chemotherapy, and radiation. There is currently no targeted therapy approved for cervical cancer. But recent data suggest that 15%–20% of individuals with cervical cancer respond to immunotherapy with checkpoint inhibitors.²⁹ Moreover, engineered T cell therapy was recently proven successful in a clinical trial.³⁰ Therefore, the characterizing of these cell lines will provide a valuable resource for the preclinical evaluation of targeted therapy and the further translation of such data into clinical practice.

Material and methods

Cell culture

The cell lines were obtained from the American Type Culture Collection (ATCC) and the Korean Cell-Line Bank and Cancer Research Center, Seoul National University.³¹ Cells were cultured in EMEM or RPMI-640 media with 10% fetal bovine serum, 1% Pen-Strep (10,000 units/mL of penicillin, 10,000 μg/mL of streptomycin, and 25 μg/mL of amphotericin B) until 70%–80% confluent. Cells were washed with 10 mL PBS and harvested using 2 mL trypsin per T-75 flask. All cell lines were confirmed by Identifier analysis and regularly shown to be mycoplasma negative.

PDX cell methods

The tissues for PDX-CK3489 were obtained from the NCI patient-derived models repository. The tissue was obtained from a 74-year-old male with HPV+ head and neck squamous cell carcinoma (HNSCC) (collected on 05/2014) (https://pdmdb.cancer.gov/web/apex/f?p=101:3:0::NO:3:P3_PATIENTSEQNBR:244). The frozen tissue (2–3 mm cube) was thawed on ice and washed with ice-cold PBS to remove DMSO (a component of the freezing media). We implanted the washed tissues under the skin of the right flank of the next-generation sequencing (NGS) mice. The animal was euthanized when tumors reached a maximum of 10% of body weight (∼2,000 mm³). The tumor was harvested and frozen in 10% DMSO and 90% FBS freezing media until further use or analysis.

DNA and RNA extraction

Cells or tissues were washed with 10 mL PBS and harvested using 2 mL trypsin per T-75 flask. DNA was extracted and purified using a Gentra Puregene kit from Qiagen. Cell line RNA was prepared from 30 million cells using Trizol (ThermoFisher) and Poly-A+ RNA purified by DYNAL Dynabeads (Invitrogen). DNA was quantified by Nanodrop (Thermo Scientific) and Qubit (Thermo Scientific) and stored at 4°C; RNA was quantified by Qubit and stored at −80°C.

PCR amplicon and Sanger sequencing validation

The Primer3 online program³² was used to design primers for PCR amplification. Overlapping primers were designed to span the entire 7.9-kb HPV genomes. Cell line DNA was amplified using a long-range PCR kit from New England BioLabs. The samples were run on a 1.5% agarose gel at 100 V for 1 h and imaged with a BioRad ChemiDoc Image System. To determine the HPV lineage and sublineage, we purified the PCR products and sequenced them on an Applied Biosystems 3500xL Genetic Analyzer from Thermofisher Scientific. We analyzed the sequences using DNASTAR SeqMan Ultra and SeqBuilder Pro software.

HLA typing by DNA sequencing

HLA typing was performed using targeted NGS with locus-specific primers used to amplify a total of 26 polymorphic exons of HLA-A and B (exons 1 to 4), C (exons 1 to 5), E (exon 3), DPA1 (exon 2), DPB1 (exons 2 to 4), DQA1 (exon 1 to 3), DQB1 (exons 2 and 3), DRB1 (exons 2 and 3), and DRB3/4/5 (exon 2) genes with Fluidigm Access Array (Fluidigm Corporation, South San Francisco, CA 94080, USA). The 26 Fluidigm PCR amplicons were pooled and subjected to sequencing on an Illumina MiSeq sequencer (Illumina, San Diego, CA 92122, USA). HLA alleles and genotypes are called using the Omixon HLA Explore (version 2.0.0) software (Omixon Biocomputing Ltd., Budapest, Hungary).

Long-read DNA sequencing

We used the Ligation Sequencing kit (SQK-LSK109, SQK-LSK110, Oxford Nanopore Technologies [ONT]) to perform long-read DNA sequencing of the cell line DNA samples. We used either unsheared DNA or DNA that was sheared to 8–20 kb with a G-tube (Covaris) from 1 to 4 μg of input DNA. We applied adaptive sampling to selected samples using a reference FASTA file containing the human HG38 genome and hrHPV-type FASTA file and a BED file that specified the regions of interest, such as cancer genes, integration loci, and HPV sequences. We loaded the samples onto MinION R9.4 flow cells on a GridION instrument and performed the sequencing runs. Table S1 provides the details of each sequencing run, and Table S2 summarizes the WGS coverage of each cell line. All sequences have been deposited at the Sequence Read Archive (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA772772).

Identification and characterization of HPV integration events

WGS reads from ligation sequencing or adaptive sampling were mapped to the hg38 human genome assembly and a file containing 13 hrHPV types. Reads aligning to both human and HPV were identified, and individual segments of human and HPV DNA were identified with BLAT and BLAST. We validated the newly described junctions between human and HPV DNA by PCR and Sanger sequence, and we generated the final assemblies of the integrated HPV genomes using SeqMan 17 Ultra software (DNAStar).

Identification of somatic mutations

For cell lines sequenced by the COSMIC cell line (https://cancer.sanger.ac.uk/cell_lines) or CCLE (https://sites.broadinstitute.org/ccle/) projects, driver mutations in the 299 PanCancer Driver genes were identified from public data. We confirmed these mutations in our ONT data by manually reviewing BAM files. In previously uncharacterized cell lines, we used the long-read data to call variants in the genes with driver mutations observed in the COSMIC and CCLE cell lines. Clair3³³ was run with the default parameters to call SNVs for the uncharacterized cell lines. SNVs were then annotated with SnpEff,³⁴ and SNVs with high or moderate impact were filtered with SnpSift.³⁵ We further filtered with the list of genes with driver mutations observed in the COSMIC and CCLE cell lines. Lastly, SNVs were visually verified using Integrative Genomics Viewer (IGV). Mutations in PIK3CA exons 9 and 20 were confirmed by PCR and Sanger sequence.

Analysis of BFB events in cancer cohorts

The Cancer Genome Atlas (TCGA), AACR Genie, and Memorial Sloan Kettering Cancer Center (MSKCC) datasets were accessed through cbioportal (http://www.cbioportal.org/). Copy-number (log2) values of >2 for YAP1 and <0 for 11q telomeric genes were identified as potential BFB events. Data for age at diagnosis and race/ethnicity were obtained from the same site. Whole-genome haplotype-specific coverage plots were generated using Wakhan (https://github.com/KolmogorovLab/Wakhan).

Full-length RNA sequencing

Cell-line RNA (500 ng Poly-A+) was sequenced using the Direct RNA kit (DCS109) and Direct RNA sequencing kit (SQK-RNA002, ONT). RNA libraries were loaded onto MinION R9.4 flow cells on a GridION instrument (ONT). Transcriptome fastq files were aligned and reads assigned to genes using the Exome workflow in EPI2ME (EPI2ME Dashboard [nanoporetech.com]). Reads were normalized as reads per million (RPMs). HPV reads were identified by alignment to a file with 13 hrHPV sequencing plus HPV26 and HPV30 using EPI2ME. HPV reads were counted and normalized as RPMs, and E6 splicing was assessed by manual counting of spliced and unspliced reads in BAM file produced in the Cancer Genomics Cloud (CGC) platform (https://cgc.sbgenomics.com/).

Detection of inverted reads and putative BFB events

To identify large-scale chromosomal alterations, copy-number plots of each cell line were generated by binning the start of alignment of each read to HG38 into 1 and 10 million bp bins across each chromosome. For adaptive sampling runs, it was confirmed that the plots were identical to runs without adaptive sampling. This is because in adaptive sampling, all reads generate at least 400 bp of sequence, and each read is only counted one time. The BFB events were detected using Severus (https://github.com/KolmogorovLab/Severus). Briefly, clusters of split-read alignments were used for each cell line to identify breakpoints. Then, clusters with foldback inversions and change in coverage were selected as BFB candidates. Identified breakpoints were further manually investigated using IGV.

Validation of inverted sequences

Inverted reads at BFB sites were confirmed by designing primers specific to single-copy sequences at the junctions using the Primer3 tool. We performed single-primer PCR reactions with the forward primers and resolved the products on agarose gels. We selected the primers that produced bands only in the cell line with the BFB event and added barcodes to them. We sequenced the barcoded products on ONT Flongle flow cells and aligned the sequences to the HG38 human genome assembly. We mapped the reads to confirm the presence of BFB events.

Bioinformatics and statistics

We performed sequence alignment of the FASTQ files obtained from the sequencing runs to the human genome (HG38) or HPV genomes using the EPI2ME server (https://epi2me.nanoporetech.com). We used the Fastq Human Alignment GRCh38 app for human genome alignment and the Fastq Custom Alignment workflow for HPV genome alignment, providing a FASTA file of HPV16 or HPV18 or 13 other hrHPV types as a reference. We combined the alignment data for HG38 and HPV with Excel or in Filemaker (Claris) and added read-length and adaptive sampling decision data, if applicable. Reads of interest were manually extracted and mapped using BLAT³⁶ (https://genome.ucsc.edu) or BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi). BAM files produced by EPI2ME were merged and indexed using the BamTools Merge and SAMtools Index tools in the CGC.³⁷ We also ran an ONT WGS Data Processing pipeline on CGC, based on a Broad Institute pipeline, to align human and HPV reads and produce BAM files. BAM files were viewed in the IGV.³⁸

Pipelines

DNA reads were analyzed using fast5 raw data by base-calling using Guppy v.4.5.4 (https://nanoporetech.com/). Modified base-calling was performed using Megalodon v.2.3.3 (https://github.com/nanoporetech/megalodon). Structural variation calling was carried out with the Nanopore pipeline-structural-variation with modifications. The entire workflow is available at https://github.com/NCI-CGR/Nanopore_DNA-seq. RNA reads were processed using the fast5 raw data and base-called using BONITO v.0.3.7 and aligned to the HG38 genome using Minimap2 v.2.17. Isoforms were detected and quantified using Stringtie2 v.2.1.5 and Freddie (https://github.com/vpc-ccg/freddie). The Freddie program calls the Gurobi package (www.gurobi.com) to solve optimization problems. The entire workflow is available at https://github.com/NCI-CGR/Nanopore_RNA-seq.

HPV lineage classification

HPV lineage and sublineage was determined using MEGA³⁹ and ClustalOmega. Cell line FASTA files were exported from the GridION and translated into amino acid sequences using MEGA. Maximum likelihood phylogenetic trees were constructed to determine the closest HPV variant match for each cell line with Bootstrap values of >80 used. Domain sites for HPV genes E6, E7, E1, E2, L1, and L2 were set to look for amino acid changes in each sequence specified by that region. Alignments were confirmed using Clustal Omega.

Statistical analyses

Statistical analysis was performed in GraphPad. We used Fisher’s exact test to assess the relationship between HLA homozygous and heterozygous for type A, B, C. We used an unpaired t test with Welch’s correction to ascertain the relationship between the age of onset of YAP1 amplified tumors. In all calculations, a p value of 0.05 or less was deemed significant.

Results

Establishment of a cervical cancer cell line panel

To create a platform to study HPV and cervical cancer, we collected a comprehensive panel of 18 cervical cancer cell lines and four HPV-positive HNSCC lines and xenografts. We performed long-read WGS with and without adaptive sampling and full-length cDNA or direct RNA transcriptome sequencing on all cell lines (Figure S1). Table 1 summarizes the HPV types, histology, ancestry, age, and HPV integration status of the cell lines. Figure 1A shows the diverse ancestry of the subjects: 52% East Asian, 43% European, and 5% African American. In addition, there are eight different HPV types represented, with HPV16 present in 50% (11/22) of our panel.

Table 1.

HPV and cervical cancer cell lines and their identifying features

Cell lines	Cancer type	Histology	Ethnicity	Age/sex	Cellosaurus	HPV type	Sublineage	Integration status	Integration site	Integration site reference
SiHa	CESC	SCC	East Asian (Japan)	55 (f)	CVCL_0032	HPV16	A1	integrated	one, 13q22.1	Kalu et al.40
CaSki	CESC	SCC	European	40 (f)	CVCL_1100	HPV16	A2	integrated	many	Kalu et al.40
C33A	CESC	SCC	European	66 (f)	CVCL_1094	none	–	–	–	NA
HeLa	CESC	adeno	African American	30 (f)	CVCL_0030	HPV18	Aa	integrated	8	Kalu et al.40
SCC090	HNSCC	SCC	European	46 (m)	CVCL_1899	HPV16	A2	integrated	9	Kalu et al.40
SCC152	HNSCC	SCC	European	47 (m)	CVCL_C058	HPV16	A2	integrated	9q22.33 TRMO, HEMGN, FOXE1	Kalu et al.40
ME180	CESC	SCC	European	66 (f)	CVCL_1401	HPV68B	C1	integrated	18 ZBTB7C (APM-1)	Reuter et al.41
C4-I	CESC	SCC	European	41 (f)	CVCL_2253	HPV18	A4	integrated	8q21.2 PSKH2, ATP6V0D2	Kalu et al.40
MS751	CESC	SCC	European	47 (f)	CVCL_4996	HPV45	A3a	integrated	18q11.2, large cluster	Kalu et al.40
SCC154	HNSCC	SCC	European	54 (m)	CVCL_2230	HPV16	D3	integrated	21q11.2 NRIP1, USP25, SAMSN1	Kalu et al.40
SW756	HNSCC	SCC	European	46 (f)	CVCL_1727	HPV18	B3	integrated	12q34 HMGA2	Kalu et al.40
HT-3	CESC	SCC	European	58 (f)	CVCL_1293	HPV30	–	integrated	13q14.2	this study
SNU-1000	CESC	SCC	East Asian (Korean)	43 (f)	CVCL_5000	HPV16, episomal	A4	episomal + integrated	11q22.1 CEP126	Rossi et al.42
SNU-17	CESC	SC	East Asian (Korean)	40 (f)	CVCL_5029	HPV16	A4	integrated	19p13.13 TRMT1	this study
SNU-487	CESC	small cell	East Asian (Korean)	38 (f)	CVCL_5068	HPV18	A1	integrated	5q34 TENM2	this study
SNU-682	CESC	SCC	East Asian (Korean)	50 (f)	CVCL_5082	HPV33	A1	integrated	5q34 TENM2	this study
SNU-703	CESC	SCC	East Asian (Korean)	37 (f)	CVCL_5085	HPV16	A4	integrated	Xp22.11 ZFX	this study
SNU-778	CESC	SCC	East Asian (Korean)	62 (f)	CVCL_5092	HPV31	C2	integrated	10p12.1 RAB18	this study
SNU-902	CESC	SCC	East Asian (Korean)	57 (f)	CVCL_5107	HPV16	A4	integrated	20p12.1 MACROD2	this study
SNU-1005	CESC	SCC	East Asian (Korean)	58 (f)	CVCL_5001	HPV16	A4	integrated	14q24.1 RAD51B	this study
SNU-1245	CESC	SCC	East Asian (Korean)	61 (f)	CVCL_5019	HPV18	A1	integrated	1q32.2	Rossi et al.42
SNU-1299	CESC	SCC	East Asian (Korean)	72 (f)	CVCL_5021	HPV16	D3	integrated	3q26.32 150 KB from TBL1XR1	this study

HPV cell line information including HPV type, HPV sublineage, demographics, and integration sites are listed. Integration sites were originally identified in the references listed or in this paper. Characteristics of each cell line were determined utilizing techniques outlined in the methods. For some cell lines (a), the sublineage could not be determined because the sequence length was insufficient. CESC, cervical squamous cell carcinoma and endocervical adenocarcinoma; HNSCC, head and neck squamous cell carcinoma; NA, not applicable.

Figure 1.

Cell-line demographics and HPV expression data

(A) The distribution of the ancestry of the subjects, HPV types, and cancer and histological types among the 21 cell lines is shown.

(B) The cancer driver gene mutations found mutated in at least one cell line are shown. Orange highlights are pathogenic variants and in yellow are variants of unknown impact; H, homozygous variants; WT, wild type; AMP, amplification; no CNVs, no copy-number variants were found. Genes with an asterisk represent those not evaluated across all cell lines.

(C) The cDNA or RNA reads per million (RPMs) mapped to the HPV genome is shown.

(D) The percentage of HPV E6/E7 containing reads that are unspliced in the E6 gene.

We performed WGS using ligation sequencing on the Oxford Nanopore platform. We used sheared DNA, which generated 8 to 14 RPMs with an N50 value of 9–11 kb, and unsheared DNA, which generated 8 RPMs with an N50 value of 30 kb. Each DNA sample was sequenced on a single MinION flow cell (Table S1). We used adaptive sampling to enrich for sequences relevant to HPV and cervical cancer.^43,44 These regions included 299 genes that are frequently mutated or altered in cancer, genomic loci that are prone to HPV integration, and 13 hrHPV types. This approach yielded 7–24 RPMs on single MinION flow cells, with 5–10× coverage of most sampled genes and HPV integration breakpoints (Table S2). Rejected reads from adaptive sampling had a mappable length of 400–500 bp and were used to generate high-resolution copy-number profiles. As PromethION flow cells became available, we repeated some cell lines, achieving 20- to 30-fold coverage with an N50 of 10–11 kb (Tables S1 and S2).

To identify somatic driver mutations, we analyzed all variants in the 299 frequently mutated genes identified in the TCGA pan-cancer study.⁴⁵ Previously annotated somatic mutation data for 12 cell lines we studied were available from COSMIC (https://cancer.sanger.ac.uk/cell_lines). We also performed variant calling of the genes with driver mutations in the previously uncharacterized cell lines (Figure 1B). Consistent with prior research on cervical tumors,^46,47 PIK3CA is the most frequently mutated gene in the cell lines, and there are mutations in STK11, CASP8, ZFHX3, RB1, HLA-A, FBXW7, and KRAS in more than one cell line. In total, 18/22 cell lines have mutations in three or fewer driver genes. However, C33A, an HPV-negative cell line,⁴⁸ has nine known driver mutations, and HT-3, a cell line with an unknown-risk HPV (HPV30),⁴⁹ has five mutations. Therefore, our cell-line panel recapitulates the mutational spectrum of primary cervical tumors. Furthermore, this result suggests that cell lines without an hrHPV type require additional driver mutations for cancer progression.

Both direct cDNA and direct RNA methods were used to characterize HPV and cellular gene expression. Direct cDNA sequencing yielded an average of 2.6 RPMs (range 0.45–10 million) obtained per cell line on a single MinION flow cell and Direct RNA 0.8–1.2 RPMs (Table S1). The cell lines display a wide range in the level of HPV RNA expression, as determined by reads aligning to the HPV genome RPMs. The CaSki and ME180 cell lines have the highest HPV expression, while SNU-17 and SNU-703 have the lowest (Figure 1C). One of the oncogenic functions of hrHPV E6 protein is to degrade the tumor suppressor p53. However, a specific splicing event within the E6 gene occurs in tumors and cell lines that deletes a portion of E6, making it unable to degrade p53. Furthermore, the E7 protein has been shown to be exclusively translated from this spliced mRNA.⁵⁰ SNU-487 and SNU-1245 showed the highest frequency of E6 gene splicing at 60% and 36%, respectively, while SNU-682 and CaSki exhibited only 3% and 2% splicing frequency (Figure 1).

An MHC haplotype is frequently deleted in cervical cancer cell lines

The expression of HLA molecules is essential for the recognition and elimination of tumor cells by the immune system. Therefore, we characterized the HLA genotype of our cell lines (Figure 2A). A total of nine out of 18 cell lines have a homozygous genotype across all class I and class II HLA genes, indicating a deletion of one MHC haplotype. The percentage of homozygosity for each HLA class I gene was 56% for HLA-A and HLA-B and 61% for HLA-C (Figure 2B). This frequency is significantly higher than seen for 159 non-cervical cell lines (X² = 8.0, p = 0.0047).⁵¹ In addition, at least two cell lines, SW756 and HT-3, have loss-of-function mutations in HLA-A. These results indicate a strong selective pressure to eliminate an MHC haplotype or allele in cervical tumors.

Figure 2.

HLA class I homozygosity in cervical cancer cell lines

(A) The genotype for each class I and II HLA gene is shown with loci homozygous across the genes in the locus indicated in green and across 1–3 genes in yellow. The number and percentage homozygosity for each gene is indicated at the bottom.

(B) The percentage of tumor samples homozygous for individual HLA class I genes (X² = 8.0, p = 0.0047). The data for non-cervical cancer cell lines are from Boegel et al.⁵¹

Location and impact of HPV integration

To determine the spectrum of HPV integration sites, we mapped their locations in the human genome (Figures 3A and S2; Table 1). As observed in cervical tumors, the HPV integration sites in the cell lines are widely distributed and found on all human chromosomes.⁵² Each cell line has a unique integration site, and 10 out of 20 are in HPV integration hotspots. In addition, six cell lines have an integration near a known super enhancer active in cervical cells.⁵³ In SNU-487 and SNU-682, integration occurs at chromosomal region 5q34, near TENM2. TENM2 mediates cell adhesion and may represent a new cervical cancer hotspot.⁵⁴

Figure 3.

HPV integration and gene expression

(A) A map of all integration events in the human genome is shown for 20 HPV+ cell lines. Each point represents an integration event of which there may be one or multiple integration breakpoints. For the cell lines CaSki and SCC152, with multiple integrations, only the transcriptionally active site is shown. Each point lies in an approximate location of the integration event. The exact locations are included in Tables 1 and S2.

(B) Diagram of the HPV integration locus and flanking genes in four cell lines. The red arrows indicate a gene at the integration locus that is overexpressed in that cell line.

(C) The relative gene expression of each gene at or near an integration site was calculated as a Z score; bold boxes are genes that are near the HPV integration. Darker green colors indicate higher values, showing that those genes are highly expressed in those cell lines. Genes with an asterisk represent those where HPV integrated inside the gene. The white boxes show Z-score values that were removed as HPV hybrid reads could have inflated those numbers.

To investigate the impact of HPV integration on host genes, we examined the expression level of genes within approximately 1 Mb of integration events and compared them to the same genes in the rest of the cell-line panel. We observed that HPV integration activates the expression of one or more genes in the region in many cell lines. For example, in SCC152, the FOXE1 transcription factor gene is highly expressed, and in SNU-778, RAB18 (part of the RAS oncogene family) is overexpressed. The HPV integration occurs within a gene body in several cell lines, and HPV splices into one or more exons. We predict that this will result in insertional inactivation of that gene. For example, in SNU-1005, the RAD51B DNA repair gene is inactivated by insertion of HPV31 (Figure 3B). We quantified the relative expression of genes near integration sites in all cell lines using a Z-score statistic (Figure 3C). Many cell lines have at least one gene highly expressed relative to the other cell lines without integration at that locus. In contrast, only 1.5% of 23 randomly selected genes in chromosome arms not participating in HPV integration had elevated expression in a cell line (Z score >3, Table S3). Many cell lines have integration activating at least one gene of oncogenic importance, such as the MYC or YAP1 oncogenes and RUNX2, which encodes a transcription factor.

Gene expression in cervical cancer cell lines

We also identified oncogenes overexpressed independent of HPV integration. The MS751 cell line has gene amplification and mRNA over-expression of the EGF receptor (EGFR). Similarly, the YAP1 oncogene is overexpressed and amplified in SNU-1000 and SCC154 cells. YAP1 encodes a downstream nuclear effector of the Hippo signaling pathway involved in development, growth, repair, and homeostasis^55,56 (Figure S3).

To understand the molecular basis for cervical cancer in the absence of HPV or the presence of unknown-risk HPV types, we analyzed C33A, an HPV-negative cell line, and HT-3, a cell line with the unknown-risk HPV30 virus. C33A cells are homozygous for the pathogenic c.817C>T (p.Arg273Cys) (rs121913343) mutation in TP53 and homozygous for the c.1961−1G>A splice site mutation in RB1 (GenBank: NM_000321). Therefore, somatic mutations can substitute for the primary oncogenic function of HPV (Figures 4A and 4B).

Figure 4.

TP53 and RB1 mutations in HPV-negative and moderate-risk HPV cell lines

(A) The integration site of HPV30 in HT-3 cells demonstrates that the integration is 19 kb 3′ to RB1 within RCBTB2.

(B) A model showing that the HPV E6 and E7 proteins inhibit p53 and pRB, respectively, and mutations in RB1 (c.1331A>G [p.Gln444Arg]) and TP53 (c.817C>T [p.Arg273Cys], rs121913343) and (c.734G>T [p.Gly245Val], rs121912656) occur almost exclusively in cell lines without HPV or HPV of unknown risk.

In HT-3 cells, long-read WGS revealed that the unknown-risk HPV type, HPV30, is integrated on chromosome 13, 517 kb from the RB1, within RCBTB2. This cell line is also homozygous for a TP53 mutation (c.734G>T [p.Gly245Val], rs121912656) and is heterozygous for the RB1 mutation c.1331A>G (p.Gln444Arg), known to affect the splicing of exon 13.⁴⁸ In the 20 cell lines with hrHPV types, SNU-17 cells contain a deletion in one allele of RB1, and SNU-778 cells have a homozygous 4-bp insertion in RB1. However, none of these 20 cell lines have a TP53 mutation (Figures 1B and 4B). Therefore, somatic mutations in TP53 and RB1 can result in cervical cancer with unknown/low-risk HPV types or the absence of HPV.

BFB events identified in cell lines

We generated copy-number plots for each chromosome using the ONT DNA reads for each cell line. We observed that some chromosomes showed a focal gain in coverage, followed by a steep drop in coverage and LOH near the telomere (Figures S4A–S4I). Junction analysis identified inverted sequences consistent with BFB events (Figures 5A–5D). Additionally, we used Severus (https://github.com/KolmogorovLab/Severus) to identify inversions characteristic of BFB. We found 13 BFB events in nine cell lines (Table S4).

Figure 5.

Structure of BFB events

(A) The alignment of reads at the inversion site of a type I BFB event on chromosomal region 17q in HT-3 cells is shown. The coverage drops from 95× (segment A) to 54× (segment B) to 16-fold (reads from non-rearranged allele). All soft-clipped portions of reads are inverted in relation to the aligned portion of the reads. Arrows mark the start and end of the segment shown as B in the diagram below. All the reads spanning the junction are consistent with the fusion of inverted chromatids with staggered ends.

(B) A proposed model for the formation of the BFB junction of chromosome 2 in SNU-1000 cells, a type II BFB event. Following the deletion of one of the ends (a 1,052-bp deletion), a segment of chromosome 7 is inserted in between the joined chromosomes. The coverage drops from 52× (segment A) to 33× (segment B) to 11×.

(C) A type III BFB event on chromosome 11p in SNU-682 cells, inside WT1. The coverage drops from 75× to 6×. The junction contains a complex sequence likely derived from sequences adjacent to the breakpoint (see Figure S4G).

(D) Model for a BFB type I event. A telomere deletion results in a pair of deleted chromatids during mitosis. The free ends are subject to exonuclease digestion, and uneven digestion generates staggered ends. Fusion results in a lower copy number (one half) of the B segment. Type II events are formed when a segment from another chromosome (orange) is inserted at the junction, and type III events involve insertion of a sequence derived from the sequences flanking the breakage site (purple).

To explore the molecular mechanism underlying the formation of BFB events, we analyzed the long-read DNA sequences at the junction of the amplification events. At each junction site, there is a stepwise drop in coverage along with inverted reads. Figure 5A shows an example of a telomere deletion at chromosomal region 17q24.3 in HT-3 cells. We found that the junctions were composed of inverted reads of different lengths, indicating that one of the chromatids had a deletion before fusion. This is consistent with the model shown in Figure 5A where the copy number of the B segment is half that of segment A. We propose that segment B was generated by a BFB cycle involving a breakage and duplication of one chromatid during anaphase, followed by nuclease digestion of the free ends and fusion of the resulting staggered ends. In the HT-3 chromosomal region 17q BFB event, the size of the B segment was 688 bp. We observed a total of 13 BFB events in nine of the cell lines, which we termed and classified into three types based on the origin and structure of the DNA segments at the junctions.

Type I BFB events (Figures 5A and 5D) were similar to the one described above, with junction segments from the same chromosome. We identified eight type I BFB events in six cell lines (Table S4). Type II BFB events (Figures 5B and 5D) involved insertion of a segment from different chromosomes inserted during fusion process. We found two type II BFB events in two cell lines. Type III BFB events (Figures 5C and 5D) involved segments from the same chromosome that were rearranged before or during the fusion process. We detected two type III BFB events in two cell lines. One of the type III events affected the WT1 tumor suppressor in SNU-682 cells, causing a deletion of five exons and a loss of WT1 expression (Figure 5C). SNU-1000 cells have a complex rearrangement on chr11 involving HPV sequences.⁴² Then, we attempted to characterize the structure of the DNA centromeric to the BFB junctions. In some cases, we observed additional inversions and rearrangements, such as in SNU-703 chr11 (Figure S4A), while in other cases, we noticed a gradual decrease in read coverage. An HPV integration event was observed within 2 Mb of a BFB event in 2/13 instances (Table S4). These concordant events were in CaSki cells with at least 20 sites of HPV integration and the previously described complex amplification in SNU-1000 cells.⁴²

Canonical telomeres are not found at BFB sites

The internal architecture of the BFB amplifications is defined by breakpoints with foldback inversions and sequence digestions. However, it is still unclear how BFB amplifications affect the overall karyotype. We hypothesized that the derived chromosome would need to acquire a telomere to reach a stable state after one or several rounds of BFB amplifications.⁵⁷ To explore this hypothesis, we generated haplotype-specific coverage plots of the four cell lines with the highest sequencing coverage (CaSki, SNU-1000, SCC152, and HT-3). In all five BFB events detected in these cell lines, we observed reduced coverage on the telomeric side of BFB compared to the centromeric side. In CaSki, the telomeric side was also characterized by LOH (Figure 6D). This suggests that the chromosomal sequence directly after the BFB event on the telomeric side was lost in all cases. In three cases that did not have LOH but instead had reduced telomeric coverage, it is possible that the cells had three or more copies of the corresponding chromosomes before the BFB event, and at least two copies (from different haplotypes) were retained after BFB.

Figure 6.

Map of the chromosomal region 11q22 with a cluster of 3 BFB events

(A) Map of the chromosomal region 11q22 containing YAP1, BIRC3, and BIRC2 is shown along with the location of three BFB events in independent models. Detail of the read coverage aligned to HG38 is shown for the HNSCC PDX line (PDX-CK3489), SCC154 HNSCC cells, and CaSki cervical cancer cells. All soft-clipped reads are inverted in relation to the aligned forward (blue) and reverse (pink) read segments.

(B) Phase reads copy-number plot of the CaSki genome displaying reads assigned to haplotype 1 (HP-1) and 2 (HP-2) as well as unphased reads. Large blocks of unphased reads are due to loss of heterozygosity (LOH).

(C) Detail of chromosome 3 is shown indicating the location of multiple HPV integrations on 3q22–27 (131–188 Mb).

(D) Plot of chromosome 11 showing the phased amplification of the YAP1 region (BFB) and the deletion of one haplotype and LOH after the BFB event extending to the telomere.

We further attempted to localize the ends of derived chromosomes after BFB amplifications. We searched for reads containing telomere motifs and grouped them in 50-kb bins. This analysis identified telomeres at the expected locations (chromosome ends) but did not find any bins with more than three reads containing telomere sequences inside the BFB amplifications (Table S5). Analysis of clusters of read alignments with loose ends did not yield any rearrangements or abrupt sequence ends, either (other than the reported foldback inversions and several focal events). This suggests that the derived chromosomes did not acquire canonical telomere sequences or fusion with another chromosome.

YAP1 amplification is caused by BFB events

Four cell lines (SCC154, CaSki, SNU-1000, and SNU-1299) have BFB events and amplify the YAP1 oncogene and the BIRC3 and BIRC2 anti-apoptotic genes adjacent to each other on chromosomal region 11q22.1. Long-read WGS of an HNSCC patient-derived xenograft (PDX) also revealed amplification of YAP1, BIRC3, and BIRC2, and copy-number variant (CNV) loss of 11q distal to this region, and a type I BFB event (Figure 6A). Analysis of phased reads in the CaSki genome revealed complex chromosome gains and losses across the genome consistent with data from short-read WGS data (Figures 6B–6D).¹⁰

As YAP1 is a potent cervical cancer oncogene and amplification of this locus is the most common oncogene amplification in cervical cancer, we examined additional genomic datasets. Analysis of the 293 cervical cancers in TCGA confirmed that 31 (11%) have YAP1 amplification. Figure 7A shows the copy-number values for genes proximal and distal to YAP1. In every tumor, the amplification occurs in a 3–4-Mb region of chromosomal region 11q22 containing YAP1, BIRC3, and BIRC2. Distal to this region, the log2 CNV values are <1 for all genes on 11q (Figure 7A). This indicates that nearly all YAP1-BIRC3-BIRC2 amplifications in cervical squamous cell carcinoma (CESC) are due to BFB events. We performed similar analyses for 20 cervical tumors in the ICGC-PCAWG dataset, 103 cervical squamous cell carcinoma (SCC) tumors in the MSKCC metastasis study, 90 Guatemalan cervical tumors,^42,47 and 391 cervical SCCs in the AACR GENIE cohort⁵⁸ (https://www.cbioportal.org/) (Figure S5). In these studies, virtually all tumors with YAP1 CNV gain show a focal gain of the YAP1 region and loss of 11q22-tel. This result strongly suggests that deletion of chromosomal region 11q, followed by BFB cycles, leads to YAP1-BIRC3-BIRC2 amplification.

Figure 7.

Analysis of YAP1 amplification in 31 cervical tumors

(A) Copy-number data for genes centromeric and telomeric to YAP1 in all TCGA cervical tumors with Log2 copy-number values for YAP1 >1. The location of YAP1, BIRC3, and BIRC2 are highlighted. The location of each gene in Mb is given below the gene symbol.

(B) Comparison of the age of diagnosis between women with YAP1-amplified and YAP1-unamplified tumors. Results are from a two-tailed unpaired t test (t = 7.1, X² < 0.0001).

(C) Self-identified race of individuals with cervical cancer with and without YAP1 amplification (X² = 12.6, p = 0.0004). Data from TCGA, MSKCC metastatic cancer, and AACR GENIE cohort were obtained from cbioportal (https://www.cbioportal.org).^59,60

Within TCGA, MSKCC, and GENIE, the frequency of YAP1 gain in cervical cancer is the highest relative to all cancer types (Figure S6; Table S6). Interestingly, YAP1 gain is also high in other cancer types caused by HPV, such as HNSCC, penile, anal, and vulvar cancers (Figure S6). In addition, in these studies and the combined dataset, the median age of diagnosis of individuals with cervical cancer with YAP1 amplification is only 36 years old, 14 years younger than those without this CNV event (Figure 7B). These data indicate that YAP1 amplification leads to rapid cervical cancer development and progression. Finally, YAP1 amplification is significantly (X² = 12.6, p = 0.0004) more frequent in cervical tumors from African, Asian, and other women of non-European ancestry (Figure 7C). Therefore, targeted therapy for the cervical cancer subtype with an amplified YAP1 region could help reduce cervical cancer health disparities.

Discussion

We performed a multi-omics analysis of the cervical and head and neck cancer cell lines, including complete HPV sequencing, HPV integration analysis, copy-number alterations, HLA gene sequencing, and human and HPV gene expression. The main goal of this project was to characterize a panel of HPV+ cervical and head and neck cancer cell lines using long-read methodology to further understand HPV integration events and genomic rearrangements. Cervical cancer is under-represented in major international cell-line databases such as the NCI-60 panel (0 cervical lines), COSMIC (19 lines), and CCLE (14 lines) (Table S7). Our panel of 22 cell lines includes 18 cervical cancer cell lines, 10 of which are not represented in any current cell-line panel (Table S7); this includes the major cervical cancer (CESC) histological types (SCC, adenocarcinoma, and small cell carcinoma) as well as a comparison group of HPV+ HNSCC cell lines. This panel of cell-line origins are mainly from individuals of European and Asian descent with one cell line being of African American descent (HeLa). We document the diversity of HPV types, HPV sublineages, and integration loci. Long-read DNA sequencing with ONT platforms was used for long-range characterization of the HPV integration loci. Finally, full-length RNA sequencing was performed using direct cDNA and direct RNA approaches for transcriptome analysis.

Cervical cancer disproportionally affects women living in poverty, especially in Sub-Saharan Africa, where it causes about 50,000 deaths every year.⁶¹ African American women also face a higher risk of cervical cancer mortality, with a two-fold increase in age-adjusted death rates compared to other ethnic groups in the United States. We found that YAP1-BIRC3-BIRC2 amplification, which is associated with earlier onset of cervical cancer, is three times more prevalent in African American women with cervical cancer. This finding suggests that this genomic alteration (YAP1-BIRC3-BIRC2 amplification) contributes to the ethnic disparity in cervical cancer mortality. Cohen et al. document that age-adjusted mortality from CESC-SCC is significantly higher (1.9-fold) in African Americans in the U.S.⁶² YAP1 overexpression in cervical cells activates EGFR and its ligands, which are potential therapeutic targets for EGFR inhibitors.⁶³ In addition, YAP1-targeted agents are currently being tested in clinical trials.⁶⁴

HPV infection and integration can cause genome instability in HPV-driven cancers, leading to genome rearrangements and the formation of extrachromosomal HPV/human hybrids.^65,66 However, little is known about other copy-number alterations in cervical cancer. In this study, we used long-read sequencing to analyze cancer cell lines. A single BFB event was observed in six cell lines, and BFB events on more than one chromosome were identified in three cell lines. Although BFB events are common mechanism of genome instability in cancer cell genomes, few studies have characterized the sequence features of the inversion sites involved in BFB cycles. We found that the breakage ends underwent exonuclease digestion and that the fusion sites showed DNA insertions and rearrangements. We categorized three types of BFB events: type I (with no insertions or rearrangements), type II (with insertion from another chromosome), and type III (with additional local rearrangement). Long-read sequencing was crucial in determining the precise structure of the insertions and rearrangements at BFB junctions. Five cell lines showed YAP1 amplification associated with BFB, making this one of the most common driver events in the cell-line panel. In addition, amplification of the YAP1-BIRC3-BIRC2 loci was one of the most frequent copy-number alterations in cervical tumors in the TCGA, MSKCC metastatic cancer, and AACR GENIE cohorts (Figure S6).^46,58 Furthermore, we found genes inactivated by BFB, including the WT1 tumor suppressor. Therefore, BFB events are essential in the progression of cervical cancer through the activating or inactivating of essential cancer driver genes.

YAP1 is an oncogene that is amplified in 15% of CESC tumors,⁴⁶ and overexpression of YAP1 in mouse cervical cells can cause cervical carcinogenesis without HPV infection.⁶⁷ Furthermore, YAP1 expression repressed differentiation of HPV-infected basal cells in the cervix.⁵⁶ Our data strongly suggest that nearly all YAP1 amplification is the consequence of BFB events. We found that YAP1 amplification is associated with a 10–14-year-earlier age of diagnosis of cervical cancer. In addition, previous immunohistochemistry (IHC) studies showed that the YAP1 protein is high in both high-grade pre-cancer and cervical cancer.⁵⁶ This result suggests that YAP1 amplification is both an early event in cervical cancer and leads to the rapid progression of the disease. Furthermore, we consistently found BIRC2 and BIRC3 amplified along with YAP1, demonstrating that these genes also play a crucial role in cancer progression. High levels of BIRC2 and BIRC3 could contribute to cervical cancer progression by conferring resistance to T cell-mediated cell death.⁶⁸

Our analysis of the telomeric side of BFB amplifications suggests that the derived chromosome did not acquire a canonical telomere sequence, nor was it fused with another chromosome. Therefore, the exact mechanisms of stabilizing the derived chromosome after BFB remain unclear. It has been hypothesized that the chromosome may remain in a double-strand break state after BFB, and our observations are consistent with this explanation.⁶⁹ We expect that additional long-read sequencing of diverse tumors at higher depth will provide further information on this critical class of chromosome alteration.

HPV infections and early pre-cancerous lesions are often rapidly cleared, indicating a robust immune response to virally infected cells.¹ Furthermore, genome-wide association studies of cervical cancer reveal strong associations with the MHC on chromosomal region 6p, principally to HLA class II and HLA class I genes.⁷⁰ In addition, HLA-A and HLA-B are frequently mutated in CESC (unpublished data). We found that 56%–61% of 19 CESC and HNSCC cell lines are homozygous across the HLA class I and II genes, significantly higher than the homozygosity in cell lines of other cancer types, supporting a key role for the MHC in cervical cancer progression.

Currently, there is no targeted therapy for cervical cancer. However, PIK3CA is the most common oncogene driver mutation, which is present in six out of 21 of our cell lines. BYL719/Alpelisib is a specific phosphoinositide 3-kinase (PI3K) inhibitor approved for treating a subset of breast cancers.⁷¹ Preclinical data show that BYL719 inhibits cell lines with PIK3CA mutations (unpublished data). YAP1 amplification is also frequent in cervical cancer and is largely mutually exclusive of PIK3CA mutation (unpublished data). Therefore, YAP1-amplified tumors represent a second cervical cancer subtype potentially amenable to targeted therapy. Our cell panel provides model systems to study targeted inhibitors and immune therapies and advance the therapeutic options for cervical cancer.

Our cell lines confirm the distribution of HPV integration across many human chromosomes, but 80% of integrations are in known HPV integration hotspots or near super-enhancers.⁵³ However, most cell lines have integration in or near a gene of potential oncogenic importance and activate the expression of at least one gene at the locus. As HPV brings two potent oncogenes to the cell, driver gene mutations are less prevalent in cervical cancer. In our cell lines, there was an increase in mutations in cancer driver genes in cell lines without HPV or with unknown-risk HPV types. Mutations in TP53, RB1, and other cancer driver genes may substitute for the absence or low activity forms of the E6/E7 oncogenes. For example, the cell line C33A is HPV negative (we confirmed this by long-read sequencing) and is homozygous for pathogenic mutations in both TP53 and RB1. HT-3 has an integrated copy of HPV30, an unknown-risk type, and has mutations in RB1 and TP53 that may potentiate the E6 and E7 proteins of HPV30.

Our study has several limitations, such as the under-representation of cell lines derived from adenocarcinomas, an aggressive subtype of cervical cancer, and only one cell line (HeLa) from a woman of African origin. There are likely unknown biases in what classes of cervical tumors were established as cell lines. Episomal-only tumors and cell lines are known to be unstable, and only SNU-1000 retains extrachromosomal HPV16. We described only 18 CESC cell lines, but 10 are unique to this panel (Table S7). For several cell lines, limited coverage did not allow the detection of SNVs genome-wide or the identification of all chromosome alterations, including BFB events. Caution must be taken in extending results from cell lines to in vivo tumors, and these results need to be extended to DNA obtained directly from cancer tissue.

In summary, using a panel of 22 cervical and HPV-positive cell lines and long-read sequencing, we comprehensively characterized the structure of HPV integrations and the consequences on gene expression. We show that HPV-negative and non-hrHPV-type cell lines have more driver mutations, including in TP53 and RB1, to compensate for the low or absent activity of HPV E6 and E7. In addition, we characterized the sequences at the inversion junctions of BFB events and provided new insight into the formation of these critical chromosome rearrangements in cancer. These cell lines can serve as models for specialized treatments of cervical cancer.

Data and code availability

The code to run Severus is available at https://github.com/KolmogorovLab/Severus. DNA sequences of cell lines have been deposited at the Sequence Read Archive under bioproject: PRNA772772 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA772772).

Published: February 1, 2024