Genomic landscape of primary and recurrent anal squamous cell carcinomas in relation to HPV integration, copy number variation, and DNA damage response genes

Abstract

The incidence of anal squamous cell carcinoma (ASCC) has been increasing, particularly in populations with HIV. HPV is the causal factor in 85-90% of ASCCs, but few studies evaluated HPV genotypes and integrations in relation to genomic alterations in ASCC. Using whole-exome sequence data for primary (n=56) and recurrent (n=31) ASCC from 72 patients, we detected HPV DNA in 87.5% of ASCC, of which HPV-16, HPV-18, and HPV-6 were detected in 56%, 22%, and 33% of HIV-positive (n=9) compared to 83%, 3.2%, and 1.6% of HIV-negative cases (n=63), respectively. Recurrent copy number variations (CNVs) involving genes with documented roles in cancer included amplification of PI3KCA and deletion of APC in primary and recurrent tumors; amplifications of CCND1, MYC, and NOTCH1 and deletions of BRCA2 and RB1 in primary tumors; and deletions of ATR, FANCD2, and FHIT in recurrent tumors. DNA damage response genes were enriched among recurrently deleted genes in recurrent ASCCs (p=0.001). HPV integrations were detected in 29 of 76 (38%) ASCCs and were more frequent in stage III-IV versus stage I-II tumors. HPV integrations were detected near MYC and CCND1 amplifications and recurrent targets included NFI and MUC genes. These results suggest HPV genotypes in ASCC differ by HIV status, HPV integration is associated with ASCC progression, and DNA damage response genes are commonly disrupted in recurrent ASCCs.

Implications:

These data provide the largest whole-exome sequencing study of the ASCC genomic landscape to date and identify HPV genotypes, integrations, and recurrent CNVs in primary or recurrent ASCCs.

Introduction

High-risk human papillomavirus (HPV) types cause 85-90% of anogenital tract squamous cell carcinomas (SCCs), and 15-40% of head and neck SCCs (HNSCCs) (1-3). The incidence of anal SCC (ASCC) has been increasing globally (4-6), and high-risk HPV types are the etiological agent in 85%-90% (3, 6, 7). ASCC is more common among women, but the incidence has been increasing among men who have sex with men, particularly those with HIV infection (6, 7). HPV-positive ASCCs exhibit fewer deleterious genetic alterations and better treatment responses compared to HPV-negative ASCCs (8). Standard treatment is chemoradiation (5-fluorouracil and mitomycin C or cisplatin with concurrent ionizing radiation) and 5-year survival rates for localized disease are 75-85% (6, 8). Tumor recurrences (20%-30%) represent a genetically distinct class, and 5-year survival rate is 15-20% following a second recurrence or metastasis (6, 9-11).

HPV is an oncogenic DNA virus that infects mucosal or cutaneous stratified epithelium and induces cellular transformation by expressing viral oncoproteins E6 and E7, integrating into the genome, and inducing chromosomal instability. The HPV genome is maintained as an extrachromosomal episome and viral replication proceeds as cells differentiate and migrate to the epithelial surface. Expression of E1^4 (E4), which is highly expressed in differentiating transit-amplifying cells, alters host cell keratin network structure, induces G2 arrest, and stimulates E1 and E2 dependent viral genome amplification (12). E6 and E7 promote malignant transformation by inactivating tumor suppressors TP53 and RB1, respectively. These effects can be augmented by E2 disruption during integration of HPV DNA into the host genome, which leads to increased E6-E7 expression and thereby promotes malignant progression (13, 14). However, there is a long latency period between HPV infection and cancer development because additional mutations in host cell genes are required to cause HPV-associated cancers in vivo.

HPV integration into the host genome can impact function or expression of tumor suppressors and oncogenes, frequently by inducing copy number variations (CNVs), as demonstrated for oncogenes such as TP63 and MYC (15-18). HPV integration is associated with malignant progression (19), and integrated HPV is detected more frequently in ASCCs compared to premalignant anal lesions (13). HPV types vary in their propensity for oncogenesis and integration into the human genome. High-risk HPV-16 causes the majority of anogenital cancers, followed by high-risk HPV-18, while low-risk HPV-6 and HPV-11 are rarely associated with these cancers (3). Among HPV-16 and HPV-18-positive anogenital cancers, HPV-16 integration is detected in 60-80% and HPV-18 integration is detected in 92-100% (20, 21), respectively.

The prevalence of HPV genotypes differs by anatomical location, and focal genomic alterations associated with HPV integration vary by cancer type (22). Although focal genomic alterations and their relationship to HPV integrations have been characterized in cervical and HNSCC, few studies have evaluated focal genomic alterations in relation to HPV infection and integration in ASCC. HIV-positive individuals have increased risk of HPV infection and ASCC in comparison to HIV-negative individuals (7, 23), and higher rates of malignant progression of neoplastic anal lesions (5, 23), yet little is known about ASCC genomes in HIV-positive compared to HIV-negative patients. Here, we characterize the virome and virus-host genomic interactions in 88 ASCCs from 72 HIV-negative and HIV-positive individuals using whole exome sequencing (WXS) data to identify sites of HPV integration in the host genome and their relationship to recurrent CNVs in primary and recurrent ASCCs.

Materials and Methods

Whole-exome data processing and metagenomic analysis

WXS data was collected from three publications (11, 24, 25). Data from 46 patients in Mouw et al., 2016, and 5 patients in Shin et al., 2018 was provided by the authors (11, 25) and data from 21 patients in Cacheux et al., 2018 (24), was downloaded from the NCBI Sequence Read Archive (SRA: SRP119025, https://www.ncbi.nlm.nih.gov//bioproject/PRJNA412145). Written informed consent from the patients was obtained and studies were conducted in accordance with ethical guidelines in the Declaration of Helsinki and approved by the respective Institutional Review Boards at Dana-Farber Cancer Institute, Catholic University of Korea College of Medicine, and Institut Curie. All three studies used Agilent SureSelect Exon Capture (versions 2, 4, and 5 for Mouw et al., Shin et al., and Cacheux et al., respectively) and Illumina HiSeq sequencing. Raw FASTQ sequence files were trimmed of adapter sequences using Trimmomatic and aligned to the human genome (GRCh38.p12) using BWA (v0.7.15) with the following non-default parameters: Seed length (k) = 25, mismatch penalty (B) = 10, gap open penalty (O) = 15, gap extend penalty (E) = 3. Alignments were deduplicated using Samtools’ (v1.9) markdup function and all unmapped, discordant (reads with only one mate mapped, without proper 5` to 3` orientation of pairs, or with unexpected insert size), clipped (reads only partially aligning to the reference), or reads with more than 3 mismatches from the reference were classified as potentially non-human. Low-complexity sequences were masked using the NCBI BLAST (v2.6.0) dustmasker function and reads containing more than 33% masked nucleotides were filtered out. The remaining reads were screened against the Kraken (v1.0) viral kmer database and filtered using Kraken’s filter function set at 0.25. The number of reads for any HPV genotype in a given sample was divided by total input reads to generate normalized HPV reads/million.

HPV integration site analysis

HPV-associated reads identified by Kraken and their pair-mates from a given sample were aligned to a composite reference consisting of the human genome and either HPV-16 (NC_001526.4), HPV-18 (NC_001357.1), or HPV-6 (FR751330.1), depending on the genotype. Discordant and/or chimeric reads aligning to both human and HPV genomes were extracted and mapped to genes using Samtools, Bedtools (v2.27.1), and Bedops (v2.4.30). A custom script was developed to search for additional chimeric reads at junctions identified in the original human alignment files. Integration sites with at least 4 reads spanning a junction were confirmed using Integrated Genome Viewer (IGV, v2.4.9).

HPV genome coverage and de novo assembly of HPV integrants

Coverage of HPV genomes in samples with evidence of integration was quantified and displayed using IGV. For each HPV integration site identified, sequence reads within 20 kb on either side of the human-HPV junction were extracted from the human alignment files, concatenated with all HPV reads from the same sample, and used as the input for generating de novo assemblies using Velvet (v1.2.10). HPV genes within assembled contigs and human-HPV hybrid contigs were identified by screening sequences against the NCBI “Others (nr etc.)” nucleotide BLAST database. Read coverage and Human-HPV junctions of assembled contigs were visualized using Tablet (v1.17.08.17).

Copy number analysis

Coverage counts in 1 kb windows of the human genome were generated for sample alignments using Bedtools’ makewindows and coverage functions. For estimation of tumor purity, minor allele frequencies were estimated from single nucleotide polymorphism (SNP) counts with quality scores of at least 20 generated by FreeBayes (v1.2.0-4-gd15209e). Inputs for the following CNV analysis were generated by a custom R (v3.5.1) function for each tumor sample in comparison to a synthetic pooled normal sample (n=2-6 total normal samples) matched for the same study and sample type (frozen or formaldehyde-fixed paraffin embedded (FFPE)) and restricted to the corresponding version of Agilent SureSelect All Exon targets. The R package SynthEx (https://github.com/ChenMengjie/SynthEx) was used to generate the appropriate synthetic normal comparator, estimate coverage ratios, and normalize by tumor purity. The R package DNAcopy (v1.56.0) was used for data smoothing (non-default parameters: smooth.region = 10, outlier.SD.scale = 4 , smooth.SD.scale = 3) and performing circular binary segmentation (non-default parameters: alpha=0.01, undo.splits="sdundo", undo.SD=2). Segmented sample data was concatenated and uploaded to the GISTIC_2.0 (v6.15.28) module on the GenePattern server (https://cloud.genepattern.org) to assess recurrent CNVs across all tumor samples (7 samples exhibiting hyper-segmentation were excluded from the analysis by setting the “max sample segs” parameter to 1000). To reduce false positives from common germ-line variants, each normal sample was compared to the corresponding synthetic normal and significant recurrent CNVs (n=165) found in GISTIC2.0 analysis (q=0.25, amplitude ≥0.3) of normal samples were filtered out of the tumor dataset. For individual comparisons and plotting, additional GC content and tumor ploidy normalization using the “closest” method were performed using the R package Canner (v1.28.0). GC content information and Agilent SureSelect All Exon V2 and V4 target coordinates were updated from hg19 to hg38.p12 using liftOver and Bedops.

Gene Enrichment Analysis

A curated list of 981 cancer-related genes was assembled from the COSMIC database of cancer-related genes (724 genes, downloaded March 6, 2019 from https://cancer.sanger.ac.uk/census) and five studies of cancer drivers and DNA damage response genes (257 additional genes) (26-30). Genes within CNVs classified as recurrently amplified or deleted were analyzed for functional enrichment by screening against the gene universe of 981 cancer-related genes in the curated list and their mapped ontological categories. Among the curated gene list, 902 Entrez gene IDs were mapped to ontological categories in the org.Hs.eg.db database (ftp://ftp.geneontology.org/pub/go/godatabase/archive/latest-lite/) using the topGO R package, and enrichment analysis was performed using the limma package in R with false discovery rate (FDR) controlled at 5% using the Benjamini-Hochberg method.

Results

Samples and clinical data

The study included 72 patients with 88 ASCC tumor samples, including 56 primary tumors, 27 recurrent tumors, 4 metastases, and 1 uncategorized. WXS data from these samples was from three published studies conducted in the United States, France, and South Korea (11, 24, 25). All three studies used similar exome capture and sequencing technology (average depth of coverage reported in each study was 87.1-107.3X, 119X, and 134.9X, respectively). WXS data from matched normal samples was available for 38 patients. Clinical and demographic characteristics of the cohort are shown in Table 1. The median age was 59.5 years (IQR 52.5-65.2), 51 (70.8%) were female, and 9 (12.5%) were HIV-positive. Among 24 samples with available clinical tests for HPV genotype, HPV-16 was detected in 22 (91.6%), while HPV-18/6 and HPV-6 were each detected in 1 (4.2%). Compared to HIV-negative patients, HIV-positive patients were younger and more likely to be male (median [IQR] age 47 [45-58] vs. 61 [54-66] and 88.9% vs. 20.6% male for HIV-positive (n = 9) vs. HIV-negative (n = 63) patients, respectively). HIV-positive patients were also more likely to be diagnosed with lower stage tumors (66.7% vs. 31.8% stage I or II, respectively) or with tumor recurrence (66.7% vs. 39.6% non-recurrence, respectively), although these differences did not reach statistical significance (Fisher’s exact test, p=0.063 and p=0.16, respectively (Supplemental Table S1).

Table 1.

Demographic and clinical characteristics of cohort

Characteristics	ASCC Patients (n=72)
Age, median [IQR]	59.5 [52.5-65.2]
Gender
Female	51 (70.8)
Male	21 (29.2)
HIV status
Positive	9 (12.5)
Negative	63 (87.5)
HPV status*
Positive	62 (86.1)
HPV-16	22 (30.6)
HPV-18/6	1 (1.4)
HPV-6	1 (1.4)
Unknown	38 (52.8)
Negative	8 (11.1)
Unknown	2 (2.8)
Tumor Stage
I	1 (1.4)
II	25 (34.7)
III	32 (44.4)
IV	7 (9.7)
Unknown	7 (9.7)
Patients with recurrence	31 (43.1)
Patients with metastasis	5 (6.9)
Initial Therapy
Chemoradiation	61 (84.7)
Surgery/radiation	3 (4.2)
Radiation alone	1 (1.4)
Surgery alone	1 (1.4)
Unknown	6 (8.3)
Sample Type
Frozen	23 (31.9)
FFPE	49 (68.1)
Patients with matched normal	38 (52.8)

All data are n (%) unless otherwise indicated. Abbreviations: ASCC, Anal Squamous Cell Carcinoma; FFPE, Formaldehyde Fixed Paraffin Embedded

^*HPV status determined by p16 IHC, HPV FISH, or PCR

Analysis of the ASCC virome

To detect HPV and other viral sequences, HPV integration sites, and CNVs in ASCC WXS data, we developed a custom NGS and bioinformatic pipeline (Figure 1). After subtracting human reads and filtering low-complexity reads, the remaining unmapped reads were screened against the Kraken viral kmer database. Reads assigned to a viral taxon were then realigned to a composite reference of the human genome and virus of interest to identify breakpoint junctions. Integration events were defined by junctions mapped between HPV and human genomic sequences with at least 4 total reads including 1 chimeric HPV-human read spanning the breakpoint. Analysis of the tumor virome of 72 patients detected between 0.016 to 114 HPV reads/million in ASCCs from 63 (87.5%), of which HPV-16 was detected in 57 (90.5% of patients with HPV detected in WXS data), while HPV-18 and HPV-6 were each detected in 4 patients (6.3%) (Supplemental Table S2). Overall, we detected HPV sequences in ASCC WXS samples from 58 of 62 (93.5%) patients with a positive clinical HPV test, 4 patients with a negative clinical HPV test, and in normal tissue from one patient with no clinical HPV test data. Among 38 pairs of matched tumor and normal samples, HPV was detected in WXS data from 33 (86.8%) tumor samples in comparison to only 6 (15.8%) normal samples. Among the 6 normal samples positive for HPV, the number of detected HPV reads was on average < 1% of the number of reads found in tumors (between 0.0176 and 0.243 reads/million (Figure 2A and Supplemental Table S2)). Among tumor samples positive for HPV, we did not detect statistically significant differences in number of detected HPV reads/million between late stage vs. early stage tumors, primary vs. recurrent tumors, or primary tumors that did vs. did not recur (Mann-Whitney test, all p-values >0.30).

Figure 1. Overview of study design.

Raw FASTQ reads were trimmed of adapter sequences, deduplicated, and aligned to the human genome (GRChg38.p12). Aligned reads were used to generate CNV profiles in tumor samples after smoothing, denoising, and comparison to pooled synthetic normal profiles where no matched normal was available. Unmapped reads were filtered of low-complexity sequences and screened against Kraken’s viral kmer database. Reads assigned by Kraken to HPV taxa were further analyzed for breakpoints between human and HPV chromosomal sequences by realignment to human and HPV composite genomes.

Figure 2. Detection of viral sequences and junction mapping in ASCC.

A. Whole-exome virome analysis of 88 tumor samples and 38 matched normal samples. Columns represent individual samples. Histograms (top) shows total normalized reads/million for HPV genotypes detected in individual samples. Histograms to the left of each plot indicate total proportion of samples positive for each virus. HPV integrations were identified by mapping at least 4 or more reads with at least one chimeric HPV/human read spanning the junction. Clinical information, including results of HPV testing by p16 staining, FISH, or PCR, is displayed at the bottom. HIV status included some individuals classified as “presumed HIV-negative” on the basis of clinical history, immune profiles, and no risk factors. B. Circos plots representing positions of junctions between HPV and human genomic sequences. Lines represent integration breakpoints mapped between the human genome and either HPV-16 (left, 26 tumors from 23 patients) or HPV-6 (right, 2 tumors from 1 patient) with at least 4 reads and 1 chimeric HPV/human read spanning the junction (genomes not shown to scale).

Next, we evaluated HPV genotypes in ASCC in relation to HIV status. HPV-16, HPV-18, and HPV-6 were detected in 82.5%, 3.2%, and 1.6% of HIV-negative patients vs. 55.6%, 22.2%, and 33.3% of HIV-positive patients, with non-HPV-16 genotypes detected more frequently in ASCCs from HIV-positive vs. HIV-negative patients (4 of 9 (44.4%) vs. 3 of 63 (4.8%) patients, respectively; Fisher’s exact test, p=0.0036) (Supplemental Table S1). A single HPV genotype was detected in all but one tumor from an HIV-positive patient, in which both HPV-16 and HPV-6 were detected in the primary tumor and only HPV-18 was detected in the recurrent tumor (Figure 2A and Supplemental Table S2). Authors of the original study describing primary and recurrent tumors from this patient noted they did not have a similar profile of mutations and suggested the recurrence may be a second primary tumor (11); our WXS analysis is consistent with this scenario.

In addition to HPV, we detected several herpesviruses in ASCC samples, including alphaherpesvirus 1 (HSV-1), betaherpesvirus 5 (CMV), human herpesvirus type 6 (HHV-6B) (Roseola virus), and gammaherpesvirus 4 (EBV/HHV-4) (Figure 2A). However, the number of reads assigned to these viruses was low in comparison to HPV, with only 1-5 total reads detected in tumor or normal samples. Among normal tissue samples, only HIV-positive patients had reads assigned to herpesviruses.

Analysis of HPV integration events

Given that rates of HPV integration in tumors differ by HPV genotype and anatomical location, we next evaluated HPV integration sites in ASCC tumors. HPV integration events were detected in 29 of 76 HPV-positive ASCC samples, of which 26 were HPV-16 positive, two were HPV-6 positive, and one was HPV-18 positive (corresponding to 39%, 33%, and 20% of samples in which HPV-16, HPV-6, or HPV-18 were detected in WXS, respectively); one tumor was positive for both HPV-16 and HPV-6. As expected, HPV integrations were detected more frequently in samples with higher overall HPV genome coverage (median reads/million = 9.71 in samples with integrations (n=29) vs. 0.54 without integrations (n=47), Figure 2A). Positions of 88 breakpoints mapped between HPV-16 or HPV-6 and the human genome with at least 4-reads supporting a junction are shown in Figure 2B.

Next, we evaluated which elements of the HPV genome were detectable in ASCC tumors. The majority of HPV positive samples had linear read coverage for most of the HPV genome (Figure 3A and Supplemental Figure S1), with greatest read depth in E4 (which overlaps E2), E7, and E6 genes in samples with HPV-16 (Figure 3B). Integration site breakpoints were distributed across the HPV genome, with integration sites most commonly occurring within E4, L1, L2, and E2. Similar patterns were observed for HPV-16 and HPV-6 (Figure 3A and 3B). Two patients with HPV-6 had two tumor samples each (primary and recurrence, or recurrence and metastasis). For the primary-recurrence pair, no HPV integrations were detected in either sample, while for the recurrence-metastasis pair, the same set of integrations was detected in each tumor and additional integrations were detected in the metastasis.

Figure 3. Read coverage and integration sites of HPV-16 and HPV-6 in ASCC.

A. IGV output showing depth of read coverage at sites within HPV-16 (top) and HPV-6 (bottom) for selected patients with the highest total genome coverage. Colored bars in histograms represent variations from reference sequence. Cyan flags mark the position of HPV/human breakpoints. B. Quantification of average read depth for HPV-16 genes, normalized to gene length.

To further characterize HPV integrations, we performed de novo assembly of human and HPV-derived reads to resolve regions surrounding breakpoints. In two tumors, assembly resulted in one 2,704 bp contig and one 8,330 bp contig that contained both flanking human DNA sequence and integrated HPV genomic sequence. In the first case, an HPV integration containing both E6 and E7 was mapped to a position 135 kb upstream of the MYC oncogene (Figure 4A, top). In the second case, an integration containing partial sequences of L1 and L2 was mapped within the 3^rd intron of the NFIX transcription factor gene (Figure 4A, bottom). In most cases, assembly did not generate chimeric human-HPV contigs; most contigs were from one genomic source or the other. With respect to HPV, the most common contigs contained either E6 and/or E7 or mapped within L2-L1 (Figure 4B).

Figure 4. De novo assembly maps integration of HPV genes into NFIX and MYC.

A. Genomic mapping of 2 ASCC cases with HPV integration near MYC and within NFIX. Top: Human genomic region surrounding integration sites (coordinates marked by red flags). Human reads within the region marked by the blue bars were used for de novo assembly. Middle: chimeric contigs containing both human and HPV sequences were assembled using Velvet. Read depth along each contig is displayed. Bottom: All reads mapped to the HPV genome were used for de novo assembly (red bar). Light blue and red regions indicate sections of the assembled contig derived from human and HPV sequences, respectively. B. HPV-16 genome assembled contigs. All contigs greater than 300 bp in length generated by Velvet assembly of HPV-16 reads from the indicated ASCC cases are displayed corresponding to HPV-16 genome map as in A. Green bars represent contigs containing E6 and/or E7, purple bars represent contigs containing sequences from L1 and/or L2.

Regions targeted by HPV integration include cancer-related genes

Next, we characterized regions of the human genome containing HPV integration sites. A summary of integration sites with the highest numbers of reads spanning breakpoints (≥6) and their associated chromosomal locations and genomic features is shown in Table 2. HPV integrations were detected in or near several genes with documented functional roles in cancer and classified as tier 1 in the COSMIC database including MYC, MET, EML4, and NCOR1 (Table 2 and Supplemental Table S2). Additionally, HPV was recurrently integrated into MUC12 (6 tumors from 5 patients) and MUC1 (1 tumor from 1 patient), HRNR (3 tumors from 2 patients), GOLGA6L22/HERC2P2 (3 tumors from 2 patients), REXO1L pseudogene locus (2 tumors from 2 patients), and GAGE12B/F (2 tumors from 2 patients). The integration with the most reads covering the junction was found in an HIV-negative patient with both recurrent and metastatic tumor samples bearing integrations of HPV-6, a low-risk HPV genotype rarely linked to oncogenesis (31, 32). In this patient, HPV-6 integrations in the CCSER1, HRNR, MUC12, and GOLGA6L22 genes were mapped at the same breakpoint junctions in both tumor samples (Supplemental Table S2). Recurrent integrations were identified within some gene families including Nuclear Factor I family members NFIX and NFIA, A Disintegrin and Metalloproteinase with Thrombospondin family members ADAMTS6 and ADAMTS20, NOD-Like Receptor Protein genes NLRP1 and NLRP11, WASP-family Homologs WASHC2A and WASH7P (1 tumor each from 2 HIV-negative patients), and Neuroblastoma Breakpoint Fusion family members NBPF14 and NBPF20 (tumors from 2 HIV-positive patients), raising the possibility that related genes within these families could play a role in viral infection or might be selected for in tumors.

Table 2.

Summary of all HPV-human junctions with 6 or more reads spanning the breakpoint.

Patient	Age	Gender	HIV Status	Tumor Type	Tumor Stage	HPV Type	Reads at Junction	Chromosome Location	Gene
1	70	Female	Negative	Metastasis & Recurrent	III	HPV-6	154	chr4:90883274	CCSER1
				Metastasis & Recurrent			14	chr1:152216110	HRNR
				Metastasis & Recurrent			8	chr7:100999089/100999381/101004040	MUC12
				Metastasis & Recurrent			8	chr15:22465194	GOLGA6L22
				Metastasis			6	chrX:7843725	VCX
				Metastasis			6	chr7:116739848/116757490	MET
				Recurrent			6	chr10:46822289	LOC107984026
2	55	Female	Negative	Primary	IV	HPV-16	144	chr21:36744883/36745756	SIM2
				Primary			74	chrX:2293227/2293261	DHRSX
				Primary			6	chr18:13789515	[RNMT+24Kb]
3	53	Female	Negative	Primary	UNK	HPV-16	74	chr11:67417261	CARNS1
4	61	Female	Negative	Primary	III	HPV-16	42	chr8:127563720/127573357/127574389	[MYC-135Kb]
				Primary			20	chr6:143511477	FUCA2
5	72	Female	Negative	Primary	UNK	HPV-16	36	chr19:13037379/13047737	NFIX
6	70	Female	Negative	Primary	UNK	HPV-16	36	chr13:66304840	PCDH9
7	66	Female	Negative	Primary	III	HPV-16	34	chr2:42301261	EML4
				Primary			14	chr3:47661389	SMARCC1
				Primary			6	chr17:16075501/16075530/16075558	NCOR1
				Primary			6	chr1:37887353	INPP5B
				Primary			6	chr3:114295610	TIGIT
8	45	Male	Positive	Primary	II	HPV-16	12	chr1:152213926/152215378/152215981/152220199	HRNR
9	54	Female	Negative	Primary	III	HPV-16	10	chr8:85661955/85744974/85815197	REXO1L1P/LOC101929627/REXO1L5
				Primary			6	chr7:100997821	MUC12
10	49	Female	Negative	Recurrent	III	HPV-16	8	chr8:85802820/85803558	REXO1L4P
11	54	Male	Negative	Recurrent	III	HPV-16	8	chr5:70441454/70907766	GTF2H2B/SERF1A
				Recurrent			6	chr12:110027733	ANKRD13A
12	53	Female	Negative	Primary	II	HPV-16	6	chr15:22523457	HERC2P2
13	68	Female	Negative	Primary	II	HPV-16	6	chr1:155188351	MUC1
14	33	Male	Positive	Recurrent	II	HPV-18	6	chr1:148585206	NBPF14
15	65	Female	Negative	Recurrent	UNK	HPV-16	6	chr17:46929530	GOSR2
16	58	Female	Negative	Recurrent	IV	HPV-16	6	chr7:100996064/100998690/101000735	MUC12
17	63	Female	Negative	Recurrent	III	HPV-16	6	chr1:61094894	NFIA
18	57	Female	Negative	Primary	II	HPV-16	6	chr5:157757038	LSM11

Bold genes are cancer-related genes from the COSMIC Database; Kauffman et al., 2008; Vogelstein et al., 2013; Chae et al., 2016; Bailey et al., 2018; and Knijnenburg et al., 2018

[ ] Integration outside of known genes with closest gene and distance (+ for downstream, - for upstream) in bracket. Abbreviations: UNK, unknown.

Relationship between HPV integration and recurrent copy number alterations

Next, we evaluated whether HPV-integration was associated with focal genome amplifications or deletions. To characterize recurrent CNVs, we used GISTIC2.0 analysis for all tumors combined (Supplemental Figure S2) or stratified into primary and recurrent or metastatic (Figure 5A). In primary tumors (n=49 after eliminating 7 samples with hyper-segmented profiles), tumor suppressor genes (TSGs) BRCA2, FOXO1, and RB1 were recurrently deleted within a common loss on 13q14 (28.6%), while oncogenes CCND1 (20.4%), MYC (36.7%), and NOTCH1 (28.6%) were recurrently amplified. In recurrent or metastatic tumors (n=30), TSGs ATR (10%), FANCD2 (13.3%), and FHIT (20%) were recurrently deleted, while TERT was recurrently amplified (13.3%). CNVs on chromosome 3 were common, including recurrent deletion of TSGs MLH1 and SETD2 within losses on 3p in both primary and recurrent or metastatic tumors (20.4% of primary and 20.0% of recurrent or metastatic tumors), and recurrent amplification of PIK3CA on 3q (61.2% of primary and 33.3% of recurrent or metastatic tumors). Recurrent deletion of tumor suppressor APC was detected in both primary (12.2%) and recurrent or metastatic tumors (6.7%). Among paired primary and recurrent tumors available from 10 patients in Mouw et al. (11), MYC and PIK3CA amplifications and APC deletions were shared in 40%, 20%, and 20%, respectively, while CCND1 and TERT amplifications and RB1 deletions were each shared in 10% of paired samples, indicating that putative driver mutations were maintained.

Figure 5. Copy number variations in primary and recurrent/metastatic ASCC tumors and association with HPV integration sites.

A. GISTIC2.0 output showing recurrent copy number deletions (blue) and amplifications (red) in primary (left) and recurrent/metastatic ASCC tumors (right). Selected genes from the curated list of cancer-related genes (Supplemental Table S3) are shown within genomic alterations that include deletion of tumor suppressors or amplification of oncogenes. Events shared between primary and recurrent tumors are shown in bold, and genes which contain HPV integration events are shown in red. B. Selected examples of copy number amplifications closely associated with HPV integration sites. Plots show whole chromosome copy number states for individual tumors compared to normal references (red line = average CN). Specific amplification events are enlarged to show position of HPV integration sites (blue arrows) with respect to genes affected by the event.

We analyzed locations of HPV integration sites in relation to CNVs and found that integrations were more likely to be located near CNVs (within 150 kb) than expected by random chance (OR 1.34, 95% CI 1.00-1.72; p=0.047, n=29 samples with detected HPV integrations). HPV integrations occurred in or near amplifications in MYC, SIM2, and NFIX; however, these genes were also recurrently amplified in tumors without evidence of HPV integration (SIM2 in 6.1% and NFIX in 30.6% of primary tumors) (Figure 5B). Another nuclear factor gene, NFIC, was recurrently amplified in primary tumors (34.7%), but HPV-integrations were not detected at this locus. An HPV integration site was detected in one sample near amplification of CARNS1 and ~2 Mb upstream of a focal amplification containing CCND1, FGF3, FGF4, and FGF19; this region was recurrently amplified in other tumor samples as well, independent of HPV integrations (Figure 5B). Other HPV integrations were located near focal amplifications that included cancer-related genes EML4 and NBPF14, but these genes were not recurrently amplified.

Functional enrichment analysis of genes within recurrent copy number variations

Functional enrichment analysis was performed for genes within recurrent focal amplifications or deletions in either primary tumors (n=49), recurrent or metastatic tumors (n=30), or all tumors (restricted to one sample per patient; n=65) (Figure 6A). This analysis was restricted to a curated list of 981 cancer-related genes, of which 902 mapped to gene ontology (GO) terms (Supplemental Tables S3 and S4). Top GO terms enriched in the set of 79 genes within recurrent CNV deletions in all tumor samples included “cellular response to DNA damage stimulus” and “DNA repair” (FDR-adjusted p=0.00016 and p=0.0004, respectively; Supplemental Table S4). DNA damage response genes within these GO terms and recurrently deleted in either primary and recurrent or metastatic tumors are shown in Figure 6B. Among recurrently deleted DNA damage response genes (n=41), the most common in both primary and recurrent tumors (12.2% - 28.6%) included APC, BRCA2, FANCD2, MLH1, MSH3, PARP3, RAD18, RAD50, RB1, SETD2, and XPC (Figure 6B). When recurrent deletions were analyzed separately in primary and recurrent or metastatic tumors, enrichment of “cellular response to DNA damage stimulus” and “DNA repair” remained significant for the set of 48 genes within recurrent CNV deletions in recurrent or metastatic tumors (adjusted p-values=0.001 and 0.002, respectively; Supplemental Table S4), but not the set of 37 genes within recurrent CNV deletions in primary tumors, suggesting enrichment of recurrently deleted DNA damage response genes was largely driven by deletions in recurrent or metastatic tumors. Recurrent DNA damage response gene deletions exclusively detected in recurrent tumors included ATR, MBD4, and TOPBP1 (10%). Similar analyses of recurrent amplifications detected enrichment of the term “negative regulation of RNA metabolic process” (adjusted p-value=0.015) for the set of 28 genes within recurrent CNV amplifications in all tumors, including BCL6, CCND1, MYC, NFIX, NOTCH1, SIM2, TERT, and TP63, while “cell differentiation” showed a trend toward enrichment (adjusted p-value =0.075) (Supplemental Table S4).

Figure 6. Venn diagram of overlap between recurrent copy number variations and HPV integrations.

A. All genes with HPV integrations within 150 kb and 4 or more reads supporting a breakpoint junction including at least one chimeric read are shown, with overlaps between cancer-related genes within recurrently amplified and deleted regions in either primary or recurrent and metastatic tumors. Cancer-related genes from the curated list in Supplemental Table S3 are in bold, and DNA damage response genes from lists in Supplemental Tables S3 and S4 are in red. B. DNA damage response genes in recurrent amplifications or deletions from (A) stratified by percent of primary and recurrent or metastatic ASCC with the indicated CNV as determined by the GISTIC algorithm.

Relationship of HPV integrations and recurrent copy number variations to clinical outcomes

More HPV integration events were detected in late compared to early stage tumors (2.69 integrations per sample for stage III and IV tumors (n=29 HPV positive tumors) compared to 0.91 integrations per sample for stage I and II tumors (n=22 HPV positive tumors) (Mann-Whitney test, p=0.008). More integration events were also detected in recurrent or metastatic tumors compared to primary tumors (1.88 integrations per sample (n=31) vs. 1.10 integrations per sample in primary tumors (n=56)), although the difference only trended toward statistical significance (p=0.092). We also detected a trend of more advanced tumor stage in samples with HPV integrations detected near cancer-related genes compared to HPV-negative tumors (median tumor stage III vs. median tumor stage II, n=11 and n=7, respectively; p=0.066).

For 28 primary tumors with available data on recurrence or non-recurrence during follow up and passing quality filters (all from Mouw et al. (11) and Shin et al. (25)), we performed Fisher’s exact test comparing the frequency of recurrent CNVs between primary tumors that recurred vs. did not recur. Among the most common CNVs that were recurrent in both primary and recurrent tumors, there was no statistically significant difference between primary tumors that did (n=14) vs. did not (n=14) recur (14.3% vs. 42.8% for PIK3CA amplifications, 21.4% vs. 7.1% for APC deletions, and 14.3% vs. 14.3% for MLH1 and SETD2 deletions, respectively; p=0.21, p=0.33, and p=0.99, respectively). We did not detect statistically significant differences in the frequency of recurrently deleted DNA damage genes when comparing primary tumors that did vs. did not recur for the set as a whole (n=41 genes shown in Figure 6B; 42.8% vs. 71.4%, respectively; p=0.25) or individually (all p-values > 0.30) (Supplemental Table S5). While copy number losses on 13q14 that included BRCA2, FOXO1, and RB1 were more common in primary tumors that did not recur (42.8% vs. 21.4% in primary tumors that recurred), the difference was not statistically significant (p=0.42).

Discussion

Here, we report the largest study to date of WXS from ASCCs including analysis of the tumor virome, HPV integration sites, and CNVs in primary and recurrent or metastatic tumors. HPV integrations were detected in 38% of tumors, more prevalent in later stage tumors, and associated with CNVs including amplified oncogenes and deleted tumor suppressors. Recurrent CNVs detected in ASCCs included amplification of PI3KCA, MYC, and CCND1, and deletion of APC (10, 17, 33-39). Chromosome 3 was frequently disrupted by large scale amplifications and deletions, a common finding in other HPV-related malignancies (11, 22, 24, 25, 40). RB1 was recurrently deleted in primary ASCC, consistent with studies of HPV-positive oropharyngeal cancers (10, 41). BRCA2 and RB1 deletions within 13q14 losses were more frequent in primary ASCCs that did not recur vs. those that recurred (42.8% vs. 21.4%, respectively), although the difference did not reach statistical significance. DNA damage response genes were statistically enriched among recurrently deleted genes in ASCCs, particularly in recurrent or metastatic tumors, suggesting that disruption of genes involved in DNA damage control may contribute to treatment resistance. TERT amplifications and FANCD2, FHIT, and SETD2 deletions were common in recurrent tumors (10-20%), while ATR, MBD4, and TOPBP1 deletions were detected exclusively in recurrent or metastatic tumors (10%). These findings suggest that CNVs involving DNA damage response gene deletions warrant further evaluation in primary and recurrent ASCCs to understand their prognostic significance.

While HPV-16 was the most common genotype detected in ASCCs overall, HPV-18 and HPV-6 were significantly more common in HIV-positive compared to HIV-negative ASCC cases. These findings are consistent with prior studies reporting higher detection rates of non-HPV-16 genotypes including HPV-18, −45, −58, −73 in anogenital neoplastic lesions in HIV-positive patients (7). Low-risk HPV-6 was detected in ASCCs from 3 HIV-positive patients, and multiple HPV-6 integration sites were detected in 2 tumors from an HIV-negative patient, including integrations in CCSER1, MUC12, HRNR, and GOLGA6L22. These findings, together with prior studies reporting rare cases of ASCC, cervical cancer, and HNSCC in which HPV-6 was the only genotype detected (31, 32, 42, 43), suggest that HPV-6 can be an oncogenic factor in some ASCCs.

Our data suggest that HPV integration is linked to ASCC oncogenesis through activation of oncogene expression. HPV integrations were frequently associated with CNVs, including amplification events containing MYC, an oncogene associated with HPV integration in cervical cancer and HNSCC (17, 18, 44), in addition to CCND1, EML4, and NBPF14. SIM2, another amplified HPV integration target we identified, may also promote oncogenesis (45, 46). HPV integrations were more frequent in stage III-IV compared to stage I-II ASCC and in recurrent or metastatic compared to primary ASCC, consistent with studies showing that HPV integration increases as infected cells progress to malignancy (16, 47, 48). HPV replication relies on dysregulation of the DNA damage response pathway (49-51), raising the possibility that increased HPV integration might be a consequence of chromosomal instability.

Nuclear Factor I family genes NFIX and NFIA were targets of HPV integration, and NFIX and NFIC were recurrently amplified in ASCC. NFI transcription factors have been previously linked to oncogenesis and may promote tumor heterogeneity by epigenetic modulation of nucleosomes in addition to altering gene-expression directly as transcription factors (52). A prior study with subjects overlapping the Cacheux et al. cohort included herein identified NFIX integrations in five ASCC cases using a PCR-based strategy to detect human-viral junctions and reported that NFIX integrations were associated with CNV gains (53). Two HPV integrations were between 663 kb and 2.3 Mb upstream of NFIX and three were detected within the first intron of the NFIX gene. Our study mapped three integration junctions to the 4^th intron of NFIX in one patient, and one integration junction to the 2^nd intron of NFIA in a second patient. These findings identify Nuclear Factor I genes as recurrent targets of HPV integration in ASCCs and highlight potential relevance of these genes for oncogenesis in HPV-related cancers.

MUC genes were a recurrent target of HPV integration in ASCC, including MUC12 (5 patients) and MUC1 (1 patient). Furthermore, MUC4 and MUC20 were recurrently amplified in primary and recurrent or metastatic tumors (45% of all tumor samples). MUC genes have been linked to tumor development in pancreatic and colon cancer (54, 55). MUC genes are characterized by an array of tandem repeats in a central exon; repetitive elements have been suggested as targets of HPV integration, a potential mechanism that might underlie these recurrent events (16).

Several herpesviruses were detected in a subset of tumor and normal samples, including EBV, CMV, HSV-1, and roseola (HHV-6B). Given that these viruses do not typically infect epithelial or squamous cells in the anal canal, their detection may originate from tumor-infiltrating lymphocytes (56). In normal tissue samples, these viral sequences were detected only in HIV-positive subjects, possibly due to an immunocompromised state.

We acknowledge several limitations of our study. Recurrent ASCCs were over-represented in our cohort compared to the general ASCC population, raising the possibility that some of our findings may not be generalizeable. The sample size and availability of some clinical data limited our ability to detect genomic differences between primary tumors that did vs. did not recur after chemoradiotherapy. Integrated HPV was detected in 39% of HPV-16 positive ASCC and 20% of HPV-18 positive ASCCs; these rates are lower than expected, potentially reflecting predominance of exon sequences in WXS data. There were many low-coverage (one read pair) chimeric reads in FFPE samples, which we eliminated by setting detection threshold at 4 reads or higher. We applied this threshold because sample fixation can induce both DNA strand breaks and crosslinks (57, 58), potentially yielding false positive chimeric reads. However, some true positives may have been lost due to this filtering. In samples with no direct evidence of human/HPV breakpoint junctions, our method could not differentiate between episomal and integrated viral DNA. The likely underestimate of HPV integrations due to incomplete genome coverage in WXS data and high background of false positives in FFPE samples suggests future studies utilizing whole genome sequencing of frozen samples will be informative. Genomic profiling combined with RNA sequencing will also be important to evaluate expression of cancer-related genes, viral genes, and fusion transcripts in relation to findings described herein.

In conclusion, we characterized the viral and genomic landscape of ASCC using WXS data. We found marked differences in prevalence of ASCC-associated HPV genotypes by HIV status, with lower prevalence of HPV-16 and higher prevalence of HPV-18 and HPV-6 among HIV-positive cases. HPV integrations were mapped within or near MYC and CCND1 amplifications and recurrent targets included Nuclear Factor I and MUC genes, suggesting direct links between HPV integration and oncogenesis in ASCC. We identified differences in CNVs between primary and recurrent or metastatic tumor samples, including detection of recurrent amplification of TERT and recurrent deletion of ATR, FANCD2, and FHIT in ASCC recurrences or metastases. Our finding that DNA damage response genes were enriched among recurrent deletions in recurrent ASCCs suggests that disruption of these genes may play a role in recurrence of these tumors following chemoradiotherapy and underscores the need for future studies to evaluate the prognostic significance of these mutations and potential therapeutic opportunities.