Genome-Wide Profiling of Acquired Uniparental Disomy Reveals Prognostic Factors in Head and Neck Squamous Cell Carcinoma

Abstract

Acquired uniparental disomy (aUPD) leads to homozygosity facilitating identification of monoallelically expressed genes. We analyzed single-nucleotide polymorphism array-based genotyping data of 448 head and neck squamous cell carcinoma (HNSCC) samples from The Cancer Genome Atlas to determine the frequency and distribution of aUPD regions and their association with survival, as well as to gain a better understanding of their influence on the tumor genome. We used expression data from the same dataset to identify differentially expressed genes between groups with and without aUPD. Univariate and multivariable Cox proportional hazards models were performed for survival analysis. We found that 82.14% of HNSCC samples carried aUPD; the most common regions were in chromosome 17p (31.25%), 9p (30.13%), and 9q (27.46%). In univariate analysis, five independent aUPD regions at chromosome 9p, two regions at chromosome 9q, and the CDKN2A region were associated with poor overall survival in all groups, including training and test sets and human papillomavirus (HPV)-negative samples. Forty-three genes in areas of aUPD including PD-L1 and CDKN2A were differentially expressed in samples with aUPD compared to samples without aUPD. In multivariable analysis, aUPD at the CDKN2A region was a significant predictor of overall survival in the whole cohort and in patients with HPV-negative HNSCC. aUPD at specific regions in the genome influences clinical outcomes of HNSCC and may be beneficial for selection of personalized therapy to prolong survival in patients with this disease.

Introduction

Head and neck squamous cell carcinoma (HNSCC) is the seventh most common cancer worldwide; more than half a million new patients are diagnosed each year [1]. Incidence has increased, especially among young patients, because of increasing prevalence of human papillomavirus (HPV) [2], [3]. The 5-year overall survival (OS) rate is better in patients with HPV-associated HNSCC than in those whose tumors are not associated with HPV [4].

Loss of heterozygosity (LOH) results from loss of one of two parental alleles present in each genome. In most cases LOH results in cells having a single copy of one parental allele and loss of the other allele. Acquired uniparental disomy (aUPD) also called copy-neutral LOH) is a subset of LOH wherein a chromosomal region or whole chromosome is lost and reduplicated. aUPD is not associated with changes in copy number. Thus each cell harbors two copies of a single parental allele rather than one copy each of two parental alleles. Both regulatory and open reading frames are monoallelic and any alterations in promoter, enhancer or regions either as the result of germline SNPs or methylation that are included in the aUPD could alter the expression or stability of mRNAs or the stability of function of their protein products. aUPD thus has the potential to expose effects of homozygosity for existing germline and somatic aberrations including mutations, deletions, methylation (hypo- or hyper-), complex structural alterations, and imprinted genes [5], [6], [7], [8], [9], [10]. aUPD can be a consequence of mitotic recombination that usually results in segmental aUPD where only a portion of the chromosome arises from a single parent. Whole chromosome aUPD is usually the consequence of anaphase lagging of one chromosome, and with duplication of the whole chromosome [11], [12], [13], [14], [15]. Moreover, loss of chromosomal segments or whole-chromosome, and frequently duplication of the retained allele in subsequent replication can be consequence of breakage-fusion-bridge (BFB) events in cancer [16]. Thus, a UPD can be a result of BFB and thereafter replication of the retained allele [16]. Genomic copy number alterations, gene expression, miRNA expression, and protein expression are well studied in HNSCC. Accumulating data have shown that genomic events including loss of heterozygosity (LOH) and epigenetic changes present in tumors can be used as prognostic biomarkers for cancer [17], [18], [19]. However, genome-wide profiling of aUPD in HNSCC is very limited with most studies having small sample sizes [20], [21], [22]. Of note, aUPD profiling and association between aUPD regions and survival have been reported in a variety of malignancies including MDS, MDS/MPD, and secondary AML [23]. However, allele-based level changes in the genome and their association with clinical outcome and survival are poorly characterized, and acquired uniparental disomy (aUPD) has not been studied as a potential prognostic factor in HNSCC.

Previously, we have shown that smallest overlapping regions of aUPD were associated with etiologic factors such as alcohol intake, smoking, HPV and TP53 mutation status of HNSCCs [24]. In the current study, we profiled genome-wide aUPD to determine the frequency and distribution of aUPD in HNSCC and to determine whether any smallest overlapping regions (SORs) of aUPD were associated with survival and clinical characteristics of disease in a large data set. Importantly, we assessed expression of mRNA for genes located in areas of aUPD to determine effects of monoalellic gene expression due to aUPD on gene expression. This represents the first large scale comprehensive study of aUPD regions and their association with survival in HNSCC.

Materials and Methods

Patient Samples

Clinical and patient demographic data were retrieved from TCGA (http://portal.gdc.cancer.gov/).). We noted HPV status as reported by TCGA and by Nulton et al. [25], [26]. Patient characteristics are summarized in Table S1a, and all samples are listed in Table S1B. Overall survival (OS) was calculated from the date of diagnosis of head and neck cancer to the date of death or last follow-up. Recurrence-free survival (RFS) was calculated from the date of diagnosis of HNSCC to the date of recurrence or last follow-up. Sample and clinical data were based on a November 2015 freeze from TCGA data portal.

Genomic Data and aUPD Analysis

Genomic data (CEL files) were retrieved from TCGA data portal. Genotyping console software (Affymetrix) was used to perform quality control, and then to generate CHP files. Affymetrix contrast QC threshold was used for both tumor and matching samples. Data that did not pass the quality control check were removed from further analysis; 448 samples (448 tumor and 448 matching normal) passed the quality control check and were used in the study. aUPD analysis and detection of SORs of aUPD were performed using Copy Number Analyzer for GeneChip (CNAG v4.0) software (http://www.genome.umin.jp) by using tumor and matching normal data as described previously [10], [27]. The SORs of UPD are described based on the 3′ and 5′ endpoints of aUPD regions. The Dec 2013 human genome browser (NCBI Build 38/hg38; http://genome.ucsc.edu) was used for identification of gene localization. Telomeric aUPD was defined as aUPD occurring in the telomeric region with one breakpoint. When at least two breakpoints appeared, it was defined as centromeric aUPD (also called interstitial). Segmental aUPD was defined as all centromeric and telomeric regions. When aUPD occurred in whole-p arm or whole-q arm of chromosome, it was defined as whole-p arm or whole-q arm aUPD. If aUPD occurred in the whole chromosome, it was considered whole-chromosome aUPD (Figure S1A). Total aUPD represented both segmental and whole-chromosome aUPD. We recruited normalized HiSeq gene expression data from Cancer Genome Browser (https://xena.ucsc.edu/) to determine differentially expressed genes between samples with and without aUPD.

Statistical Analysis

The nonparametric Kruskal-Wallis test was used to analyze the correlation between aUPD (total, telomeric, centromeric, segmental, whole-chromosome, whole-p arm, and whole-q arm) and clinical characteristics (stage, grade, HPV status, and gender). Univariate Cox proportional hazards models were used to determine the effects of SORs of aUPD, gender, age, stage, and grade on OS and RFS. Kaplan-Meier analysis and log-rank p values were calculated to identify survival differences between groups. A multivariable Cox proportional hazards model was used to select prognostic markers. This study complied with REMARK criteria [28]. A two-tailed Student t test was used to compare expression of genes between samples with aUPD and those without aUPD for identified SORs. To correct for multiple comparisons, we adjusted the p values by obtaining the Benjamini-Hochberg false discovery rate [29]. A finding was considered significant when the two-sided P value was less than .05. Statistical analyses were performed in STATA v10 (STATA Corp., College Station, TX).

Results

Profiling SORs of aUPD

We analyzed single-nucleotide polymorphism-based arrays generated by The Cancer Genome Atlas (TCGA) from 448 HNSCCs to identify the frequency and distribution of segmental and whole-chromosome aUPD. A total of 1591 aUPD regions were found across all of the chromosomes (per sample range 0 to 27, mean 3.56, median 3.0), including 1407 segmental and 184 whole-chromosome events in the entire data set. The frequency of HNSCC samples harboring at least one aUPD was 82.14% (368 of the 448 samples). The most common instances of aUPD were found in chromosomes 17p (31.25%), 9p (30.13%), 9q (27.46%), and 13q (18.53%) (Figure S1B).

Next we tested for differences in the frequency of total, telomeric, centromeric, segmental region, whole-chromosome, whole-p arm, and whole-q arm aUPD among patient samples by disease stage, grade, and HPV status. In Kruskal-Wallis tests, the frequency of total, telomeric, centromeric, segmental aUPD was significantly higher in patients with stage III and IV disease than in those with stage I and II disease (total: P = .005, telomeric: P = .011, centromeric: P = .021, segmental: P = .002), but no differences were found by disease stage in whole-p arm (P = .278), whole-q arm (P = .375), and whole-chromosome (P = .4821; Figure 1). In contrast to disease stage, the frequency of aUPD was not correlated with grade (total: P = .733, telomeric: P = .793, centromeric: P = .133, segmental: P = .826, whole-chromosome: P = .819, whole-p arm: P = .675, whole-q arm: P = .352; Figure S2). The frequencies of total (P = .036) aUPD were significantly higher in HPV-negative patients than in HPV-positive patients. However, no differences were found according to HPV status for telomeric (P = .472), centromeric (P = .962), segmental (P = .102), whole-chromosome (P = .188), or whole-p arm (P = .810), and whole-q arm (P = .389) aUPD (Figure S3).

Figure 1.

Frequency of total, telomeric, centromeric, segmental, whole-chromosome, whole-p arm, and whole-q arm acquired uniparental disomy (aUPD) in patients with head and neck squamous cell carcinoma, stratified by disease stage. The frequency of total, telomeric, centromeric, and segmental aUPD was significantly higher in patients with stage III and IV disease than in those with stage I and II disease, but no differences were found by disease stage in whole-p arm, whole-q arm, and whole-chromosome.

Association of Recurrent SORs of aUPD With Survival

We identified 23 SORs of aUPD, including CDKN2A, across all chromosomes. Independent SORs were identified at chromosomes 2q (two regions), 6p (one region), 9p (10 regions), 9q (four regions), 11q (two regions), 13q (one region), and 17p (two regions), as well as CDKN2A (at chromosome 9p) (see Table 1 and Supplemental Table S2). Next, we tested whether any of these SORs were associated with OS or recurrence-free survival (RFS). First we randomly divided samples into a training and a test set (Table S1) and assessed whether the SORs were associated with survival. In univariate analysis, SORs of aUPD at chromosome 9p (9p24.3; P = .044, 9p24.1; P = .020, 9p23-p22.3; P = .009, 9p22.3-p22.2; P = .014, 9p21.3_1; P = .017, 9p21.3_2; P = .024, 9p21.3-p21.2; P = .044, 9p21.2; P = .008, 9p21.1; P = .044, and 9p13.3; P = .029, respectively; Table 1, Figure 2, Figure 3, Table S2a, and Figure S4) and CDKN2A (P = .045; Table 1, Figure 3), as well as gender (P = .022; Table 1, Figure S4), were associated with shorter OS in the training set. In the test set, one SOR of aUPD at chromosome 6p12.3 (P = .018), seven SORs at chromosome 9p (9p24.3; P = .008, 9p24.1; P = .007, 9p23-p22.3; P = .029, 9p22.3-p22.2; P = .011, 9p21.3_1; P = .010, 9p21.3_2; P = .018, and 9p21.3-p21.2; P = .033, respectively), two SORs at 9q (9q33.2; P = .025; and 9q34.13; P = .026), and one SOR at the CDKN2A (9p21.3; P = .008) were associated with reduced OS (Table 1, Figure 2, Figure 3, Table S2B, Supplementary Figure S4, Supplementary Figure S5). In the test set, only two SORs of aUPD at 11q (11q22.3; P = .008; and 11q25; P = .002) were associated with reduced RFS (Table S2B). In the training set, none of the SORs of aUPD were associated with RFS (Table S2a). In multivariable analysis, 9p21.2 (P < .0001) in training set, and 9p24.1 (P < .0001) and 9p23-p22.3 (P < .0001) in test set were associated with worst OS, while none of SOR of aUPD was a significant predictor of RFS in training and test sets (Table 2).

Table 1.

Univariate analysis of clinical variables and the SORs of aUPD serving as covariates of survival in patients with head and neck squamous cell carcinoma

	Training set		Test set		All samples		HPV-negative
Covariate	HR (95% CI)	Pa	HR (95% CI)	Pa	HR (95% CI)	Pa	HR (95% CI)	Pa
OS
Age >50 vs. ≤50	1.15 (0.57-2.31)	.700	0.80 (0.48-1.31)	.370	0.88 (0.59-1.31)	.530	0.87 (0.57-1.33)	.523
Stage I&II vs. III&IV	1.30 (0.75-2.27)	.350	1.07 (0.67-1.71)	.779	1.19 (0.83-1.70)	.354	1.13 (0.77-1.67)	.529
Grade 1 vs. 2&3&4	1.96 (0.84-4.56)	.120	1.24 (0.69-2.23)	.464	1.42 (0.89-2.29)	.145	1.32 (0.78-2.24)	.297
Gender	1.71 (1.08-2.72)	.022	1.43 (0.92-2.22)	.115	1.55 (1.13-2.13)	.007	1.58 (1.12-2.22)	.009
aUPD 6p12.3	0.43 (0.11-1.77)	.242	1.99 (1.13-3.54)	.018	1.41 (0.84-2.36)	.196	1.32 (0.76-2.29)	.332
aUPD 9p24.3	1.65 (1.01-2.69)	.044	1.76 (1.16-2.68)	.008	1.62 (1.21-2.17)	.001	1.75 (1.25-2.45)	.001
aUPD 9p24.1	1.77 (1.09-2.87)	.020	1.79 (1.17-2.72)	.007	1.81 (1.32-2.74)	<.0001	1.82 (1.30-2.55)	<.0001
aUPD 9p23-p22.3	1.89 (1.17-3.05)	.009	1.61 (1.05-2.47)	.029	1.74 (1.27-2.39)	.001	1.79 (1.27-2.51)	.001
aUPD 9p22.3-p22.2	1.80 (1.13-2.86)	.014	1.73 (1.13-2.65)	.011	1.66 (1.24-2.23)	.001	1.85 (1.32-2.60)	<.0001
aUPD at CDKN2A	1.71 (1.01-2.88)	.045	2.15 (1.23-3.78)	.008	1.96 (1.34-2.85)	<.0001	1.98 (1.31-2.97)	.001
aUPD 9p21.3_1	1.77 (1.11-2.84)	.017	1.75 (1.14-2.67)	.010	1.80 (1.31-2.46)	<.0001	1.82 (1.29-2.55)	.001
aUPD 9p21.3_2	1.72 (1.07-2.76)	.024	1.68 (1.09-2.57)	.018	1.73 (1.26-2.37)	.001	1.76 (1.25-2.47)	.001
aUPD 9p21.3-p21.2	1.63 (1.01-2.63)	.044	1.60 (1.04-2.46)	.033	1.65 (1.20-2.26)	.002	1.71 (1.22-2.41)	.002
aUPD 9p21.2	1.91 (1.19-3.08)	.008	1.43 (0.92-2.20)	.108	1.64 (1.19-2.25)	.002	1.79 (1.27-2.52)	.001
aUPD 9p21.1	1.67 (1.01-2.73)	.044	1.29 (0.82-2.01)	.269	1.45 (1.04-2.02)	.027	1.51 (1.06-2.15)	.024
aUPD 9p13.3	1.74 (1.06-2.86)	.029	1.56 (0.99-2.47)	.057	1.64 (1.17-2.29)	.004	1.62 (1.13-2.33)	.009
aUPD 9q22.33	1.56 (0.94-2.59)	.082	1.38 (0.87-2.19)	.171	1.47 (1.05-2.07)	.025	1.58 (1.10-2.27)	.014
aUPD at 9q31.3	1.62 (0.99-2.65)	.054	1.52 (0.97-2.38)	.067	1.58 (1.13-2.19)	.007	1.70 (1.20-2.43)	.003
aUPD at 9q33.2	1.48 (0.90-2.44)	.120	1.67 (1.07-2.60)	.025	1.59 (1.14-2.21)	.006	1.66 (1.16-2.37)	.005
aUPD at 9q34.13	1.34 (0.81-2.20)	.259	1.65 (1.06-2.57)	.026	1.52 (1.10-2.12)	.012	1.62 (1.14-2.30)	.007
DEL at CDKN2A	1.24 (0.71-2.18)	.454	1.54 (0.88-2.70)	.132	1.38 (0.94-2.02)	.097	1.28 (0.85-1.93)	.235
RFS
aUPD 11q22.3b	1.25E-14 (0-0)	1.000	3.62 (1.41-9.33)	.008	3.60 (1.44-9.00)	.006	3.66 (1.45-9.23)	.006
aUPD 11q25b	1.25E-14 (0-0)	1.00	4.07 (1.68-9.75)	.002	4.04 (1.74-9.43)	.001	3.66 (1.45-9.23)	.006

Abbreviations: SOR, smallest overlapping region; aUPD, acquired uniparental disomy; HR, hazard ratio; CI, confidence interval; HPV, human papillomavirus; OS, overall survival; RFS, recurrence-free survival; DEL, deletion.

^aP < .05 was used to select features; boldface indicates statistically significant variables.

^baUPD-positive sample size was small in these two variables.

Figure 2.

Kaplan-Meier plots of overall survival (OS) for acquired uniparental disomy (aUPD) at chromosomes 9p23-p22.3, 9p22.3-p22.2, and 9p21.3_1 have shown worse OS than the samples without aUPD in the training set and test set from patients with head and neck squamous cell carcinoma, as well as in all samples and human papillomavirus (HPV)-negative patients only.

Figure 3.

Kaplan-Meier plots of overall survival for acquired uniparental disomy (aUPD) at chromosomes 9p21.3_2 and 9p21.3-p21.2 and CDKN2A have shown shorter OS than the samples without aUPD in the defining regions in the training set, test set, all samples from patients with head and neck squamous cell carcinoma, as well as human papillomavirus (HPV)-negative patients only. *Samples with aUPD at CDKN2A region was compared with samples without aUPD and deletion for the same region; aUPD-Neg; aUPD-Pos.

Table 2.

Multivariable analysis of clinical and genetic covariates for OS and RFS in patients with head and neck squamous cell carcinoma

Variable	HR (95% CI)	Pa	q
OS
Training set
9p21.2	2.95E+09 (8.67+ 07-1.01+ 11)	<.0001	<.0001
Test set
9p24.1	4.54E+07 (5.88E+06-3.50E+08)	<.0001	<.0001
9p23-p22.3	5.06-08 (1.95E-09-1.32E-06)	<.0001	<.0001
All samples
9p21.3_1	61.34 (1.59-2369.73)	.027	.037
9p21.2	7.73E+08 (1.05E+08-5.67E+09)	<.0001	<.0001
9p21.1	0.26 (0.09-0.79)	.017	.029
CDKN2A	0.13 (0.02-0.85)	.034	.037
HPV-negative samples
9p21.3-p21.2	7.85E-10 (1.05E-10-5.88E-09)	<.0001	<.0001
9p21.1	0.29 (0.09-0.88)	.029	.037
CDKN2A	0.13 (0.02-0.92)	.041	.041
RFS
All samples
11q25	8.63 (1.18-63.18)	.034	.037
HPV-negative samples
11q22.3	3.66 (1.45-9.23)	.006	.012

Abbreviations: OS, overall survival; RFS, recurrence free survival; HR, hazard ratio; CI, confidence interval; q, Benjamini-Hochberg false discovery rate.

Only variables that were significant in univariate analysis were included in multivariate analysis.

^aP < .05 was used to select features.

Then we tested all samples to determine whether the SOR were associated with survival. In univariate analysis, aUPD in 14 independent SORs was associated with reduced OS in all samples; 10 regions at chromosomes 9p (9p24.3; P = .001, 9p24.1; P < .0001, 9p23-p22.3; P = .001, 9p22.3-p22.2; P = .001, 9p21.3_1; P < .0001, 9p21.3_2; P = .001, 9p21.3-p21.2; P = .002, 9p21.2; P = .002, 9p21.1; P = .027, and 9p13.3; P = .004) and four at chromosome 9q (9q22.33; P = .025, 9q31.3; P = .007, 9q33.2; P = .006, and 9q34.13; P = .012; Table 1, Table S2C, Figure 2, Figure 3, Supplementary Figure S4, Supplementary Figure S5). Conversely, only two SORs of aUPD at chromosome 11q (11q22.3; P = .006, and 11q25; P = .001) were associated with reduced RFS in all HNSCC samples (Table 1, Figure S4). Next, we analyzed associations between aUPD or deletion at the CDKN2A region and survival. We found that aUPD at the CDKN2A region (P < .0001) was associated with poor OS (Figure 3, Table 1), but deletion in the same region was not associated with OS (P = .097) in all samples. Moreover, we tested whether age, gender, stage, and grade were associated with survival and found that only gender was associated with reduced OS (P = .007; Figure S4); but age (P = .530), grade (P = .145), and stage (P = .354) were not associated with OS. In multivariable analysis, four aUPD regions at chromosome 9p (9p21.3_1; P = .027, 9p21.2; P < .0001, and 9p21.1; P = .017, and CDKN2A; P = .034) were significant predictors of OS, and one region at chromosome 11q (11q25, P = .034) was a significant predictor of RFS (Table 2).

When we tested only HPV-negative samples, we found that the same 14 SORs of aUPD were associated with shorter OS: chromosome 9p (9p24.3; P = .001, 9p24.1; P < .0001, 9p23-p22.3; P = .001, 9p22.3-p22.2; P < .0001, 9p21.3_1; P = .001, 9p21.3_2; P = .001, 9p21.3-p21.2; P = .002, 9p21.2; P = .001, 9p21.1; P = .024, and 9p13.3; P = .009) and chromosome 9q (9q22.33; P = .014, 9q31.3; P = .003, 9q33.2; P = .005, and 9q34.13; P = .007; Table 1, Figure 2, Figure 3, Table S2D, Figure S4). Similar to all samples, only two SORs at chromosome 11q (11q22.3; P = .006, and 11q25; P = .006) were associated with shorter RFS. In addition, the SOR of aUPD at CDKN2A (P = .001; Figure 3) was associated with shorter OS, but deletion at the same region was not associated with OS (P = .235). Gender (P = .009) was also associated with reduced OS. In multivariate analysis, SORs of aUPD at chromosome 9p21.3-p21.2 (P < .0001), 9p21.1 (P = .029) and CDKN2A (9p21.3; P = .041) were significant predictors of OS, and only SOR of aUPD at chromosome 11q22.3 (P = .006) was a significant predictor of RFS (Table 2). In contrast to all samples and HPV-negative samples, none of the SORs were associated with OS or RFS in HPV-positive samples.

Association of Differentially Expressed Genes in the SORs of aUPD with Survival

Seventeen of 22 SORs in all samples, including the CDKN2A, and additional one region in the test set, were associated with survival. The 18 SORs contained 135 genes (Table S3A). The number of genes in these 18 SORs varied. Two SORs at chromosome 9p (9p21.3_2 and 9p21.1) did not contain any open reading frames (ORF), however both contain long non-coding RNA and pseudogenes, and 9p21.3_2 also harbors regulatory elements (Supplementary Table S3B). One of independent SOR at 6p12.3, four at chromosome 9p (9p24.3, 9p23-p22.3, 9p21.3_2, and 9p21.3-p21.2, one at 9q31.3, and one at 11q25 contained only one protein-coding gene; the remaining SORs consisted of multiple genes (e.g., 9p13.3 had 51 genes; Table S3A). Moreover, miRNAs or and non-coding RNAs as well as pseudogenes and regulatory elements (promotor or enhancers) are mapped in the 18 regions explored in this manuscript and may contribute to the correlations with outcomes (Supplementary Table S3B).

Next, we tested whether the expression of these 135 genes differed between samples with and without aUPD. Fifty-six genes were significantly differentially expressed between samples with and without aUPD. With Benjamini-Hochberg false discovery rate correction, only 43 genes were significantly differentially expressed. Forty-one of these 43 genes had significantly higher expression in SORs of aUPD-positive samples than in aUPD-negative samples. Significantly overexpressed genes included CD274 (also known as PD-L1; q = 8.99E-06) in SOR at chromosome 9p24.1, and DCTN3 (q = 2.86E-07) and VCP (q = 2.86E-07) at chromosome 9p13.3. We found that only two of the 43 genes had significantly lower expression in aUPD-positive samples than in aUPD-negative samples: CDKN2A at chromosome 9p21.3 (q = 7.58E-04) and GGTA1 at chromosome 9q33.2 (P = 3.78E-04; Table S3).

Discussion

Our results indicate that aUPD is widespread in HNSCC, and specific SORs of aUPD in the genome influence survival. We also found that genes within these SORs were differentially expressed between those with and without aUPD.

The mechanisms influencing cancer development may vary among the telomeric, centromeric, whole arm and whole chromosomes aUPD. Thus, in the current study, we analyzed the data for the whole-p and whole-q arm, whole chromosome, telomeric and centromeric aUPD. We found that the frequency of total aUPD, as well as telomeric and centromeric was associated with stage. However, the frequency of aUPD was not associated with grade. These findings contrast with those of our previous report in high-grade serous epithelial ovarian cancers, in which the frequency of aUPD was associated with grade but not with stage [10]. This suggests that aUPD can mediate different functions across cancer lineage.

We identified a total of 18 SORs of aUPD that were associated with shorter OS or RFS. Recently aUPD has been shown to occur at areas encompassing imprinted genes in tumors [8]. The 18 SORs encompass 135 genes and the expression of 43 of the 135 genes within the defined aUPD regions was significantly different compared with samples without aUPD in the same regions. Thus the monoallelic expression of the 43 differentially expressed genes may alter expression. Differentially expression of the 43 genes identified to be located in aUPD SOR in the current study may confer an advantage for the tumor, and may contribute to tumor aggressiveness. Forty-one of the 43 differentially expressed genes had significantly higher expression and only two had lower expression in samples with aUPD compared with those without aUPD. If these genes are in areas encompassing imprinting [8], only the non-imprinted gene was selected. This finding indicates that many of these genes may exhibit gain of function activity. PD-L1 hypomethylation has been observed in HNSCC in TCGA and other data sets. PD-L1 hypomethylation is inversely correlated with mRNA and protein expression [25], [30]. If a hypomethylated PD-L1 allele undergoes reduction to homozygosity through aUPD, this could result in increased PD-L1 mRNA expression as is observed in samples with aUPD compared to those without aUPD.

Six SORs of aUPD at chromosome 9p (9p23-p22.3, 9p22.3-p22.2, 9p21.3_1, 9p21.3_2, 9p21.3-p21.2, and the CDKN2A) were associated with shorter OS in all groups, including the training and test sets, all samples, and HPV-negative samples. Moreover, seven SORs at chromosome 9p (9p24.3, 9p24.1, 9p23-p22.3, 9p22.3-p22.2, 9p21.3_1, 9p21.3_2, 9p21.3-p21.2) and two SORs at chromosome 9q (9q33.2 and 9q34.13) were associated with reduced OS in the test set, all samples, and HPV-negative samples. In contrast, only two SORs of aUPD at chromosome 11q (11q22.3 and 11q25) were associated with RFS in the test set, all samples, and HPV-negative samples. This may be partly because data for RFS in some samples were missing decreasing the power to detect associations. Although only a few samples had SORs of aUPD at chromosome 11q (11q22.3 and 11q25), the overall concordance of the data among the training set, test set, all samples, and HPV-negative samples was notable.

Previously aUPD (in 3 out16 samples) and deletion (in 13 out16 samples) at chromosome 9p were reported in HNSCC [22], and at chromosome 3p and 17 [20]. However these regions were not refined and association between aUPD regions and survival was not assessed due to the small sample sets. LOH at 8p21.2 and 9p21.2 has been reported to be associated with shorter survival in HNSCC samples [17]. LOH at 9p was also found to be predictive for local relapse in HNSCC [19]. However, in both LOH studies microsatellite markers were employed to identify LOH in limited regions in HNSCC samples and the LOH was not further segregated into copy number loss and aUPD [17], [19]. Genome-wide LOH analysis was performed in the TCGA HNSCC paper based on deletion, but LOH or aUPD were not analyzed for association with gene expression or survival in the TCGA HNSCC paper [25].

CD274 (also known as PD-L1), RCL1, PDCD1LG2 (also known as PD-L2), KIAA1432, JAK2 at chromosome 9p24.1, and VCP, DCTN3, STOML2, C9orf23, and GALT at chromosome 9p13.3, and CDKN2A at chromosome 9p21.3 and TTF1 at chromosome 9q34.13 provide examples of significantly differentially expressed genes between samples with and without aUPD that were associated with OS in all samples, HPV-negative samples, and the test set. CD274 (also known as PD-L1) encodes programmed cell death protein-1 ligand 1, which is an immune inhibitory receptor ligand, and interaction of this ligand with its receptor, PD-1 (programmed cell death protein-1), inhibits T-cell activation and cytokine production and enables immune escape. PD-L1 is primarily expressed in cancer cells, parenchymal cells, and myeloid cells, whereas PD-1 is primarily expressed in tumor-infiltrating lymphocytes. Activation of the PD-1/PD-L1 axis occurs in tumors either through innate immune resistance or adaptive immune resistance [31]. Previously, overexpression of PD-L1 has been shown in a variety of cancers, including HNSCC [32], gastric cancer [33], cervical cancer [34], and squamous carcinoma of the cervix and vulva [35], and PD-L1 overexpression confers resistance to radiation in HNSCC [36]. Notably, targeting PD-L1 with anti-PD-L1 monoclonal antibody decreased PD-L1 expression in a variety of tumors, including HNSCC [32], [36], [37]. Our results indicate that PD-L1 and PD-L2 are significantly overexpressed in samples with aUPD compared with those without aUPD. These findings provide potential insight into the mechanisms of PD-L1 and PD-L2 overexpression in HNSCC and support previous reports. Our results also suggest that identification of aUPD in defined regions may help to select patients for individualized therapy. TTF1, which encodes transcription termination factor 1 was also overexpressed in samples with aUPD in defined regions. TTF1 expression and blood vessel invasion were shown to correlate with PD-L1 expression in sarcomatoid lung carcinoma [38]. Moreover, overexpression of TTF1 has been shown to be associated with poor prognosis in colorectal cancers [39].

Of note, we found that aUPD at the CDKN2A region was associated with OS in all samples and in HPV-negative samples, but not in HPV-positive samples. Indeed, expression of CDKN2A in samples with aUPD was significantly lower than in samples without aUPD. In contrast, deletion at the same region was not associated with OS or RFS in any of the sample groups. Other genes with significantly higher expression in samples with aUPD in defined regions compared with those without aUPD included mitochondrial genes (HINT2, ACAT1), genes involved in galactose (GALT, galactose-1-phosphate uridyltransferase) and carbohydrate metabolism (GBA2, glucosidase, beta [bile acid] 2), genes involved in nucleotide exchange (KIAA1432 and C9ORF100), genes involved in oxidative stress defense (ERMP1, endoplasmic reticulum metalloprotease 1), ion channel and DNA excision repair genes (XPA, xeroderma pigmentosum, complementation group A), and non-receptor tyrosine kinase gene (JAK2). In other studies, down-regulation of ERMP1 and C9orf100 significantly reduced cell proliferation and migration [40], [41], [42]. CAT1 encodes a mitochondrially localized enzyme that catalyzes the reversible formation of acetoacetyl-CoA from two molecules of acetyl-CoA and regulates pyruvate dehydrogenase complex [43]. Inhibition of ACAT1 decreases cell proliferation and tumor growth [43].

In addition, VCP (valosin-containing protein), a member of the AAA-ATPase protein family, encodes a protein that interacts with other proteins to regulate endoplasmic reticulum–associated protein degradation. This involves multiple cellular functions during mitosis, including regulation of the spindle pole body, vesicular trafficking, and membrane fusion [44]. Expression of VCP, DCTN3, and STOML2 independently plays a role in increasing cell growth and anchorage-independent growth during the development of invasive oral carcinoma [45]. VCP also promotes growth, invasion, and metastasis in colorectal cancer through activation of STAT3 signaling [46]. Inhibition of VCP expression suppresses West Nile virus infection [47], as well as suppressing the cell cycle, inducing endoplasmic reticulum stress, and inducing caspase-mediated cell death in ovarian cancer cells [48]. ANP32B encodes histone chaperone acidic nuclear phosphoprotein 32B (ANP32B) and plays an anti-apoptotic role. Overexpression of ANP32B has been shown to lead to accumulation of henipavirus matrix (Hendra virus; HeV M) and nuclear proteins (Nipah virus; NiV M) [49]. Down-regulation of ANP32B induces apoptosis in myeloid leukemia cells [50]. Collectively, with recent evidence support roles for the microbiome and a number of viruses in human cancer, it is possible that HeV M and NiV M may be involved in HNSCC development. Further, inhibition of ANP32B expression may enhance apoptosis in HNSCC cells. Our findings indicate that genes involved in galactose, carbohydrate and mitochondrial metabolism, nucleotide exchange, oxidative stress defense, and ion channels may be also involved in the pathogenesis of HNSCC, and inhibition of expression of these genes may reduce cell growth or invasion.

In summary, we demonstrated associations with patient outcome and identified genes in specific SORs of aUPD that are differentially expressed in patients with aUPD, compared with patients without aUPD. These differentially expressed genes may be an indicator of tumor aggressiveness, and in turn affect survival. Therefore, aUPD analysis may be a useful tool to select targeted therapy for this heterogeneous disease.

The following are the supplementary data related to this article.

Supplementary Figure S1

Representative figure for segmental aUPD analyzed by CNAG (A). Distribution of acquired uniparental disomy (aUPD) across the whole-genome. Each red line represents an aUPD region (B).

Supplementary Figure S2

Frequency of total, telomeric, centromeric, segmental, whole-chromosome, p arm, and q arm acquired uniparental disomy (aUPD) in patients with head and neck squamous cell carcinoma, stratified by grade.

Supplementary Figure S3

Frequency of total, telomeric, centromeric, segmental, whole-chromosome, p arm, and q arm acquired uniparental disomy (aUPD) in patients with head and neck squamous cell carcinoma, stratified by human papillomavirus (HPV) status.

Supplementary Figure S4

Kaplan-Meier plot of RFS for aUPD at 11q22.3 and 11q25 in all HNSCC, test set and HPV-negative HNSCC samples. In addition, Kaplan-Meier plot of OS for aUPD at 9p24.3, 9p24.1, 9p21.2, 9p21.1, 9p13.3 and gender in training set, 6p12.3 in test set, 9p21.2, 9p21.1, 9p13.3, 9q22.33, 9q31.3 and gender in HPV-negative HNSCC samples.

Supplementary Figure S5

Kaplan-Meier plots of overall survival for acquired uniparental disomy (aUPD) at chromosomes 9p24.3, 9p24.1, 9q33.2, and 9q34.13 in all samples from patients with head and neck squamous cell carcinoma, as well as the test set and human papillomavirus (HPV)-negative patients only.

Supplementary Table S1a

Demographic and clinical characteristics of the patients with head and neck squamous cell carcinoma whose samples were used for our analysis.

Supplementary Table S2a

Univariate analysis of survival in the training set of samples from patients with head and neck squamous cell carcinoma.

Supplementary Table S3A

Smallest overlapping regions of aUPD that were associated with overall or recurrence-free survival, and differentially expressed genes in those regions.

Supplementary Table S3B

Smallest overlapping regions of aUPD that were associated with overall or recurrence-free survival, and miRNAs, long non-coding RNAs, pseudogenes and number of enhancers in those regions.