Analysis of nuclear receptor expression in head and neck cancer

Abstract

Objective:

Studies of squamous cell carcinoma of the head and neck (HNSCC) have demonstrated the importance of nuclear receptors and their associated coregulators in the development and treatment of HNSCC. We sought to characterize members of the nuclear receptor super family through interrogation of RNA-Seq and microarray data.

Materials and Methods:

TCGA RNA-Seq data within the cBioportal platform comparing HNSCC samples (n = 515 patients with RNA-Seq data) to normal tissue (n = 82 patients) was interrogated for significant differences in nuclear receptor expression. Affymetrix microarray analysis of HNSCC tumors relative to normal oral mucosa (41 tumor, 13 normal) was analyzed.

Results:

Of the 48 NR genes and 19 NR cofactors examined, 99 % of tumor samples in the TCGA had some form of NR gene ‘alteration’ compared to normal tissue. These alterations predominantly encompass expression changes. NR genes (PPARG) and (RORC), and the NR cofactor, (NCOA1), were differentially expressed and downregulated in tumors compared to normal tissue.

Conclusion:

We have discovered significant decreases in PPARG expression with co-occurring changes in genes involved with lipid metabolism and cell cycle progression in HNSCC. We are targeting PPARγ with thiazolidinediones in a series of clinical trials to restore normal signaling via differentiation to hopefully reverse carcinogenesis. We also observed several receptors with differential expression associated with clinical factors that may become the focus of interest in future targeting efforts. These data provide evidence for nuclear receptors playing a role in the dysregulation of gene expression in HNSCC and illustrate the utility of current bioinformatic tools for interrogating complex, high throughput data sets.

Introduction

Squamous cell carcinoma of the head and neck (HNSCC) occurs through malignant transformation of basal layer epithelial cells of the upper aerodigestive tract and oral cavity. HNSCC is cause for significant burden of disease globally, resulting in nearly 200,000 deaths in 2022 worldwide [1]. Advanced stage disease at initial presentation due to rapid growth, lack of robust screening programs and guidelines, and locoregional relapse in HNSCC patients following treatment are major causes of morbidity and mortality. This has motivated scientists and clinicians to interrogate various signal transduction pathways as molecular targets for chemoprevention/therapeutic development as well as for markers of early diagnosis, prognosis, and long-term surveillance.

In cancer prevention, the retinoid nuclear receptor axis is one such signal transduction pathway that has been extensively studied and targeted. Hong et al. first demonstrated the success of isotretinoin (13-cis-retinoic acid) at reversing dysplasia in oral leukoplakia [2]. Isotretinoin, originally used to target retinoic acid receptor beta (RARβ), inhibits carcinogenesis in animal models of oral cancer, suppresses pre-malignant oral leukoplakia, and prevents second primary tumors in HNSCC patients in early phase trials [3]. However, robust reversal of carcinogenesis long term has been an elusive goal with this compound class. Toxicity of effective isotretinoin doses is one barrier to long-term efficacy [4]. We hypothesize other nuclear receptor family members are more broadly implicated in head and neck cancer carcinogenesis, and sought to characterize the nuclear receptor signaling axis to gain insight in other potential targets for chemoprevention.

Nuclear receptors function as transcription factors for a diverse set of fat-soluble hormones, vitamins, and fatty acids. The superfamily now consists of 48 different protein-encoding genes in humans [5-7]. The receptor proteins are made up of structurally related molecules which function as ligand dependent transcription factors with a role in regulation of both normal physiological processes and tumor development [8,9]. Studies have identified several nuclear receptors and their co-regulator proteins as important molecular agents in cell growth and proliferation, cell differentiation, apoptosis, angiogenesis, and metastasis of several types of cancer, including HNSCC [10-22]. An in vitro study of the effects of estrogen exposure in HNSCC cell lines demonstrated a potential effect of estrogen on the differential synthesis of growth-promoting enzymes in cancer [23]. In oral leukoplakia, activation of the PPARγ receptor with the Type II diabetes drug pioglitazone results in upregulation of growth suppressor p21 as well as the inhibition of cell proliferation, an important milestone implicating nuclear receptors in head and neck carcinogenesis [24].

While the aforementioned studies and others have demonstrated the importance of nuclear receptors in carcinogenesis as well as potential chemoprevention strategies [4,10,15,18,20,21,25-27], there lacks a thorough understanding of nuclear receptor and associated co-regulator alterations in HNSCC. To better characterize nuclear receptor signaling in head and neck carcinogenesis, we performed an in-depth interrogation of nuclear receptor expression and mutations. We previously performed microarray analysis of HNSCC tumors and reported a list of 2890 genes demonstrating differential expression in HNSCC tumors when compared with normal oral mucosa [28]. Here, we have created a nuclear receptor atlas in HNSCC from a query of 48 nuclear receptor and 19 nuclear receptor cofactor genes summarizing composite findings from The Cancer Genome Atlas (TCGA) tumor tissue data of 523 patients compared to normal tissue (analysis within cBioportal and the University of Alabama at Birmingham’s Cancer Data Analysis Portal (UALCAN)) in conjunction with Affymetrix data analyzing tumors (41 patients) and normal tissue (13 patients) from the University of Minnesota.

Methods

cBioportal and TCGA:

The TCGA PanCancer Atlas within cBioportal [29,30] was utilized for this study. Data within the platform includes 523 tumor samples, of which 515 have RNA-Seq data, with 82 adjacent normal samples for comparison. ‘Alterations’ to nuclear receptors were defined within cBioportal as genetic events, mutations, structural variants, copy number variations, or mRNA expression changes. Mutation frequency across all NR genes and cofactors in tumor specimens were compared to reference genome GRCh37 within cBioportal. Mutation calls included information on variant type, allele frequency, protein-level changes, and functional impact.

TCGA RNA-Seq data within the cBioportal platform comparing HNSCC samples (n = 515 patients/samples) to normal tissue (n = 82 patients/samples) was interrogated for significant differences in nuclear receptor expression. Genes with a z-score relative to normal samples (log RNA-Seq V2 RSEM normalized) greater than +2 or less than −2 in ≥50 % of cases were considered significant. Clinical impact and survival, methylation, expression correlations, and co-expression or mutual exclusivity of known or predicted effector proteins was subsequently analyzed for genes with significantly differential expression in tumor vs. normal. Unpaired, two-tailed t-tests assuming unequal variance (Welch’s t-test) were used to analyze comparisons in nuclear receptor genes’ z-scores across multiple clinical and pathologic parameters, p or q < 0.05 was deemed significant. These analyses were done only among subsets with at least 30 patients to ensure sufficient statistical power. Statistical stringency for gene differential expression conformed to cBioportal and UALCAN recommendations [31].

Network analyses were conducted using TCGA expression data. A Pearson correlation matrix was first created, with significant correlations (absolute correlation coefficient ∣r∣ > 0.5, p < 0.05) used for network construction. The Fructerman-Reingold layout was applied to improve readability. The clustered genes were sorted using the Louvain algorithm to uncover groups of genes that are more densely connected to each other than to nodes in other groups, depicted visually by the color of the nodes.

Patient Characteristics and Biopsy Samples from University of Minnesota cohort:

Surgical resection specimens from patients undergoing surgery for HNSCC were collected under standardized procedures and buccal mucosa punch biopsies of were obtained from healthy patients with no history of premalignant lesions or periodontal disease as described in Ginos et al. [28].

Microarray methods:

RNA extraction, cDNA conversion, and Affymetrix hybridization was performed as described in Ginos et al. [28]. Affymetrix array data was uploaded into the GeneData Cobi 4.0 database. Gene expression intensity for each array was scaled to a value of 1500 intensity units. Each intensity unit value was determined (Microarray Suite 5.0 algorithm by Refiner). Gene expression was compared between HNSCC samples and the control samples, using Welch’s t-test with p < 0.05 indicating significant difference. For genes with multiple expressed sequence tags (ESTs), low intensity ESTs were discarded if they were <20 % expression of the maximum.

Ethical and analytical considerations:

All human subjects research was conducting in accordance with the Institutional Review Board at the University of Minnesota and Clinical Protocol Review Committee approval of the Masonic Cancer Center. Statistical analysis was conducted within the cBioportal interface, Microsoft Excel, and in R Studio [32]. Portions of the code used in this project were generated with assistance from Open AI’s GPT model (ChatGPT).

Results

Forty-eight NR genes and 19 NR cofactors were clustered based on functional subtype, including thyroid hormone receptor-like (Subgroup 1A-1I), retinoid X receptor-like (Subgroup 2A-2E), estrogen receptor-like (Subgroup 3A-3C), nerve growth factor 1B-like (Subgroup 4A), steroidogenic factor-like (Subgroup 5A), germ cell factor-like (Subgroup 6A), atypical (Subgroup 0B), and nuclear receptor cofactors (Table 1) [5]. The subset of NR genes and cofactors analyzed within cBioportal were complementary to genes represented in our HNSCC genomic database from 41 tumor samples (13 normal oral mucosa samples) on a U133A Affymetrix chip.

Table 1.

Percent alteration among nuclear receptor subfamilies and nuclear receptor cofactors in HNSCC (n = 515 tumor samples).

Thyroid Hormone Receptor-Like (Subgroup 1A-1I)			Retinoid X Receptor- Like (Subgroup 2A-2E)			Estrogen Receptor-Like (Subgroup 3A-C)			Nerve GF-LIke (Subgrroup 4A)			Steroidogenic Factor- Like (Subgroup 5A)			Germ Cell Factor-Like (Subgroup 6a)			Atypical Family (Subgroup 0B)			Cofactors
Gene name	% altered		Genes	% altered		Genes	% altered		Genes	% altered		Genes	% altered		Genes	% altered		Genes	% altered		Genes	% altered
NR1C3 (PPARg)	68%		NR2B2 (RXRb)	39%		NR3C2 (MR)	43%		NR4A2 (TINUR)	16%		NR5A1 (SF1)	21%		CAR (SPG7)	8%		NR0B1 (DAX)	8%		NCOA1 (SRC1)	48%
NR1F3 (RORc)	67%		NR2C1 (TR2)	27%		NR3B3 (ERRg)	39%		NR4A1 (NAK1)	3%		NR5A2 (LRH-1)	3%		NR6A1 (GCNF)	7%		NR0B2 (SHP)	1.10%		PPARg coactivator 1a	39%
NR1F1 (RORa)	38%		NR2A1 (HNF4a)	23%		NR3C4 (AR)	39%		NR4A3 (MINOR)	2.30%		Overall	23%		Overall	15%		Overall	9%		HDAC6	38%
NR1F2 (RORb)	33%		NR2E1 (TLX)	20%		NR3B1 (ERRa)	37%		Overall	21%											TRAP37 (MED27)	33%
NR1A2 (TRb)	29%		NR2F6 (EAR2)	20%		NR3A2 (ERb)	19%														HDAC5	32%
NR1H2 (LXRb)	19%		NR2B1 (RXRa)	17%		NR3A1 (ERa)	17%														HDAC11	29%
NR1C2 (PPARd)	18%		NR2E3 (PNR)	13%		NR3C3 (PR)	12%														RAC3	26%
NR1B2 (RARb)	15%		NR2B3 (RXRg)	11%		NR3C1 (GR)	9%														NCOA3 (SRC3)	20%
NR1D2 (REV-ERBβ)	13%		NR2F1 (EAR3)	8%		NR3B2 (ERRb)	2.30%														TRAP95 (MED16)	22%
NR1H3 (LXRa)	12%		NR2C2 (TR4)	7%		Overall	87%														TRAP80 (MED17)	18%
NR1A1 (TRa)	10%		NR2A2 (HNF4g)	4%																	NRIP1	18%
NR1C1 (PPARa)	10%		NR2F2 (COUPTF2)	4%																	NCOR1	17%
NR1D1 (REV-ERBα)	8%		Overall	87%																	NCOR2 (SMRT)	17%
NR1I2 (PXR)	7%																				NCOA2 (SRC2)	15%
NR1B1 (RARa)	4%																				CBP (P300)	13%
NR1B3 (RARg)	3%																				7PPAR-binding protein (TRAP220)	13%
NR1I1 (VDR)	2.50%																				HDAC4	10%
NR1H4 (FXR)	1.70%																				HDAC7A	7%
Overall	95%																				HDAC9	6%
																					Overall	95%

Nearly all tumor samples (99 %) had some form of nuclear receptor gene alteration, a majority of which consist of expression changes relative to normal tissue (Table 1). Previous reports of the TCGA HNSCC data have demonstrated relatively high somatic mutation rate per sample, with the study ranking ninth out of the TCGA’s 80+ projects [33] and HNSCC more broadly having comparable rates of mutations to other tobacco-exposure related malignancies [34]. Somatic mutations among these tumor specimens (n = 496 patient specimens with mutation data) in NR genes and cofactors were less prevalent compared to gene expression changes, with the highest number of mutations occurring in RARG (thyroid-like, 13 mutations), NR2F2 (retinoid-like, 9 mutations), and NR3C2/MR (estrogen-like, 9 mutations). The 48 NRs ranked anywhere from the 21st-95th percentile (average 55th percentile, [n = 46, NR2C1 and NR2E3 had zero identified mutations]) for number of mutations compared to the 15,854 genes analyzed for mutations in the TCGA HNSCC data (Supplementary Figure 1, Panel A). Functional impact of the mutations across these three genes revealed a majority were somatic SNPs occurring in protein-coding domains and were functionally ‘deleterious’ or ‘likely deleterious’ based on summative analysis of AlphaMissense, SiFT, and Polyphen 2 (Supplementary Table 1). NR cofactors had a higher number of mutated samples overall with percentile ranking of the 15,854 genes in the TCGA HNSCC data ranging from 21st-99th, average of 74th percentile [n = 19], including CBP/P300 (37 mutations), NCOR1 (19), and NC0R2/SMRT (19) (Supplementary Figure 1, Panel B). These data suggest that the expression decreases we have previously reported for PPARG (confirmed below) [28] might be further affected by mutated cofactors in the transcriptional activation complex for these genes by conformational or other reasons.

NR expression changes were identified within HNSCC compared to normal tissue (Fig. 1, panel A). Peroxisome proliferator-activated receptor gamma (PPARG) and RAR Related Orphan Receptor C (RORC) were identified as significantly downregulated (average z-scores of −2.91±1.79 and −2.65±1.34, respectively, n = 515 tumor specimens). These expression changes were validated in UALCAN (n = 520) for both genes in HNSCC using the same TCGA dataset but different statistical methods (average z-score vs. ERL module to calculate IQR values after filtering outliers, followed by Welch’s T test).

Fig. 1.

Nuclear receptor expression changes in HNSCC relative to normal tissue. Panel A: mRNA expression z-scores relative to normal samples (log RNA seq V2 RSEM). The x-axis of the heatmap corresponds to the 515 patient samples with gene expression data relative to adjacent normal tissue. Genes in heat map are listed in descending order from most- least average expression change. Greater than +2 or less than −2 in >50 % of cases were considered significant (average z-score >+2 or <−2). PPARG and RORC were identified as significantly downregulated (average z-scores of −2.91 and −2.65, respectively) in the cBioPortal analysis. Of note, RORC and PPARG also had the highest gene alteration frequency of the queried nuclear receptors, with 68 % and 69 % altered for each gene (n = 515 patients/samples). More red samples indicate gene upregulation, more blue samples indicate gene down-regulation. Panel B: Expression of PPARG across cancer types (red bar) compared to normal tissue (blue bar), was significantly down regulated (red asterisk) in breast invasive carcinoma, colon adenocarcinoma, lung adenocarcinoma, lung squamous cell carcinoma, and thyroid carcinoma. Expression is significantly upregulated (green asterisk) in kidney chromophobe carcinoma, kidney papillary carcinoma, hepatocellular carcinoma, pancreatic adenocarcinoma, and stomach adenocarcinoma. PPARG methylation is significantly greater in HNSCC specimens with decreased PPARG expression. Panel C: Expression of RORC across cancer types (red bar) compared to normal tissue (blue bar), significantly down regulated (red asterisk) in cholangiocarcinoma, colon adenocarcinoma, esophageal adenocarcinoma, kidney renal clear cell carcinoma, kidney renal papillary carcinoma, lung squamous cell carcinoma, rectal adenocarcinoma (p = 6.40 × 10⁻³), and thyroid carcinoma (p = 3.54 × 10^−l2). RORC expression is significantly upregulated (green asterisk) in breast invasive carcinoma, hepatocellular, lung adenocarcinoma, prostate adenocarcinoma, and paraganglioma and pheochromocytoma. RORC methylation is significantly greater in HNSCC specimens with decreased RORC expression.

Differences in specific NR expression varied significantly with HPV status, organ site (oral cavity vs larynx/hypopharynx), tumor size (T1/2 vs T3/4), sex, tumor staging (stage I/II vs. stage III/IV); lymph node status (N0 vs. N), smoking history (smokers [93 % with >10 pack years] vs. non-smokers), and race (white vs. Black/African American) (Table 2A and 2B). Comparison groups that would be of interest but did not have sufficiently large sample sizes included metastatic status, oropharyngeal cancers, and T4b stage. Comparisons related to alcohol use were not included, as there was no validated consumption quantification in TCGA. Of the genes in our initial inquiry that demonstrated significant differential expression in tumor compared to normal adjacent tissue, we also found that PPARG and RORC were both significantly more downregulated in T3/4 tumors than in T1/2 tumors. RORC was also significantly more downregulated in Stage III/IV patients at presentation compared to Stage I/II patients at initial presentation.

Table 2A.

Significant gene expression changes relative to different clinical and pathologic factors for the 48 analyzed nuclear receptors (n = 515). Average z-scores for each gene and subset are listed, with significant changes (via Welch’s t-test) in comparison groups’ P values listed. Red cells = upregulation, blue cells = down regulation. Bold = significantly up- or down-regulated genes in that cohort (z-score ≥ +2 or ≤ −2).

	T1/2 (n=186)	T3/4 (n=321)	Welch’s P value			NO (n=243)	N+ (n=258)	Welch’s P value			Smokers (n=293)	Non- smokers (n=230)	Welch’s P value			HPV− (n=415)	HPV+ (n=72)	Welch’s P value			Stage I/II (n=117)	Stage III/IV (n=392)	Welch’s P value			Oral cavity (n=387)	hypophal ynx, larynx (n=124)	Welch’s P value			White (n=448)	Black/Africa n American (n=47)	Welch’s P value			Females (n=141)	Males (n=382)	Welch’s P value
LXRa	0.417026	0.165127	0.032826		LXRa	0.102366	0.426649	0.005095		SF1	1.333362	0.755893	0.034423		ESR2	0.769127	1.31081	0.000624		TR2	1.27485	0.834802	0.005548		ESR2	0.744804	1.207426	0.000897		COUPTF2	−0.182636	0.134711	0.006647		PPARD	0.936604	0.493516	0.000201
TR2	1.274201	0.737163	0.000193		RXRA	0.116896	−0.340222	0.000239		VDR	0.365755	0.581625	0.004562		LXRa	0.181391	0.762565	0.000834		SF1	0.351902	1.331996	0.000428		HNF4a	0.253237	1.06779	0.000211		THRB	−1.2636684	−0.7917574	0.03287783		VDR	0.716728	0.368317	6.15E-06
SF1	0.501664	1.465042	0.000418		ESR1	−0.50512	−0.231509	0.028509		RXRG	−0.802071	−0.978137	0.01888		TR2	0.888339	1.41635	0.009926		VDR	0.62393	0.411608	0.019094		SF1	0.52196	2.694507	1.31E-11		RORC	−2.589236	−3.107781	0.008268		MR	−1.847953	−1.626372	0.018026
PPARD	0.353978	0.753118	0.000853		ESRRB	−0.764156	−0.607416	0.012327							REV-ERBα	0.082567	0.481214	0.002565		RXRA	0.217654	−0.19549	0.002793		GCNF	0.078704	0.3418	0.017308							RORB	−1.133433	−0.606187	0.00029
GCNF	−0.004915	0.248675	0.009666		RARB	−0.959241	−0.73193	0.01555							HNF4a	0.605548	−0.201558	0.00492		ESRRG	−1.356477	−1.563021	0.021383		VDR	0.589447	0.105759	5.77E-08							RXRB	−1.389706	−0.44564	1.28E-05
ESR1	−0.071671	−0.55137	0.000244		RARG	−0.195154	−0.311998	0.015592							PPARD	0.818772	−0.430175	1.48E-12		RARG	−0.13391	−0.281531	0.007731		HNF4g	−0.574171	0.043397	3.55E-09							SPG7	−0.408056	−0.159584	0.015542
ESRRG	−1.305608	−1.633214	3.78E-05		RORB	−0.936745	−0.524456	0.009516							RARA	0.238242	−0.390521	2.74E-11		RORC	−2.39577	−2.721889	0.017605		COUPTF2	−0.199864	0.032743	0.00582							THRA	−0.880542	−0.610956	0.006333
FXR	−1.082544	−1.230875	0.011513												VDR	0.549584	−0.108099	2.41E-09		RXRG	−0.623928	−0.964876	0.000214		AR	−1.584973	−1.253655	0.003482
TR4	−0.10899	−0.381384	0.014544												ESR1	−0.575404	0.642363	9.52E-07							ESR1	−0.21801	−0.821426	2.40E-05
EAR3	−0.208334	−0.417889	0.040387												HNF4g	−0.529904	0.147488	1.36E-06							ESRRB	−0.655747	−0.785559	0.041798
COUPTF2	−0.016302	−0.221835	0.006551												LXRb	−0.172723	0.551738	0.000632							ESRRG	−1.594409	−1.29116	0.000419
MR	−1.372968	−1.853155	8.19E-07												PXR	−0.690565	0.023321	2.82E-05							EAR3	−0.428962	−0.076256	0.001764
PGR	−0.657631	−0.826038	0.044891												TR4	−0.447586	0.37435	2.85E-07							MR	−1.727789	−1.527631	0.033692
PPARG	−2.662045	−3.053017	0.020254												EAR3	−0.432816	0.042139	0.00127							NAK1	−0.820336	−0.656412	0.039025
RORB	−0.470014	−0.884266	0.015773												COUPTF2	−0.198419	0.053374	0.032685							PPARA	−0.672731	−0.282711	0.00027
RORC	−2.423389	−2.784851	0.003883												LRH-1	−0.198329	0.106158	0.018959							PPARG	−3.126324	−2.267456	2.71E-06
RXRB	−0.425859	−0.855861	0.034346												RORB	−1.166324	1.417844	4.58E-13							RARB	−0.956867	−0.596714	0.000919
RXRG	−0.735082	−0.970325	0.002806												RXRB	−0.897161	0.545272	3.35E-07							RXRA	−0.007423	−0.333526	0.02449
REV-ERBβ	−0.657645	−1.016518	6.93E-05												AR	−1.598136	−1.185481	0.027014							RXRG	−0.817038	−1.071329	0.004122
															ESRRB	−0.753222	−0.346378	0.002331							THRA	−0.792207	−0.362948	1.19E-05
															NR0B2	−1.26196	−0.929935	0.037346							TRB	−1.374314	−0.928753	0.0080671
															FXR	−1.210301	−0.958858	0.004755
															EAR2	−0.78487	−0.236803	0.004914
															MR	−1.87446	−0.709042	3.07E-12
															TINUR	−1.024941	−0.389285	4.09E-06
															PPARA	−0.675472	−0.223421	0.000376
															PPARG	−3.012223	−2.380939	0.007432
															RARB	−0.973554	−0.315568	5.68E-06
															REV-ERBβ	−1.089078	−0.029224	1.50E-14
															RORC	−2.776511	−2.112401	6.58E-05
															RXRG	−0.816103	−1.309553	1.35E-08

Table 2B.

Significant gene expression changes relative to different clinical and pathologic factors for the 19 analyzed nuclear receptor cofactors (n = 515).

	White (n=448)	Black/African american (n=47)	Welch’s P value			Females (n=141)	Males (n=382)	Welch’s P value			Smokers (n=293)	non- smokers (n=230)	Welch’s P value			T1/2 (n=186)	T3/4 (n=321)	Welch’s P value			Stage I/II (n=117)	Stage III/IV (n=392)	Welch’s P value			NO (n=243)	N+ (n=258)	Welch’s P value			HPV+ (n=72)	HPV− (n=415)	Welch’s P value			oral cavity (n=387)	Hypopharynx, larynx (n=124)	Welch’s P value
RAC3	0.95678023	1.53773617	0.02624534		MED17	0.73213111	0.06433974	0.00013929		RAC3	1.19747285	0.74389821	0.00108201		RAC3	0.81066995	1.10623354	0.0380405		MED27	1.19472348	1.60556788	0.02145617		MED17	0.43105436	0.03066429	0.00595571		HDAC9	0.05479028	0.43702482	0.00104804		MED1	0.39797658	0.82835285	0.00255556
NRIP1	0.26058432	−0.4607638	0.01103707		NRIP1	0.59275111	0.01965263	2.48E-05		HDAC11	−1.3770485	−1.0826737	0.02947658		MED27	1.312183	1.635322	0.04679931		HDAC5	−1.6048113	−1.3536715	0.0455584		NCOA1	−1.8831336	−2.146402	0.04239997		NCOA3	0.1071	0.78105928	9.29E-05		RAC3	0.84505658	1.39569431	0.00052221
					RAC3	0.51207111	1.17360053	1.88E-05							NRIP1	−0.0630923	0.31950791	0.00741279		HDAC4	−0.574653	−0.7476135	0.04650946							HDAC6	0.62705	−1.5381419	8.92E-17		EP300	−0.0849795	0.28651707	0.00106124
					HDAC5	−1.6950274	−1.3478626	0.0021209							HDAC11	−1.0099781	−1.3903332	0.00687767												MED17	−0.7941542	0.42922193	1.05E-12		HDAC11	−1.1572332	−1.579278	0.01658293
															HDAC4	−0.6030219	−0.7659513	0.04049116												NRIP1	−0.8857222	0.34281398	5.53E-10		NCOA1	−2.2092895	−1.5565756	3.72E-06
															HDAC6	−0.8161541	−1.4658646	0.00019996												HDAC11	−0.6448819	−1.347572	0.00175164		NCOR1	−0.7406155	−0.4629683	0.02341431
															PPARGC1A	−1.4923661	−1.6972804	0.03573539												HDAC5	−1.1669097	−1.4831231	0.01620406
																														NCOA1	−1.5816569	−2.1673913	4.13E-05

PPARG, significantly downregulated in HNSCC tumor specimens compared to normal tissue, is part of the thyroid hormone receptor-like family (Subgroup 1) of nuclear receptors. Activation results in heterodimerization with retinoid X receptors (RXRs), and its role is best characterized in the context of adipocyte differentiation and metabolic syndromes, though it has also been researched in aerodigestive cancer prevention [15-20]. Subsequent comparisons in cellular and clinicopathological (i.e. survival, histologic grade, tumor staging) characteristics were conducted, comparing patient specimens with low PPARG expression (‘altered,’ n = 356) to those with normal expression (‘unaltered,’ n = 159). There were no survival differences in HNSCC patient tumor specimens with low PPARG expression compared to normal PPARG expression specimens (Supplemental Figure 2). Likewise, no significant correlations in histologic grade or cancer staging were observed in patients with low vs normal PPARG expression. PPARG expression in tumors is inversely correlated with Winter and Buffa hypoxia scores. PPARG methylation [cg04632671] is significantly higher among tumors with decreased PPARG expression (q = 4.66 × 10⁻²¹). PPARG expression is also significantly downregulated compared to normal tissue in breast invasive carcinoma (p ≤ 1 × 10⁻¹²), colon adenocarcinoma (p = 4.36 × 10⁻⁷), lung adenocarcinoma (p = 1.62 × 10⁻¹²), lung squamous cell carcinoma (p = 1.62 × 10⁻¹²), and thyroid carcinoma (p = 4.44 × 10⁻³) (Fig. 1, panel B).

To identify putative mechanisms underlying efficacy of PPARγ agonists in HNSCC cancers, 77 known downstream effector genes of PPARγ [17] were examined to discern significant concurrent expression changes. ACOX1 (log2 odds ratio [log2OR] =1.16, q = 0.004), ACADL (log2OR=1.762, q < 0.001), LIPE (log2OR=1.205, q = 0.005), LIPG (log2OR=2.04, q = 0.049), and GPD1 (log20R = 1.022, q = 0.013) are functionally implicated in PPARγ dependent lipid metabolism and were significantly co-expressed with PPARG in HNSCC. CTNNB (log2OR = 1.219, q = 0.011), a downstream component of the Wnt signaling cascade, and CDK6 (log2OR = 0.865, q = 0.042), a critical element for cell cycle progression, were also significantly co-expressed with PPARG in HNSCC (Fig. 2, Panel A). In addition, we utilized the open-source database STRING to assess for possible functional interactions with PPARγ in Homo sapiens. The ten identified proteins had a significant gene expression correlation with PPARG in HNSCC, including five NR cofactors (NCOA1–2, NCOR1–2, PPARGC1A), CREBBP, MED1, RELA, RXRA, and SIRT1. However, only the NR cofactors NCOA1 (log2OR = 1.887, q < 0.001) and PPARGC1A (log2OR= 1.225, q < 0.001) had significant co-occurring expression with PPARG. The three genes with overall highest correlation with PPARG expression are RRAGD (related to mTOR and PI3K/AKT pathways, Spearman’s correlation=0.601, q = 1.20 × 10⁻⁴⁷), EYA2 (related to homologous and double strand repair pathways, Spearman’s correlation=0.596, q = 7.46 × 10⁻⁴⁷), and GULP1 (related to lipid and cholesterol transport, phagocytosis of apoptotic cells, Spearman’s correlation=0.573, q = 2.28 × 10⁻⁴²), all of which also had significant co-occurring expression changes with PPARG.

Fig. 2. PPARG related genes subgroup analysis in HNSCC.

Panel A: Co-occurring and mutually exclusive relationships between PPARG and known interacting genes by functional class in head and neck cancer specimens. Log2OR = log2 odds ratio. Panel B: Network analysis demonstrating correlations between hypothesized PPARG-related genes, with edges denoting significant correlations between gene pairs. Only correlations with an absolute Pearson correlation coefficient > 0.5 and a p-value <0.05 are shown. The clustered genes were sorted using the Louvain algorithm to uncover groups of genes that are more densely connected to each other than to nodes in other groups, displayed visually with different node colors.

These PPARG-related genes were subject to a network analysis of co-expression in the 515 TCGA HNSCC patients relative to normal tissue, depicted in Fig. 2, Panel B. The three genes with highest degree of centrality were NCOA2, CREBBP, and PTPN1. There were numerous genes with close association with PPARG, including ones not previously implicated in head and neck carcinogenesis. This includes protein encoding genes within various metabolic pathways like acyl-CoA dehydrogenase, long chain (ACADL) adiponectin (ADIPOQ), Glycerol-3-Phosphate Dehydrogenase 1 (GPD1), solute carrier family 2 member 4 (SLC2A4) and family 27 member 4 (SLC2A4), apolipoprotein E (APOE), fatty acid binding protein 5 (FABP5), and dock protein 1 (D0K1). Interestingly, immune related protein encoding genes like toll-like receptor 4 (TLR4) and leukotriene B4 receptor (LTB4R) also demonstrated relative proximity and connectivity to PPARG. Other genes with closer proximity to PPARG include canonical cancer-related genes, including BCL2, CDK inhibitors, and WNT1.

RORC was also significantly downregulated in HNSCC tumor specimens compared to normal tissue and is part of the thyroid hormone receptor-like family (Subgroup 1) of nuclear receptors. Upon activation, RORs bind as monomers to their nuclear response elements, and are implicated in regulation of circadian rhythm, normal thymopoiesis and cell-mediated immunity, disruption of which are hallmark features of cancer [35]. As with PPARG, subsequent comparisons in cellular and clinicopathological characteristics were conducted between patients with low RORC expression (‘altered,’ n = 353) relative to those with normal expression (‘unaltered,’ n = 162). There were no survival differences in HNSCC patient tumor specimens with RORC altered expression compared to unaltered specimens (Supplemental Figure 2). Likewise, no significant correlations in histologic grade or cancer staging were observed in patients based on RORC alteration. RORC methylation [cg25112191] is significantly higher among tumors with decreased RORC expression (q = 2.71 × 10⁻⁸). Down-regulation of RORC is correlated with higher Winter and Buffa hypoxia score (q = 2.746 × 10⁻⁴ for both), aneuploidy score (q = 1.741 × 10⁻³), and fraction of genome altered (q = 2.177 × 10⁻⁴). Decreased RORC expression is also associated with HPV negativity (q = 8.45 × 10⁻³). The three genes with highest expression correlation with RORC are LING04 (integral component of cell membrane, Spearman’s correlation=0.627, q = 3.02 × 10⁻⁵³), SELENBP1 (binds selenium, exhibiting potent anticarcinogenic effects, Spearman’s correlation=0.454, q = 1.56 × 10⁻²³), and PLIN4 (coating intracellular lipid deposits, Spearman’s correlation=0.448, q = 6.06 × 10⁻²³). RORC is also significantly downregulated compared to normal tissue in cholangiocarcinoma (p = 3.73 × 10⁻⁵), colon adenocarcinoma (p = 1.53 × 10⁻⁵), esophageal adenocarcinoma (p = 4.77 × 10⁻²), kidney renal clear cell carcinoma (p = 2.24 × 10⁻²), kidney renal papillary carcinoma (p = 7.80 × 10⁻⁹), lung squamous cell carcinoma (p ≤ 1 × 10⁻¹²), rectal adenocarcinoma (p = 6.40 × 10⁻³), and thyroid carcinoma (p = 3.54 × 10⁻¹²). (Fig. 1, panel C)

Among the 19 cofactors analyzed (Fig. 3), NCOA1 was significantly downregulated in tumors compared to normal tissue (Z=−2.05 ± 1.44). Decreased NCOA1 expression is associated with a significantly decreased progression free survival (PFS) (q = 0.0226), disease specific survival (DSS) (p = 0.0226), and overall survival (OS) (q = 0.0406) (Fig. 3, Panel C). Decreased NCOA1 expression is also correlated with higher Winter and Buffa hypoxia scores (q < 10⁻¹⁰ for both), higher microsatellite instability (q = 4.658 × 10⁻⁶), and HPV- status (q = 7.552 × 10⁻³). The three genes with highest expression correlation with NCOA1 are SOS1 (guanine exchange factor for RAS proteins, Spearman’s correlation=0.745, q = 7.34 × 10⁻⁸⁸), ATAD2B (ATPase, Spearman’s correlation=0.718, q = 1.12 × 10⁻⁷⁸), and BIRC6 (facilitates ubiquiti-nation, Spearman’s correlation=0.705, q = 7.85 × 10⁻⁷⁵). NCOA1 is also significantly downregulated compared to normal tissue in bladder urothelial carcinoma (p = 1.04 × 10⁻³), breast invasive carcinoma (p= <1 × 10⁻¹²), cervical squamous cell carcinoma (p = 1.05 × 10⁻²), colon adenocarcinoma (p = 1.09 × 10⁻¹²), glioblastoma multiforme (p = 2.48 × 10⁻²), kidney chromophobe (p = 1.47 × 10⁻⁶), kidney renal clear cell (p = <1 × 10⁻¹²), kidney renal papillary carcinoma (p = 1.48 × 10⁻¹³), lung adenocarcinoma (p = 2.62 × 10⁻¹¹), lung squamous cell carcinoma (p = 1.83 × 10⁻⁵), pancreatic adenocarcinoma (p = 7.20 × 10⁻³), rectal adenocarcinoma (p = 2.58 × 10⁻³), thyroid carcinoma (p = 9.84 × 10⁻¹⁰), and uterine corpus endometrial carcinoma (p= <1 × 10⁻¹²) (Fig. 3).

Fig. 3. Nuclear receptor cofactor expression changes in HNSCC relative to normal tissue.

Panel A: mRNA expression z-scores relative to normal samples (log RNA seq V2 RSEM). The x-axis of the heatmap corresponds to the 515 patient samples with gene expression data. Genes in heat map are listed in descending order from most-least average expression change. Greater than +2 or less than −2 in >50 % of cases were considered significant (average z-score >2+ or <−2). NCOA1 was identified as significantly downregulated (average z-scores of −2.05) in the cBioPortal analysis. Panel B: Expression of NCOA1 across cancer types (red bar) compared to normal tissue (blue bar), was significantly down regulated (red asterisk) in bladder urothelial carcinoma, breast invasive carcinoma, cervical squamous cell carcinoma, colon adenocarcinoma, glioblastoma multiforme, kidney chromophobe, kidney renal clear cell, kidney renal papillary carcinoma, lung adenocarcinoma, lung squamous cell carcinoma, pancreatic adenocarcinoma, rectal adenocarcinoma, thyroid carcinoma, uterine corpus endometrial carcinoma. Expression is significantly upregulated (green asterisk) in cholangiocarcinoma and hepatocellular carcinoma. Panel C: Progression free survival (n = 514, p = 0.0128, q = 0.0301), disease-specific survival (n = 489, p = 0.015, q = 0.0301), and overall survival (n = 514, p = 0.0243, q = 0.0324) all were significantly decreased in patients with decreased RORC expression (altered group, denoted in red) compared to patients with comparable RORC expression (unaltered group, denoted in blue) relative to normal adjacent tissue. Disease-free survival was not significantly altered (n = 128, p = 0.257, q = 0.257).

We subsequently constructed a gene network based on the correlation of nuclear receptor and nuclear receptor cofactor expression profiles across TCGA samples. This revealed an interesting cluster of genes, including apparent co-expression between RARB (a known and well-studied target of retinoids’ chemopreventive effect in HNSCC) with multiple estrogen receptor-like, thyroid receptor-like, and other retinoid receptor-like family members. This cluster indeed harbored numerous NR cofactors, including NCOA1. PPARG and RORC did not demonstrate significant co-expressivity with other NRs or cofactors (Fig. 4).

Fig. 4. Nuclear receptor and cofactor network analysis in HNSCC.

Network analysis demonstrating correlations between 48 NR and 19 NR cofactor genes, with edges denoting significant correlations between gene pairs. Only correlations with an absolute Pearson correlation coefficient > 0.5 and a p-value <0.05 are shown. The clustered genes were sorted using the Louvain algorithm to uncover groups of genes that are more densely connected to each other than to nodes in other groups, displayed visually with different node colors.

The significant changes in expression observed in the TCGA were also observed upon microarray analysis of 41 tumors compared to 13 normal, with PPARG, RORC, and NCOA1 also significantly downregulated in these specimens. Interestingly 23/43 NRs and 14/19 NR cofactors were significantly differentially expressed in the Affymetrix data. Seven NRs were significantly increased in expression and 16 significantly decreased in expression (Table 3).

Table 3.

Affymetrix chip based Nuclear Receptor Analysis. Nuclear receptors are characterized as demonstrating a change in expression in HNSCC, either Increased or Decreased, if >50 % of the examined expressed sequence tags (ESTs) demonstrated p < 0.05.

Increased	Decreased		No Change
RARa	TRb	NR2F6	TRa	TR4
TR2	RARg	ERRa	RARb	NR2F1
TLX	PPARd	MR	PPARa	ERa
GR	PPARg	AR	RORb	ERb
NR4A1	RORa	DAX-1	LXRb	ERRb
NR4A2	RORg	SHP	FXR	ERRg
NR4A3	RXRa		VDR	PR
	RXRb		HNF4A	LRH-1
	PNR		HNF4G	SF1
	NR2F2		RXRg	GCNF

Discussion

We identified nuclear receptor and nuclear receptor cofactors that demonstrated significant alteration and differential expression in HNSCC compared to normal tissue utilizing a combined approach with publicly available TCGA data within cBioportal and Affymetrix data from patients at the University of Minnesota. In the Affymetrix data, a total of 43 nuclear receptors were analyzed of which 7 receptors had increased expression, 19 receptors had decreased expression, and 19 were not significantly altered. In the TCGA data, the same 43 receptors were analyzed (in addition to five others). We felt it vitally important to interrogate this data to possibly generate additional proteins and processes that may be targeted for chemoprevention, treatment, or biomarker development. To the author’s knowledge, this type of comprehensive, high-level mutational and gene expression analysis of NR genes had not been conducted previously.

We discovered downregulation of multiple nuclear receptor family members, including RORC and PPARG. This is concordant with prior studies of carcinogenesis demonstrating global de-differentiation of cancer cells. Comparing patients with more-aggressive to less-aggressive disease phenotypes (i.e. T1/2 vs T3/4, stage I/II or Stage III/IV) revealed numerous nuclear receptor expression changes, including lower expression of PPARG and RORC. This corroborates previous work demonstrating that functional activation of PPARγ signaling decreases cellular proliferation and clonogenic potential [20,28]. Downregulation of RORA has been shown to decrease p53 activity in cancer cells, resulting in cellular proliferation [14]. Treatment with a RORα agonist reverses this effect, demonstrating potential utility for HNSCC or other cancers with demonstrated RORA downregulation. While less studied overall, there is emerging data exploring the utility of RORC’s impact on immune responsivity among different cancer types [36], In our NR gene and cofactor network analysis, PPARG and RORC did not demonstrate significant co-expressivity with any other nuclear receptors. However, we did identify a cluster that implicates a wider array of nuclear receptors that are significantly co-expressed with RARB. This unlocks mechanistic insights into the complex nuclear-receptor-driven regulatory network that may underly RARB and retinoid efficacy in HNSCC chemoprevention.

The PPARγ findings are particularly interesting for our work. The PPAR family of nuclear receptors is important for cellular differentiation, as well as metabolism and tumorigenesis, with variable effects on carcinogenesis depending on the disease type. Past work has demonstrated that restoration of PPARG expression or PPARG activation is favorable for oral cancer chemoprevention [10,12,18,20,21,25,37-44]. In HNSCC, studies have demonstrated a 35 % reduction in both aerodigestive cancer incidence in patients with diabetes treated with thiazolidinediones [45,46], which are known to act via agonism of the PPARγ receptor [47]. There have been multiple pre-clinical cellular studies, animal trials, and human clinical trials for thiazolidines (pioglitazone is the principal thiazolidine used clinically) [39,42,48-56]. We have led four cancer prevention targeted NCI-funded clinical trials investigating pioglitazone (Actos^®) for leukoplakia reversal (NCI CN-01–15,000, NCI CN-01–35,153), in a presurgical setting (ACTOplus MET XR^®) (NCT02917629) or in combination with metformin (Glucophage) (NCT05727761). Others have been performed in individuals with high risk of lung carcinoma [37,38,40,41,43,44]. An in vitro study of HNSCC cells demonstrated reduced proliferation on exposure to PPARγ agonist ciglitazone and RAR agonist 9-cis-retinoic acid [12]. These findings signify potential utility of both the RAR and PPAR pathways for head and neck cancer chemoprevention, independently and in tandem. In the TCGA HNSCC specimens, downregulation of PPARG was associated with co-occurring changes found via network analysis with effectors implicated in lipid metabolism and cell cycle progression. We hypothesize that PPARG’s complex interplay between glucose and lipid metabolism, cell growth, apoptosis, and immunomodulation garner pioglitazone’s chemopreventive efficacy. Based on the TCGA results, plausible mechanisms for PPARG downregulation include gene methylation, upstream cofactor expression changes, or genomic aberrations and/or somatic copy number aberrations [57]. Further studies of PPARγ alteration are warranted to better understand thiazolidinediones’ mechanism of action in tumor suppression and to identify cancers and patients most likely to garner benefit.

Somewhat expectedly, the TCGA database (523 patients, 2 genes downregulated) identified fewer differentially expressed genes than the Affymetrix data (41 patients, 23 genes differentially expressed). The differences may be attributed to strengthened statistical power due to larger sample size. Though the two studies used different technologies for analysis (RNA-Seq vs. microarray), we would anticipate correlation in findings in both cases based on prior work [58]. This underscores the value of publicly available large-scale genomic data to accompany smaller populations of potential interest. The TCGA data’s larger sample size also permitted subset analyses based on patient demographics, oncogenic risk factors, and cancer staging. We identified differential expression on these grounds in a variety of genes that can help inform stratification of clinical utility of drugs like thiazolidinediones moving forward.

We also examined nuclear receptor co-factors, which play a role in recruiting additional proteins to inhibit or activate cellular proliferation. Further it has been suggested that nuclear receptor cofactors may be cancer drug targets as well [59]. The proteins encoded by NCOA1–3 are co-activators that have histone acetyltransferase activity which function as part of a multi subunit co-activation complexes to enhance transcriptional activity of hormone nuclear receptors target genes, and are important in cancer progression and prognosis [15,59,60]. Interestingly, a 2016 study by Pavòn et al. demonstrated favorable prognosis with NCOA1 overexpression in HNSCC [61]. Our work demonstrated significant downregulation of NCOA1 in HNSCC in both TCGA (with increased downregulation in patients with node positive disease) and Affymetrix data. We also observed significantly decreased survival (PFS, DSS, OS) among patients observed among those with decreased NCOA1 expression in TCGA patients, highlighting the potential clinical significance to be garnered by targeting NC0A1.

Conclusions

In conclusion, interrogation of genomic data from head and neck cancer regarding nuclear receptor signaling pathways confirmed our prior data about PPAR family member expression in head and neck cancer. Further, new tools allowed us to perform a detailed analysis of gene defects in cancer that may result in nuclear receptor overexpression or downregulation. Information derived from these targeted analyses provides insight for the design of efficient and intelligent preclinical experiments that should allow for improved translation into oral cancer prevention trials.

Abbreviations:

DNA

Deoxyribonucleic acid

DSS

Disease specific survival

HNSCC

Head and Neck Squamous Cell Carcinoma

TCGA

The Cancer Genome Atlas

IRB

Institutional review board

NR

Nuclear Receptor

OS

Overall survival

PFS

Progression Free Survival

PPAR

Peroxisome Proliferator Activated Receptor Gamma

RAR

Retinoic acid receptors

ROR

Retinoic Acid Receptor Related Orphan Receptor

RXR

Retinoid X receptors

RNA

Ribonucleic acid

UALCAN

University of Alabama Cancer data analysis portal

Data availability

Data will be made available on request.