Plasma metabolic phenotypes of HPV-associated vs smoking-associated head and neck cancer and patient survival

Abstract

Background:

Metabolic differences between human papillomavirus (HPV)-associated head and neck squamous cell carcinoma (HNSCC) and smoking-associated HNSCC may partially explain differences in prognosis. The former relies on mitochondrial oxidative phosphorylation (OXPHOS) while the latter relies on glycolysis. These differences have not been studied in blood.

Methods:

We extracted metabolites using untargeted liquid chromatography high-resolution mass spectrometry from pretreatment plasma in a cohort of 55 HPV-associated and 82 smoking-associated HNSCC subjects. Metabolic pathway enrichment analysis of differentially expressed metabolites produced pathway-based signatures. Significant pathways (P<0.05) were reduced via principal components analysis and assessed with overall survival via Cox models. We classified each subject as glycolytic or OXPHOS phenotype and assessed it with survival.

Results:

Of 2,410 analyzed metabolites, 191 were differentially expressed. Relative to smoking-associated HNSCC, bile acid biosynthesis (P<0.0001) and octadecatrienoic acid beta-oxidation (P=0.01), were upregulated in HPV-associated HNSCC, while galactose metabolism (P=0.001) and vitamin B6 metabolism (P=0.01) were downregulated; the first two suggest an OXPHOS phenotype while the latter two suggest glycolytic. First principal components of bile acid biosynthesis (HR=0.52 per standard deviation, 95% CI:0.38–0.72, P<0.001) and octadecatrienoic acid beta-oxidation (HR=0.54 per sd, 95% CI:0.38–0.78, P<0.001) were significantly associated with overall survival independent of HPV and smoking. The glycolytic vs OXPHOS phenotype was also independently associated with survival (HR=3.17, 95% CI:1.07–9.35; P=0.04).

Conclusions:

Plasma metabolites related to glycolysis and mitochondrial OXPHOS may be biomarkers of HNSCC patient prognosis independent of HPV or smoking. Future investigations should determine if they predict treatment efficacy.

Impact:

Blood metabolomics may be a useful marker to aid HNSCC patient prognosis.

Introduction:

Head and neck squamous cell carcinoma (HNSCC) can be divided into two groups that are clinically, genetically, and possibly metabolically distinct: human papillomavirus-associated HNSCC and smoking-associated HNSCC (sometimes viewed as HPV-unassociated). The former consists of HPV-positive oropharyngeal tumors with little to no patient smoking history, while the latter consists of HPV-negative oropharyngeal and all other HNSCC sites from patients with a substantial smoking history (i.e., >10 pack-years); HPV-positive oropharyngeal patients with a substantial smoking history are likely a mix of HPV-associated and smoking-associated HNSCC. Clinically, HPV-associated HNSCC is less aggressive, responds better to treatment, and on average patients live two-to-seven times longer than smoking-associated HNSCC¹. Genetically, “hot-spot” locations in the PIK3CA oncogene are mutated in HPV-associated tumors, while TP53 tumor suppressor loss-mutations are in most smoking-associated tumors^2,3. These genes are known to regulate tumor metabolism. For instance, wild-type p53 dampens glycolysis and promotes mitochondrial oxidative phosphorylation (OXPHOS) by regulating glucose transporters (GLUTs) and key enzymes in the glycolysis pathway⁴.

It has been hypothesized that loss or inactivation of p53 may promote the Warburg effect⁵, i.e., the reprogramming of tumor metabolism to favor glycolysis at the expense of OXPHOS⁶. Noticed in the 1920’s by Otto Warburg, this atypical metabolic state provides the energy and macromolecules for rapid proliferation and metastasis^7–10. It links to nearly all hallmarks of cancer¹¹. The Warburg effect can be seen, literally, via 18F-Fluorodeoxyglucose positron emission tomography (18F-FDG-PET)¹². Higher tumor uptake of the fluorescent glucose marker means higher glycolytic activity, which is directly associated with poor prognosis and treatment resistance in HNSCC^13,14. In fact, the Warburg effect may be a factor in HNSCC treatment resistance. Increased glycolysis and decreased OXPHOS promotes resistance to ionizing radiation therapy¹⁵ and studies suggest modulating glycolysis may re-sensitize a tumor^16,17. Recent studies in HNSCC cell lines suggest that a Warburg-like metabolic phenotype differentiates HPV-positive from HPV-negative cells^18,19. The Warburg effect may thus explain, in part, the differences in presentation, progression, and survival observed between these HNSCC groups. However, it has yet to be investigated as a potential blood biomarker.

Our goal was to investigate the metabolic differences of HPV-associated HNSCC relative to smoking-associated HNSCC in pretreatment patient plasma using untargeted high-resolution metabolomics (HRM) and test whether those differences were prognostic of overall survival independent of HPV status. HRM is the comprehensive measurement of small molecules arising from metabolic reactions occurring in a biospecimen – i.e., metabolites²⁰. It uses gas or liquid chromatography paired with high-resolution mass-spectrometry to quantify thousands of metabolites via their mass-to-charge ratio (m/z)²¹. The metabolites are matched to reference databases²² or identified with authentic standards²³. While single metabolite investigations used to be the norm, recent advances in high-throughput measurement, bioinformatics, and pathway analysis have shifted the field towards a systems-biology approach, gaining further insight into disease via dysregulated metabolic pathways and profiles^24–26.

Materials and Methods:

Cohort:

This was a prospective cohort of HNSCC patients recruited at radiation oncology clinics affiliated with Emory University Hospital from 2013 to 2016 before undergoing intensity-modulated radiation therapy (IMRT) with or without chemotherapy²⁷. In this analysis, all patients gave written informed consent to participate in the parent study and all ancillary studies. Recruitment and data collection for the parent study and the current ancillary study were guided by the ethical principles of the Belmont Report and reviewed and approved by the Emory University Institutional Review Board. Inclusion criteria were histological HNSCC with no distant metastasis, ≥21 years of age, and no evidence of uncontrolled metabolic, hematologic, cardiovascular, renal, hepatic or neurologic disease. Patients with simultaneous primaries or major psychiatric disorders were excluded. Demographic and clinical variables were collected through chart review and standardized questionnaires at study entry. Prior to IMRT, blood was collected into chilled EDTA tubes for the immediate isolation of plasma. Plasma was separated by centrifugation at 1000g for 10 minutes at 4°C then aliquoted into siliconized polypropylene tubes and stored at −80°C.

Classification of HPV-associated and smoking-associated HNSCC:

HPV status was ascertained via review of the tumor pathology report. We deemed a subject HPV-positive if p16 was positive or if HPV DNA was detected. Smoking status was ascertained via questionnaire of tobacco habits; the questionnaire did not allow calculation of pack-years. We defined our smoking-associated HNSCC group as all current or former smokers with a non-oropharyngeal or HPV-negative oropharyngeal tumor; non-oropharyngeal tumors are rarely attributed to HPV²⁸ and thus not uniformly tested. We defined our HPV-associated HNSCC group as all subjects who were HPV-positive never smokers. We endeavored to mitigate group misclassification thus we deliberately excluded HPV-positive oropharyngeal patients with a smoking history. We believe these patients are a mix of smoking-associated and HPV-associated HNSCC and by including them would introduce misclassification. In a 2010 seminal paper, Ang et. al. demonstrated that HPV-positive oropharyngeal smokers with >10 pack-years – deemed as having “Intermediate risk” for survival – are clinically distinct from the “Low risk” group (i.e., HPV-positive, ≤10 pack-years) and the “High risk” group (i.e., HPV-negative, >10 pack-years)²⁹.

Overall survival:

Vital status was ascertained via linkage to the Georgia Comprehensive Cancer Registry maintained at Emory University. Overall survival was defined as the time from study entry to death by any cause, or to the last contact up to August 2020.

High-resolution untargeted metabolomics (HRM) of blood plasma:

We used liquid chromatography-mass spectrometry (LC-MS) protocols developed at the Emory Clinical Biomarkers Laboratory^30,31. Frozen aliquots from randomized samples were removed from −80°C, thawed, and extracted with ice-cold acetonitrile. Supernatants were added to a 4°C autosampler. Each sample was analyzed in triplicate (10μL) using hydrophilic interaction liquid chromatography (HILIC) with positive electrospray ionization and Fourier transform Orbitrap high-resolution mass spectrometry (Dionex Ultimate 3000, Q-Exactive HF, Thermo Scientific). Analyte separation was achieved using a 2.1mm × 100mm × 2.6μm Accucore HILIC column (Thermo Scientific) and a gradient elution of 2% formic acid, water, and acetonitrile starting at 10%, 10%, 80%, for 1.5 min with a linear increase to 10%, 80%, 10% at 6 min and held for 4 min per injection.

The high-resolution mass spectrometer operated at 120,000 resolution and an m/z range of 85.0000–1275.0000. Probe temperature, capillary temperature, sweep gas and S-Lens RF levels were 200°C, 300°C, 1 arbitrary units (AU), and 45 AU, respectively. Positive tune for sheath gas, auxiliary gas, sweep gas and spray voltage settings were 45 AU, 25 AU and 3.5 kV, respectively. Spectral peaks were extracted and aligned using apLCMS³² and xMSanalyzer³³. Mass, retention time, and ion intensity were recorded for unique m/z spectral features. Features with a coefficient of variation ≥75% were removed. Feature intensities were median summarized across triplicates and corrected for any batch effects³⁴. Features were log₂ transformed and quantile normalized. We restricted our investigation to m/z features annotated via the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Human Metabolome (HMDB) reference databases²².

Statistical Analysis:

We identified differences in metabolic pathways of HPV-associated HNSCC relative to smoking-associated HNSCC via feature selection (step 1) followed by metabolic pathway enrichment analysis (step 2). We reduced each significant pathway via principal components analysis (PCA) (step 3) and estimated the first principal components (PC) with overall patient survival via Cox-models (step 4). Based on the pathway results, we classified each patient, post-hoc, into a binary glycolytic or OXPHOS metabolic phenotype and estimated that classification with overall survival (step 5).

Step 1: We selected m/z features that differentiated HPV-associated vs smoking-associated HNSCC in any of three models routinely used in untargeted metabolomics analyses: age at study entry, sex, and race adjusted logistic regression using a Benjamini-Hochberg false discovery rate P<0.10; Partial least squares discriminant analysis using a variable importance in projection>2; Support vector machine using recursive feature elimination. Step 2: We identified enriched metabolic pathways from the features using Mummichog, a pathway analysis tool that maps m/z to KEGG metabolic networks and uses resampling to obtain a Fisher’s exact test P-value²⁴ – a similar methodology to Gene set enrichment analysis. Pathways with ≥3 metabolite hits and P<0.05 met our threshold for statistical significance. We note the Identification Level³⁵ of the pathway metabolites and plot a heatmap. Step 3: We performed PCA on the metabolites that comprised each significant pathway. Step 4: The 1^st PC for each pathway was z-standardized and regressed on overall patient survival using Cox proportional hazards adjusted for HPV and smoking status, age at study entry, sex, race, tumor stage and patient ECOG status; we found no statistically significant interaction by these variables. Further adjustment for body mass index, alcohol consumption, marriage status, treatment, feeding tube, and time of blood draw did not change our results and were not included in final models. The hazard ratios estimated the increase in mortality per 1 standard deviation (sd) change in the pathway-specific PC. We conducted a sub-analysis restricting to oropharyngeal and non-oropharyngeal carcinoma subjects to investigate whether results hold across tumor site. Step 5: We dichotomized the PCs at their study medians and used the three most significant pathways to classify each subject as having a glycolytic or an OXPHOS phenotype (i.e, if ≥2 of the 3 pathways indicated glycolysis the patient was classified that way and vice versa with OXPHOS). We regressed the glycolytic vs OXPHOS phenotype on overall survival with a Cox model and plotted a Kaplan-Meier survival curve.

We used publicly available R packages for feature selection (xmsPANDA 1.1.76)³⁶, pathway analysis (Mummichog 2.0)²⁴, and metabolite annotation (xmsAnnotator 1.3.2)²². PCA used the “prcomp” R package. Cox regression used SAS version 9.4³⁷.

Results:

Compared to smoking-associated, the HPV-associated HNSCC subjects on average were younger (56.8 years vs 60.1), more obese (29.0 kg/m² vs 26.1), white (87% vs 72%) and male (89% vs 62%) (Table 1). As expected, the HPV-associated tumors were predominantly in the oropharynx (96%) and less advanced (29% were stage IV vs 64% for smoking-associated). Also expected, the smoking-associated subjects had higher mortality than HPV-associated group (unadjusted hazard ratio (HR)=7.6, 95% confidence interval (CI)=2.8, 25.5; P<0.001; Kaplan-Meier curve in Supplementary Figure 1).

Table 1.

Demographics and clinical characteristics of HPV-associated and smoking-associated HNSCC groups

		HPV-associated (n=55)		Smoking-associated (n=82)		P value
Death*	reported death	4	8%	24	34%
	living/censored	47	92%	47	66%
Survival time (months)	mean (sd)	46.3	17.8	37.3	20.6	0.01
Age (years)	mean (sd)	56.8	7.8	60.1	10.5	0.05
BMI (kg/m2)	mean (sd)	29.0	4.7	26.1	5.3	0.001
Sex	Men	49	89%	51	62%	<0.001
	Women	6	11%	31	38%
Race	white	48	87%	59	72%	0.03
	All others	7	13%	23	28%
Alcohol use*	<1 drink per wk	28	52%	49	60%	0.32
	>=1 drink per wk	26	48%	32	40%
Tumor site	Oropharynx	53	96%	12	15%	<0.001
	All others	2	4%	70	85%
Tumor Stage*	I–III	38	70%	29	41%	<0.001
	IV	16	30%	52	59%
ECOG status*	0 (fully active)	39	74%	24	32%	<0.001
	1 (restricted)	13	25%	38	50%
	2 (unable to work)	1	2%	14	18%
Treatment	Radiotherapy only	10	18%	17	21%	0.43
	Chemoradiotherapy	45	82%	65	79%

HPV=human papillomavirus; HNSCC=head and neck squamous cell carcinoma; sd=standard deviation; IQR=interquartile range;

^*indicates a variable that was missing in ≥1 subject.

From 16,342 extracted features, 3,075 matched to KEGG or HMDB reference databases. After filtering for missing values, 2,410 were left for analysis. The three feature selection models identified 191 metabolites that were differentially expressed between HPV-associated and smoking-associated HNSCC (Supplementary Table 1 and Supplementary Figure 2).

From 91 pathways, pathway enrichment analysis identified six statistically significant metabolic pathways from the 191 metabolites (Figure 1 and Supplementary Table 2). Notably, all six pathways are relevant to mitochondrial OXPHOS or glycolysis – demonstrating a potential Warburg-like metabolic phenotype. The pathways related to OXPHOS, i.e., bile acid biosynthesis (P<0.0001), 21-carbon steroid metabolism (P=0.007), and octadecatrienoic acid beta-oxidation (P=0.012), were upregulated in HPV-associated HNSCC and down-regulated in smoking-associated HNSCC. The pathways related to glycolysis, i.e., galactose metabolism (P=0.001), vitamin B6 metabolism (P=0.014), and starch and sucrose metabolism (P=0.02), were the opposite. Four pathways implicated in HNSCC did not meet significance but followed the Warburg-like pattern: selenoamino acid metabolism (P=0.065), phosphatidylinositol phosphate (P=0.076), and hexose phosphorylation (P=0.091) were upregulated in smoking-associated HNSCC, while omega-3 fatty acid metabolism (P=0.065) was upregulated in HPV-associated HNSCC.

Figure 1.

The metabolic pathways that were upregulated (green) or downregulated (blue) in HPV-associated relative to smoking-associated HNSCC. P values for each pathway are labeled. The dashed red line indicates statistical significance at the P<0.05 level. The pathways were identified through the metabolic pathway enrichment analysis program, Mummichog version 2.0, with 191 metabolites that differentiated HPV-HNSCC from Smoking-HNSCC in three feature selection models (Multivariable logistic regression, partial least squares discriminant analysis, and support vector machines). The first number following the pathway is the number of significant metabolites in the pathway found in the 191. The second number is the number of metabolites Mummichog mapped to that pathway from the 2,410 total.

For subsequent analyses, we reduced the six pathways to four because of metabolite overlap (five of the six 21-carbon steroid metabolites overlapped with bile acid biosynthesis, and all three of the starch and sucrose metabolites overlapped with galactose). In Table 2, we annotate the 23 metabolites that comprised the bile acid biosynthesis (n=9), galactose metabolism (n=6), octadecatrienoic acid beta-oxidation (n=5), and vitamin B6 metabolism (n=3) pathways. Twelve features were unique (*=identified via an authentic standard): adenosine 5’-monophosphate*, dihydroxy-5beta-cholestane, cholic acid, galactosylglycerol, 3-ketolactose, lactose*, inosine diphosphate, raffinose*, stachyose*, pyridoxamine*, pyridoxic acid*, and pyridoxamine 5-phosphate. The other 11 had multiple matches. A heatmap shows the relative intensity of each metabolite and the pathways between HPV-associated and smoking-associated HNSCC (Figure 2). In general, HPV-associated patients had higher levels of bile acid biosynthesis and octadecatrienoic acid beta-oxidation and lower levels of galactose metabolism and vitamin B6 metabolism; the smoking-associated HNSCC patients had the opposite.

Table 2.

Metabolite annotation for twenty-three metabolites mapped to four different metabolic pathways that differentiate HPV-associated HNSCC from smoking-associated HNSCC

m/z	timea	KEGG compound IDb	Compound namec	Adduct	ID Leveld	FCe	PCf
Bile acid biosynthesis pathway
348.0692	102	C00020	Adenosine 5’-monophosphate	M+H[1+]	1	−0.83	−0.18
403.3568	36	C06340; C05451; C15610; C05500; C05502; C15519; C13550; C03594	Hydroxycholesterol	M+H[1+]	3	1.05	0.35
405.3726	36	C05452	Dihydroxy-5beta-cholestane	M+H[1+]	3	1.73	0.39
409.2942	36	C00695	Cholic acid	M+H[1+]	3	1.20	0.23
419.3517	36	C06341; C05458; C05453; C05501; C15518; C15520; CE5530; C05499; C05445	Dihydroxycholesterol	M+H[1+]	3	1.37	0.38
421.3673	37	C05454; C05444	Trihydroxy-5beta-cholestane	M+H[1+]	4	2.54	0.37
435.3466	36	C04554; CE4872; C01301	Dihydroxy-5beta-cholestanate	M+H[1+]	4	0.81	0.38
437.3615	37	CE0232; C05446; CE1272	Tetrahydroxy-5beta-cholestane	M+H[1+]	3	0.93	0.35
453.3571	36	CE4874; CE1277; CE1278; CE1279	5beta-cholestane-pentol	M+H[1+]	3	1.25	0.30
Galactose metabolism pathway
277.0894	74	C05401	Galactosylglycerol	M+Na[1+]	3	−1.60	0.12
341.1106	41	C05403	3-Ketolactose	M+H[1+]	3	2.79	−0.40
365.1054	99	C00243	Lactose	M+Na[1+]	1	−1.59	0.51
429.023	27	C00104	Inosine diphosphate	M+H[1+]	3	−0.59	0.10
527.1585	153	C00492	Raffinose	M+Na[1+]	1	−1.33	0.52
689.2079	249	C01613	Stachyose	M+Na[1+]	1	−1.50	0.54
Octadecatrienoic acid beta-oxidation pathway
237.1486	36	CE5311; CE5323	Hydroxy-tetradeca-trienoic acid	M[1+]	4	0.32	0.52
239.1641	36	CE5313; CE5312; CE5325; CE5324	Hydroxy-tetradeca-dienoic acid	M[1+]	4	0.47	0.49
263.1639	37	CE5314; CE5326	Hydroxy-hexadeca-tetraenoic acid	M[1+]	3	1.38	0.37
265.1796	35	CE5315; CE5316; CE5327; CE5328	Hydroxy-hexadeca-trienoic acid	M[1+]	4	1.13	0.39
281.1748	37	CE5321; CE5309	3-oxo-hydroxy-hexadeca-dienoic acid	M[1+]	4	0.69	0.44
Vitamin B6 metabolism pathway
169.0971	43	C00534	Pyridoxamine	M+H[1+]	1	−0.47	0.78
184.0597	46	C00847	Pyridoxic acid	M+H[1+]	1	−0.94	0.38
249.0611	59	C00647	Pyridoxamine 5’-phosphate	M+H[1+]	3	0.29	−0.51

m/z=mass-to-charge, time=retention time in seconds HPV=human papillomavirus, HNSCC=head and neck squamous cell carcinoma, FC,=fold change, PC=principal component

^aMetabolite extraction retention time in seconds

^bMummichog-based mapping to Kyoto Encyclopedia of Genes and Genomes (KEGG) database identification number

^cm/z features that matched to multiple reference metabolites are named as specifically as the matching allowed

^dIdentification Confidence Level for metabolite annotation^23,35 (1=laboratory confirmed standard, 3=tentative structural identification with a match to a reference database, 4=molecular formula identification via isotope abundance distribution, charge state, adduct determination)

^eLog₂ fold change association with HPV-associated HNSCC relative to smoking-associated HNSCC

^fStandardized loading factor for the 1^st principal component of each metabolic pathway

Figure 2.

Hierarchical clustering heatmap of HPV-associated (HPV) and smoking-associated (SMK) HNSCC and 23 z-transformed metabolites that comprised the four metabolic pathways: bile acid biosynthesis (n=9 metabolites), galactose metabolism (n=6), octadecatrienoic beta-oxidation (n=5), and vitamin B6 (n=3). The HPV-associated HNSCC patients, in general, have higher levels of bile acid biosynthesis and octadecatrienoic acid beta-oxidation metabolites (red) and lower levels of galactose metabolism and vitamin B6 metabolism metabolites (blue) relative to the smoking-associated patients. The smoking-associated HNSCC patients, in general, have the opposite, higher levels of galactose metabolism and vitamin B6 metabolism metabolites and lower levels of bile acid biosynthesis and octadecatrienoic acid beta-oxidation metabolites.

We reduced each pathway to its first principal component (PC) and regressed that on overall mortality using adjusted Cox regression (PCA plots in Supplementary Figure 3; standardized loading factors in Table 2). Three of the four pathways were significantly associated with overall mortality in unadjusted models (Table 3). The bile acid biosynthesis and the octadecatrienoic acid PCs were inversely associated with mortality (HR=0.52 per sd, 95% CI:0.38–0.72, P<0.001; and HR=0.54 per sd, 95% CI:0.38–0.78, P<0.001, respectively). The galactose PC was positively associated mortality (HR=1.72 per sd, 95% CI:1.09–2.72, P=0.02); Kaplan-Meier curves in Supplementary Figure 4). To clarify, a subject at the lowest 20^th percentile in the bile acid PC would have one-third the mortality of a subject at the highest 20^th percentile (HR=0.52^1.68 sd=0.33). Most importantly, the associations for bile acid and octadecatrienoic acid were independent of HPV status, smoking history, age at study entry, and sex (Table 3). In fact, bile acid biosynthesis remained statistically significant after further adjustment for race, tumor stage, and patient ECOG status (HR=0.58 per sd, 95% CI:0.37–0.92, P=0.02). Restricting to tumor site had little effect on the bile acid biosynthesis pathway estimate but modest effects on the other pathways’ estimates; fewer deaths among the oropharyngeal subjects may have precluded statistical significance (Table 3).

Table 3.

First principal components of metabolic pathways that differentiate HPV-associated HNSCC from smoking-associated HNSCC and their association with overall patient mortality via Cox models

	HR	CI	P value
Bile acid biosynthesis 1 st PC
Unadjusted model	0.52	(0.38, 0.72)	<0.001
HPV, smoking, age, sex adjusted model	0.52	(0.36, 0.76)	<0.001
Fully adjusted modela	0.58	(0.37, 0.92)	0.02
Fully adjusted model among oropharyngeal tumor subjectsa	0.45	(0.17, 1.17)	0.10
Fully adjusted model among non-oropharyngeal subjectsa	0.50	(0.27, 0.94)	0.03
Galactose 1 st PC
Unadjusted model	1.72	(1.09, 2.72)	0.02
HPV, smoking, age, sex adjusted model	1.28	(0.79, 2.10)	0.32
Fully adjusted modela	1.21	(0.71, 2.08)	0.48
Fully adjusted model among oropharyngeal tumor subjectsa	1.85	(0.53, 6.49)	0.34
Fully adjusted model among non-oropharyngeal subjectsa	1.04	(0.60, 1.80)	0.89
Octadecatrienoic acid beta oxidation 1 st PC
Unadjusted model	0.54	(0.38, 0.78)	<0.001
HPV, smoking, age, sex adjusted model	0.65	(0.44, 0.97)	0.03
Fully adjusted modela	0.75	(0.51, 1.10)	0.14
Fully adjusted model among oropharyngeal tumor subjectsa	0.29	(0.08, 1.02)	0.05
Fully adjusted model among non-oropharyngeal subjectsa	0.74	(0.47, 1.16)	0.19
Vitamin B6 1 st PC
Unadjusted model	1.26	(0.86, 1.86)	0.23
HPV, smoking, age, sex adjusted model	0.93	(0.61, 1.41)	0.74
Fully adjusted modela	1.13	(0.75, 1.72)	0.56
Fully adjusted model among oropharyngeal tumor subjectsa	0.51	(0.17, 1.53)	0.23
Fully adjusted model among non-oropharyngeal subjectsa	1.35	(0.84, 2.18)	0.22

HR=Cox hazard ratio, CI=confidence interval, HPV=human papillomavirus, PC=principal component

^aadjusted for HPV and smoking status, age at study entry, sex, race, tumor stage, and patient ECOG status

We used median dichotomized PCs from bile acid biosynthesis, galactose metabolism and octadecatrienoic acid beta-oxidation, to classify each subject with a glycolytic or mitochondrial OXPHOS phenotype (Figure 3B). The Kaplan-Meier curve shows stark separation in survival (P<0.0001) with the OXPHOS phenotype having a 93% estimated three-year survival compared to 59% for the glycolytic phenotype (Figure 3A). That survival difference was larger than what we observed for HPV-associated vs smoking-associated HNSCC, at 88% and 69%, respectively (Supplementary Figure 1). The glycolytic vs OXPHOS Cox model showed the metabolic phenotype was significantly associated with mortality independent of HPV status and smoking history, age at study entry, sex, race, tumor stage, and patient ECOG status (HR=3.17, 95% CI:1.07, 9.35; P=0.04). Among HPV-associated subjects, 24% were classified as having the glycolytic phenotype (74% OXPHOS). Among the smoking-associated subjects, 34% of were classified as having the OXPHOS phenotype (66% glycolytic). This suggests that a glycolytic/OXPHOS phenotype may indeed be an independent prognostic biomarker and not merely an extension of HPV-positivity or smoking status.

Figure 3.

The classification scheme and survival curve of the glycolytic vs OXPHOS metabolic phenotype. Panel A) Kaplan-Meier survival curve of HNSCC subjects classified as having a glycolytic or mitochondrial OXPHOS metabolic phenotype. The estimated three-year survival probability is 93% for subjects with an OXPHOS phenotype vs. 59% for subjects with a glycolytic phenotype; the survival curves are statistically significant (logrank P<0.0001). The glycolytic vs OXPHOS Cox model estimated hazard ratio adjusted for HPV and smoking status, age, sex, race, tumor stage, and patient ECOG status is HR=3.17 (95% CI:1.07, 9.35; P=0.04). Panel B) The classification scheme for the glycolytic vs mitochondrial OXPHOS metabolic phenotype based on the median-dichotomized principal components of bile acid biosynthesis, octadecatrienoic acid beta-oxidation, and galactose metabolism (High=above median, Low=below median). Each subject was classified glycolytic or OXPHOS based on the direction of ≥2 pathways.

Discussion:

In this prospective study, we observed a glycolytic vs mitochondrial OXPHOS metabolomic phenotype in HNSCC patient plasma that was associated with patient survival independent of key clinical factors including HPV and smoking. Relative to smoking-associated HNSCC, the metabolic pathways of bile acid biosynthesis and octadecatrienoic acid beta-oxidation were upregulated in HPV-associated patient plasma, while galactose metabolism and vitamin B6 metabolism were downregulated. These pathways help regulate the balance between OXPHOS and glycolysis – a balance frequently dysregulated in tumors, termed the Warburg effect.

Four prior studies have investigated serum or plasma metabolomics in HNSCC patients^38–41, but to our knowledge, this is the first study to investigate metabolic differences of HPV-associated vs smoking-associated HNSCC. Perhaps the best comparison to our study is one by Jung et. al. that studied HPV-positive versus HPV-negative HNSCC cell lines¹⁸. Jung reported that HPV-positive cells rely on mitochondrial OXPHOS for energy metabolism while HPV-negative cells rely on glycolysis – results nearly identical to ours. Jung also found HPV-negative cells had increased levels of hypoxia-inducible factor-1α (HIF-1α) and hexokinase II (HK-II), regulating enzymes of the hexose phosphorylation pathway, the first step in glycolysis. We found the same pathway upregulated in smoking-associated HNSCC though the result was not statistically significant. Importantly, Jung found that pharmacologic reversal of the glycolytic phenotype increased a cell’s sensitivity to radiation therapy, suggesting a potential drug target. The similarity of our findings in patient blood to those in cell cultures suggests that circulating metabolites may be putative biomarkers of HNSCC tumor metabolism.

The Warburg-like plasma phenotype we observed has two parts. The first part was the dysregulated pathways related to glycolysis (i.e., galactose, starch and sucrose, phosphatidylinositol phosphate, and hexokinase metabolism). Of our findings, glycolysis has the strongest prior evidence. It is a hallmark of HNSCC proliferation and survival, driven by oncogenic signaling (e.g., p53 loss and PI3KCA mutation), altered metabolic enzymes (e.g., GLUT-1, HK-II), and a hypoxic tumor environment (e.g., HIF-1α)^42,43. Tumor hypoxia activates HIF-1α, a transcription factor which increases a tumor’s reliance on glycolysis while reducing mitochondrial OXPHOS⁴³. Using gene expression profiles, Keck et al. found hypoxia to be the discriminating feature of the basal HNSCC phenotype – a phenotype made up of HPV-negative smokers⁴⁴. Higher levels of glycolytic metabolites found in the plasma of smoking-associated HNSCC patients may indeed be the downstream effect of tumor hypoxia. Hypoxia, and its glycolytic switch, is associated with aggressive disease⁴⁵, treatment failure¹⁶, and poor survival⁴⁶ in HNSCC. Glycolytic metabolites in serum are significantly associated with HNSCC relapse⁴⁰, and in our study, upregulated glycolytic pathways in plasma were associated with a more aggressive tumor type and poor survival.

The second part of the Warburg-like phenotype we observed was the dysregulated pathways related to mitochondrial OXPHOS (i.e., bile acid biosynthesis, octadecatrienoic acid beta-oxidation, and omega-3 fatty acid metabolism). Bile acid biosynthesis was our most important finding. It was significantly upregulated in HPV-associated HNSCC and was positively associated with longer survival independent of HPV, smoking, and other clinical factors. It is understudied in HNSCC but deserves closer examination. Bile acids are synthesized in the liver from cholesterol to aid in the digestion of dietary fat. That process, however, is regulated via a host of nuclear receptor signaling pathways, notably FXR, TGR5, and PPAR, that are now recognized for suppressing cancer progression^47–50. Bile acids may also be related to tumor hypoxia as they have been shown to destabilize HIF-1α in prostate, lung, and mammary cancer cell lines⁵¹. Conversely, tumor hypoxia may represses CYP7A1, a key bile acid biosynthesis enzyme though many factors may affect this pathway⁵².

Like bile acids, we found octadecatrienoic acid beta-oxidation (i.e., oxidation of 18-carbon lipids such as omega-3 and omega-6 fatty acids) upregulated in HPV-associated HNSCC and positively associated with longer survival independent of HPV and smoking. Pro-oncogenic pathways generally promote synthesis of fatty acids over oxidation because proliferating cells require lipids for biological membranes⁵³. Apart from lipoprotein(a), serum lipids appear to decrease as HNSCC progresses^54–59 and short-chain fatty acids may suppress HNSCC growth⁶⁰. Our findings agree that higher levels of lipid oxidation, in general, is a beneficial marker in HNSCC.

The Warburg effect is likely just one aspect of tumor metabolism linked to HNSCC progression. Molecular changes underlying HNSCC progression, such as p53 loss, EGFR mutation, and mutations in the signaling of PI3K-AKT, NOTCH1, and ALK1, parallel metabolic changes in the tumor^10,61–64. Metabolomics may thus provide a rich source of HNSCC biomarkers particularly if metabolic reprogramming drives treatment resistance. Resistance to radiation therapy, cisplatin chemotherapy, and cetuximab immunotherapy may be directly related to reprogrammed glucose metabolism and mitochondrial-generated oxidative stress^15,65,66. Thus, metabolic biomarkers may give an early indication of HNSCC treatment efficacy and may be targets for novel metabolic therapies^10,64.

The recruitment of the parent study limited our metabolome-wide analysis to HNSCC patient plasma and not tumor specimens. Additionally, we did not have healthy controls as a comparison group. However, we analyzed plasma metabolic differences in two important clinical subtypes and assessed those differences with survival – something no prior study has done. The exclusion of HPV-positive patients with a smoking history gave us distinct HPV-associated and smoking-associated groups but may limit our generalizability. The extracted metabolites are likely influenced by recent diet and the subjects did not fast before blood draw. Information on diet was not available, but we did adjust the models for factors that influence diet such as smoking, alcohol, age, sex, BMI, and the time of the blood draw. Lastly, untargeted metabolomics is a “shotgun” approach that measures the relative intensity of each mass-to-charge feature and we cannot be certain of the annotation, outside of laboratory standards²³. To improve confidence in our annotation, we restricted our analysis to metabolites matched to the KEGG and HMDB reference databases. However, a feature may match to multiple reference metabolites and we cannot distinguish between them.

In conclusion, we observed a Warburg-like metabolomic phenotype in HNSCC patient plasma that may aid patient prognosis. We expound on the metabolic pathways that differentiate HPV-associated from smoking-associated HNSCC, two groups who differ in their presentation, treatment sensitivity, and survival. We are working to independently validate the results we observed with bile acids, lipid oxidation, and galactose metabolism so that they warrant future investigations as predictors of treatment efficacy and as therapeutic targets for treatment resistant tumors.