RES E A R C H Open Access
Microbiomic differences in tumor and
paired-normal tissue in head and neck
squamous cell carcinomas
Hannah Wang
1,2†
, Pauline Funchain
1,3†
, Gurkan Bebek
6,7
, Jessica Altemus
1
, Huan Zhang
1,2
, Farshad Niazi
1
,
Charissa Peterson
1
, Walter T. Lee
5
, Brian B. Burkey
4
and Charis Eng
1,2,3,8,9,10*
Abstract
Background: While the role of the gut microbiome in inflammation and colorectal cancers has received much
recent attention, there are few data to support an association between the oral microbiome and head and neck
squamous cell carcinomas. Prior investigations have been limited to comparisons of microbiota obtained from
surface swabs of the oral cavity. This study aims to identify microbiomic differences in paired tumor and non-tumor
tissue samples in a large group of 121 patients with head and neck squamous cell carcinomas and correlate these
differences with clinical-pathologic features.
Methods: Total DNA was extracted from paired normal and tumor resection specimens from 169 patients; 242
samples from 121 patients were included in the final analysis. Microbiomic content of each sample was determined
using 16S rDNA amplicon sequencing. Bioinformatic analysis was performed using QIIME algorithms. F-testing on
cluster strength, Wilcoxon signed-rank testing on differential relative abundances of paired tumor-normal samples,
and Wilcoxon rank-sum testing on the association of T-stage with relative abundances were conducted in R.
Results: We observed no significant difference in measures of alpha diversity between tumor and normal tissue
(Shannon index: p = 0.13, phylogenetic diversity: p = 0.42). Similarly, although we observed statis tically significantly
differences in both weighted (p = 0.01) and unweighted (p = 0.04) Unifrac distances between tissue types, the
tumor/normal grouping explained only a small proportion of the overall variation in the samples (weighted R
2
=0.
01, unweighted R
2
< 0.01).
Notably, however, when comparing the relative abundances of individual taxa between matched pairs of tumor
and normal tissue, we observed that Actinomyces and its parent taxa up to the phylum level were significantly
depleted in tumor relative to normal tissue (q < 0.01), while Parvimonas was increased in tumor relative to normal
tissue (q = 0.01). These differences were more pronounced among patients with more extensive disease as
measured by higher T-stage.
Conclusions: Matched pairs analysis of individual tumor-n ormal pairs revealed significant differences in relative
abundance of specific taxa, namely in the genus Actinomyces. These differences were more pronounced among
patients with higher T-stage. Our observations suggest further experiments to interrogate potential novel mechanisms
relevant to carcinogenesis associated with alterations of the oral microbiome that may have consequences for the
human host.
Keywords: Head and neck squamous cell carcinoma (HNSCC), Bacteria, Microbiome
* Correspondence: engc@ccf.org
†
Equal contributors
1
Genomic Medicine Institute, Lerner Research Institute, Cleveland, OH 44195,
USA
2
Cleveland Clinic Lerner College of Medicine, Cleveland, OH 44195, USA
Full list of author information is available at the end of the article
© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Wang et al. Genome Medicine (2017) 9:14
DOI 10.1186/s13073-017-0405-5
Background
Interactions between microbes and carcinogenesis
within the host have been studied for decades. The best
example is in the context of a single microorganism:
Helicobacter pylori and its proven link with gastric can-
cer [1]. At the other end of the spectrum, and more
recently, Fusobacter ium nucleatum was described in the
gut of those with advanced colorectal cancer [2, 3]. Sub-
sequent functional studies demonstrated F. nucleatum
to be capable of both upregulating inflammatory and
oncogenic pathways in colon cancer cell lines [4] and in-
ducing adenomas in mice [5]. The bacterial microbiome,
defined as the total collection of bacteria that inhabit
any environmental niche, has been increasingly recog-
nized as an active participant in human body functions
and proposed to be an organ in its own right. At a basic
level, we have long understood that the microbiome
serves to maintain homeostasis. Dysbiosis, or disruption
of the normal flora, can result in pathogenic overgrowth
of organisms including Clostr idium and Candida in the
gastrointestinal and vaginal tracts, respectively [6–8]. Simi-
larly, the oral microbiome has long been studied in the
context of dental caries: ingestion of excessive carbohy-
drates promotes overgrowth of acidogenic and acidophilic
microbes, inducing a local drop in pH, demineralization of
enamel, and subsequent breakdown of tooth [9, 10].
While investigation of the association between microbial
dysbiosis and colorectal cancer is progressing at a rapid
pace, the study of the bacterial microbiome in other areas
of the gastrointestinal tract has lagged behind. In particu-
lar, head and neck squamous cell carcinomas (HNSCC),
which account for more than half a million cancer cases
annually around the world [11, 12], has received relatively
little attention. This may be due to the fact that HNSCC is
a heterogeneous disease entity, encompassing a variety of
cancers from different disease sites, and develops from the
mucosal linings of the upper aerodigestive tract, compris-
ing: (1) the nasal cavity and paranasal sinuses; (2) the
nasopharynx; (3) the oral cavity and oropharynx; and (4)
the hypopharynx, larynx, and trachea [13]. Additionally,
major risk factors such as smoking, alcohol consumption,
and human papillomavirus infection have already been
elucidated [13, 14].
However, recent studies have identified microbiomic
shifts in the oral cavity associated with cigarette smoking,
and in the gastrointestinal tract associated with alcohol
consumption [15, 16]. Similar to the pathogenesis of
dental caries, it is feasible that the microbiome helps
transduce an environmental exposure into a carcinogenic
effect. As there are few effective systemic therapies in
HNSCC, and toxicity of local treatment is often significant
due to the vital structures involved, identification of a
microbial pathway to disease may offer new insights into
targeted therapies and primary prevention.
Prior work investigating the microbiome of head and
neck cancer, including the largest cohort that was previ-
ously reported by our group, provided descriptive evidence
of the “in”vironment of the head and neck at higher-order
taxa and suggested that microbial variation correlates with
clinical outcomes and gene methylation status [17]. Smaller
studies that have used superficial sampling of oral cavity
cancers by means of oral swabs observed differences in
taxonomic abundance between normal and tumor surfaces
primarily at the phylum level [18, 19]. However, bacteria in
the head and neck are clearly not limited to the mucosal
surface, but in fact populate deep tissue [17, 20, 21]. These
findings are supported by our prior pilot study as well as a
recent investigation composed of 29 patients with exclu-
sively laryngeal cancer that demonstrated phylum and
genus-level changes in tumor relative to normal tissue [22].
However, the significance of the findings from these studies
is unclear. Furthermore, the microbiome content of cancer-
ous mucosal tissue compared to adjacent histologically
normal tissue has not been examined outside of the setting
of laryngeal cancer.
With increasing evidence that a rich community of
bacteria exists within head and neck tissues and may
contribute to carcinogenesis, we now seek to identify
microbiomic differen ces between tumor and histolog i-
cally normal tissue in a large cohort of patients with
HNSCC of the oral cavity, oropharynx, hypopharynx,
and larynx. In this study, we report on the largest human
tissue microbiome study in HNSCC patients, with 16S
ribosomal DNA (rDNA) amplicon sequencing of paired
normal-tumor tissue samples from 121 unrelated parti-
cipnts. With these data, we correlate whole microbiome
communities of head and neck tissue with clinical out-
come measures of HN SCC, in order to test the hypoth-
esis that microbiomes either alter or have bee n altered
by both the presence and extent of HNSCC.
Methods
Patient cohort and sample collection
From 2003 to 2014, consecutive HNSCC patients were
enrolled into a tissue biorepository collection. The tissue
banking protoco l was designed specifically to maintain
sterility for downstream microbiome analysis. All tissues
banked were required to be collected from the oral
cavity, oropharynx, hypopharynx, or larynx. In this regis-
try, 169 individuals had available paired adjacent normal
and tumor tissue. A total of 30–50 mg each of paired
tumor and norma l tissue, approximately 2 cm from the
tumor edge, were sterilely collected in the operating
room, classified via pathology review, flash frozen, and
stored at –80 °C. Relevant clinicopathologic features
were collected prospectively at the time of diagnosis.
Tumor node metastases staging was determined for each
primary tumor based on American Joint Committee on
Wang et al. Genome Medicine (2017) 9:14 Page 2 of 10
Cancer guidelines [23]. Missing data were filled in via
retrospective chart review; individuals without available
data were noted as such in Table 1.
DNA extraction
Total DNA was extracted with modifications from a
previously described protocol [17]. Bead homogenization
of tissues was performed with a TissueLyser II (Qiagen,
Venlo, The Netherlands). Also added was a yeast cell
wall lysis step using the Masterpure Yeast DNA Purifica-
tion kit (Epice ntre, Madison, WI, USA) [24]. All beads,
tubes, and non-enzymatic reagents were treated with
ultraviolet light for at lea st 30 min prior to use [25]. Re-
agent controls were confirmed by 16S polymerase chain
reaction (PCR) to be absent of contaminating bacteria.
16S rRNA gene sequencing
PCR of the V1–V4 hypervariable regions of the 16S rRNA
gene was performed with previously published primers
[17]. PCR was performed under the following conditions:
95 °C for 5 min, followed by 32 cycles of 95 °C for 1 min,
55 °C for 40 s, 70 °C for 80 s, and an extension of 72 °C
for 10 min. PCR products were electrophoresed on a
1% agarose gel, purified using a Zymoclean DNA Gel
Recovery kit (Zymo, Orange, C A, USA ), and cloned
into a StrataClone pSC ve ctor (Agilent, Santa Clara,
CA, USA) [17]. From an initial 169 pairs of patient
samples, a total of 318 tissue samples from 159 distinct
patients had positive 16S rDNA P CR product re covery.
Ninety-five colonies w ere picked per tissue sample.
Plasmid inserts were PCR amplified using standard T3/
T7 primers, then Sanger sequenced (ABI3730xl, Life
Technologies, Carlsbad, CA , USA).
Bioinformatic analysis
Reads were filtered for quality, trimmed, and compiled
using a custom python script. Depth of coverage was set
at 60 sequences or higher based on leveling off of the
Shannon diversity index at 60 reads. Due to this cutoff, a
total of 242 tissue samples from 121 distinct patients
were included in the final analysis. Subsampled open-
reference operational taxonomic unit (OTU) picking
[26] against Greengenes (version 13.8) [27, 28] at 97%
similarity threshold using UCLUST [29], alignment with
PyNAST [30], phylogenetic tree construction using
FastTree (version 2.1.3) [31], and subsequent computa-
tion of alpha (Shannon diversity index, phylogenetic
diversity) [32, 33] and beta diversity measures (weighted
and unweighted Unifrac distances) [34, 35] was per-
formed using QIIME (version 1.9.1) [36].
Statistics
Student’s t-tests and likelihood ratio tests were used to
compare continuous and categorical demographics/
Table 1 Demographics and clinical characteristics of patients
a
Variable Included
(n = 121)
Excluded
(n = 38)
p value
Age (years) 63 ± 11 62 ± 13 0.73
Male 74 (64) 25 (74) 0.31
Race 0.07
White 71 (91) 19 (100)
Black 7 (9) 0 (0)
Localization 0.32
Oral cavity/Oropharynx 72 (65) 26 (74)
Floor of mouth 5 2
Tongue 42 11
Tonsil 13 8
Oral cavity NOS 12 5
Hypopharynx/Larynx 38 (35) 9 (26)
Hypopharynx 4 2
Larynx 34 7
T-stage 0.59
Low T-stage 44 (40) 15 (45)
T0 4 2
T1–T2 42 13
High T-stage (T3–T4) 66 (60) 18 (55)
N-stage 0.58
Node negative (N0) 56 (51) 15 (45)
Node positive 54 (48) 18 (55)
N1–N2 51 18
N3 3 0
Overall stage 0.16
I–II 24 (24) 4 (13)
III–IV 78 (76) 28 (88)
Previous treatment
Operation 21 (19) 4 (12) 0.37
Chemotherapy 24 (21) 3 (9) 0.09
Radiotherapy 30 (26) 8 (24) 0.78
Smoking history 0.08
Current 18 (16) 1 (3)
Past 68 (60) 24 (71)
Never 29 (25) 9 (26)
Alcohol use 0.96
Heavy 10 (9) 4 (12)
Social 57 (50) 16 (47)
History 10 (9) 3 (9)
Never 37 (32) 11 (32)
Values are presented as means ± standard deviations or
number (percent)
a
Data are missing for the following variables, indicated as “variable
name: # missing in included group/# missing in excluded group”:
Age: 7/5, Gender: 6/4, Localization: 11/3, Race: 43/19, T-stage 9/5, N-
stage 11/5, Overall stage: 19/6, Prior operation: 8/5, Prior chemother-
apy: 5/4, Prior radiation: 5/4, Smoking history: 6/4, Alcohol use: 7/4.
Percentages are calculated from denominator of samples with
known data
NOS not otherwise specified
Wang et al. Genome Medicine (2017) 9:14 Page 3 of 10
clinical factors, respectively, between patient samples in-
cluded in the final analysis and those excluded due to in-
sufficient reads. Student’s t-tests were used to compare
Shannon index and phylogenetic diversity between
tumor and non-tumor samples at a sequencing dept h of
60 with ten iterations per sample. Distance matrices of
the tumor and non-tumor samples were compared using
the Adonis statistical method [37]. This method is simi-
lar to non-parametric analysis of variance (ANOVA) and
relies on F-tests based on sequential sums of squares
derived from 1000 permutations on the weighted and
unweighted UniFrac distance matrices, with the null
hypothesis that there is no difference in community
structure between groups. To compare relative abun-
dances of taxa between matched tumor-normal pairs, we
used the non-parametric two-sided Wilcoxon signed-
rank test. To compare relative abundances of taxa
between samples of different T-stages, we used the
Wilcoxon rank-sum.
All a nalyses were conducted in JMP Pro 12 (SAS
Institute Inc., Cary, N C , USA) or R version 3.2.2. All
statistical tests were two-sided, with a p value < 0.05
or false discovery rate (FDR) adjusted q <0.05 consid-
ered statistically significant. All graphs w ere created
using the R package lattice [38]. The cladogram was
created using GraPhlAn on Gala xy [39, 40].
Results
HNSCC microbiomes are similar on a phylum-level to
those in previous studies of human oral flora
We analyzed sterilely collected, paired fresh-frozen
normal-tumor samples from 121 patients with HNSCC.
These patients were not significantly different on any
demographic or clinical factors when compared to the
38 patients excluded based on low read count (Table 1).
The taxonomic composition of our HNSCC samples is
similar to that identified in our previous pilot study of
HNSCC [17], as well as with data from previously pub-
lished studies on the human oral microbiome [41–43].
Firmicutes is the predominant phylum, followed by
Bacteroidetes and Proteobacteria, then by Fusobacteria
and Actinobacteria, in both tumor and adjacent normal
samples from HNSCC patients as well as in prior studies
(Fig. 1). Phyla falling under 0.1% relative abundance in
our dataset were not included in this analysis.
HNSCC tumor and paired-normal tissue are not significantly
different on measures of alpha or beta diversity
The average number of reads for the 242 patient sam-
ples in the final analysis was 83 ± 11 and did not differ
between tumor (84 ± 13) and normal ( 83 ± 7) samples
(p = 0.48). The average read length wa s 745 ± 117. To
determine whether overall mean diversity was different in
tumor and adjacent normal tissue of HNSCC patients, we
compared two measures of alpha diversity: Shannon index
(H) which measures the evenness and richness of a popu-
lation; and phylogenetic diversity (PD) which takes the
phylogenetic relationship between taxa into account. We
found no significant difference in measures of alpha diver-
sity between tumor (H = mean 3.72 ± standard error 0.78,
PD = 6.42 ± 1.88) and normal (H = 3.87 ± 0.74, PD = 6.62
± 1.96) tissue (H: p = 0.13, PD: p =0.42).
To test whether ove rall bacterial taxa composition
was diff erent between tumor and normal tissue, we
used principal coordinates analy sis (PCoA) on weighted
and unweighted Unifrac distances. We found that ,
although statistically significantly different on both
weighted (p = 0.012) and u nweighted (p = 0.042) mea-
sures, the tumor/normal grouping explained only a
small proportion of the overall variation in the samples
(Fig. 2a, Additional file 1: Figure S1A, B). This differ-
ence wa s also similarly significant (weighte d p = 0.0 01,
unweighted p = 0.001) but non-explanatory when compar-
ing PCoAs of samples by whether they were from the oral
cavity/oropharynx or the hypopharynx/larynx (Fig. 2b).
Relative abundance of specific taxa differs between
tumor and paired normal tissue
Next, we compared the relative abundances of 372 individ-
ual taxa between matched pairs of tumor and adjacent nor-
mal tissue, finding differences in ten genera, 12 families,
eight orders, five classes, and three phyla by Wilcoxon
signed-rank testing (Additional file 2: Figure S2). Only 2/10
genera were significant after adjusting for FDR: Actinomyces
and Parvimonas.ThegenusActinomyces, along with its
parent family Actinomycetaceae, order Actinomycetales,
class Actinobacteria, and phylum Actinobacteria, was
depleted in tumor compared to matched normal tissue. In
contrast, the genus Parvimonas, along with its parent fam-
ily Tissierellaceae, was increased in tumor compared to
normal tissue (Fig. 3).
After identif ying taxa that were significantly different
between tumor and paired normal tissues, we performed
a stratified analysis to investigate the relationship be-
tween tumor stage and the relative abundances of these
taxa. We observed that samples from low-stage (T0–2)
patients had significantly increased relative abundance of
the genus Actinomyces compared to samples from
high-stage ( T3–4) patient s (median 3.3% versus 1.2%,
p = 0.005). The parent taxa of the genus Actinomyces
were also significantly relatively increa sed in low-stage
patients compared to higher stages, up to the phylum
level. In contrast, the genus Parvimonas was significantly
relatively decreased in samples from low-stage patients
compared to high-stage patients (median 0.0% versus
1.1%, p = 0.023). The relationship between these taxa and
T-stage remained consistent when stratifying by tumor
versus paired-normal tissue (Fig. 4a). This difference was
Wang et al. Genome Medicine (2017) 9:14 Page 4 of 10
statistically significant in the normal group (phylum Actino-
bacteria p = 0.002, genus Actinomyces p = 0.023, genus Par-
vimonas p = 0.033), but only approached significance in the
tumor group (phylum A ctinobacteria p =0.067,genus Acti-
nomyces p =0.052,genusParvimonas p = 0.247).
As T-stage was significantly associated with tissue loca-
tion (oral cavity/oropharynx versus hypopharynx/larynx),
we proceeded to stratify samples based on tissue location
(Fig. 4b). We observed that relative abundances of the
phylum Actinobacteria, genus Actinomyces, and genus
Parvimonas were consistently lower at hypopharyngeal/la-
ryngeal locations relative to the oral cavity/oropharynx.
However, when analyzing oral cavity/oropharynx samples
alone, Actinobacteria and Actinomyces approached signifi-
cance in low-stage patients relative to high-stage patients
(p =0.100, p =0.192) and Parvimonas remained signifi-
cantly relatively decreased among low-stage patients com-
pared to high-stage patients (p = 0.006). When analyzing
hypopharyngeal/laryngeal samples alone, Actinobacteria
remained significantly relatively increased in low-stage
patients (p = 0.031), while Actinomyces and Parvimonas
were not significantly different between low-stage and
high-stage groups (p =0.645,p =0.790).
Discussion
In this study, we sought to describe the oral microbiome
of individuals with HNSCC and to compare the local
microbiome of their tumors with neighboring normal
tissue. We hypothesized that tumor tissue would have a
microbiome unique from that of adjacent normal tissue
and be more pronounced in higher stage disease. The
simple comparison of tumor versus adjacent normal tis-
sue did not reveal major shifts in overall diversity (Shan-
non index or phylogenetic diversity) or in microbiomic
content. However, matched pairs analysis of individual
tumor-normal pairs revealed significant differences in
relative abundance of specific taxa, namely the genera Ac-
tinomyces and Parvimonas. These differences were more
pronounced in patients with a higher T-stage.
The phylum-level oral microbiome of individuals in our
study was similar to those reported previously. Dewhirst
et al. reported on the Human Oral Microbiome Database,
which consisted of 633 Sanger-sequenced oral 16 s rRNA
gene libraries from various head and neck sites of patients
of various states of health and disease [41]. Ahn et al. ana-
lyzed oral washes from 20 individuals (ten with malignant
or premalignant oral lesions, ten healthy controls) using
both 16 s rRNA pyrosequencing and a custom DNA
microarray [42]. Segata et al. found in their study of over
200 healthy adults that the adult digestive tract micro-
biome differed according to location of sampling; group 1
(G1) sites (buccal mucosa, keratinized gingiva, and hard
palate) had increased relative abundance of Firmicutes
and decreased relative abundance of other phyla as com-
pared to group 2 (G2) sites (saliva, tongue, tonsils, and
throat) [43]. The phyla-level composition of our study
population was most similar to Segata et al.’s G2 series,
despite the fact that they used next-generation sequencing
(NGS) instead of Sanger sequencing, used swabs instead
of surgically excised tissue, and had healthy controls in-
stead of patients with HNSCC. This was not surprising
given that the majority of our patient tissues were from
Fig. 1 Relative abundances of major phyla in the human oral microbiome. Bar plot of relative abundances of major phyla in the oral microbiome
observed in this study and three previously published series. There were similar relative abundances of the most common phyla among tumor
(orange) and adjacent normal (blue) tissue from this study. Additionally, these abundances were similar to previously published series describing
the oral microbiome
Wang et al. Genome Medicine (2017) 9:14 Page 5 of 10