RES E AR C H Open Access
Characterization of oral and gut
microbiome temporal variability in
hospitalized cancer patients
Jessica R. Galloway-Peña
1
, Daniel P. Smith
6
, Pranoti Sahasrabhojane
1
, W. Duncan Wadsworth
7
, Bryan M. Fellman
2
,
Nadim J. Ajami
6
, Elizabeth J. Shpall
5
, Naval Daver
3
, Michele Guindani
8
, Joseph F. Petrosino
6
,
Dimitrios P. Kontoyiannis
1
and Samuel A. Shelburne
1,4*
Abstract
Background: Understanding longitudinal variability of the microbiome in ill patients is critical to moving microbiome-
based measurements and therapeutics into clinical practice. However, the vast majority of data regarding microbiome
stability are derived from healthy subjects. Herein, we sought to determine intra-patient temporal microbiota variability,
the factors driving such variability, and its clinical impact in an extensive longitudinal cohort of hospitalized cancer
patients during chemotherapy.
Methods: The stool (n = 365) and oral (n = 483) samples of 59 patients with acute myeloid leukemia (AML) undergoing
induction chemotherapy (IC) were sampled from initiation of chemotherapy until neutrophil recovery. Microbiome
characterization was performed via analysis of 16S rRNA gene sequencing. Temporal variability was determined using
coefficients of variation (CV) of the Shannon diversity index (SDI) and unweighted and weighted UniFrac distances per
patient, per site. Measurements of intra-patient temporal variability and patient stability categories were analyzed for
their correlations with genera abundances. Groups of patients were analyzed to determine if patients with adverse
outcomes had significantly different levels of microbiome temporal variability. Potential clinical drivers of microbiome
temporal instability were determined using multivariable regression analyses.
Results: Our cohort evidenced a high degree of intra-patient temporal instability of stool and oral microbial diversity
based on SDI CV. We identified statistically significant differences in the relative abundance of multiple taxa amongst
individuals with different levels of microbiota temporal stability. Increased intra-patient temporal variability of the oral
SDI was correlated with increased risk of infection during IC (P=0.02), and higher stool SDI CVs were correlated with
increased risk of infection 90 days post-IC (P=0.04). Total days on antibiotics was significantly associated with increased
temporal variability of both oral microbial diversity (P=0.03) and community structure (P=0.002).
Conclusions: These data quantify the longitudinal variability of the oral and gut microbiota in AML patients, show that
increased variability was correlated with adverse clinical outcomes, and offer the possibility of using stabilizing taxa as a
method of focused microbiome repletion. Furthermore, these results support the importance of longitudinal
microbiome sampling and analyses, rather than one time measurements, in research and future clinical practice.
Keywords: Microbiome, Temporal variability, Leukemia, Chemotherapy, Antibiotics
* Correspondence: sshelburne@mdanderson.org
1
Department of Infectious Disease, Infection Control and Employee Health,
MD Anderson Cancer Center, Houston, TX 77030, USA
4
Department of Genomic Medicine, MD Anderson Cancer Center, Houston,
TX 77030, 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.
Galloway-Peña et al. Genome Medicine (2017) 9:21
DOI 10.1186/s13073-017-0409-1
Background
There is an increasing appreciation for the role the hu-
man microbiome plays in many aspects of human physi-
ology, health, and disease. Several studies of healthy
human cohorts have found that although each person
has a relatively distinct gastrointestinal microbiome sig-
nature, a healthy individual’s microbiome remains rela-
tively stable over time [1–4]. Although several factors,
such as diet, drive normal levels of day-to-day micro-
biota variability, it appears that a steady-state equilib-
rium both ecologically and functionally is required for
health. In contrast, acute perturbations of an individual’s
microbiome stability within a temporal context can lead
to an unhealthy status [5, 6]. Considering that one of the
principal aims of the microbiome research community is
to use the microbiome as either an indicator for morbid-
ity or to improve human health, an enhanced under-
standing of the kinetics and taxonomic characterization
of microbiome stability in acutely ill patients is of para-
mount importance [7–11].
Although several studies have been done in healthy
subjects, relatively scant data are available as to the stabil-
ity, resilience, and temporal dynamics of the gastrointes-
tinal microbiome in acutely ill patients [1, 3, 4, 12–16].
Many of the previous investigations examining temporal
variability of the microbiome using healthy partici-
pants have been limited by small numbers of volun-
teers [4, 12, 13], short periods of longitudinal sampling
[2, 3, 16], or by being focused on only one site of collec-
tion [1, 4, 14, 17]. On the other hand, the limited number
of temporal variability studies among ill patients have
typically been in cohorts with chronic ailments such as
atopic dermatitis or colitis [18–20]. A study of stool sam-
ples from 14 patients under intensive care described
rapid shifts in microbiome composition to ultra-low di-
versity communities comprised of four or less taxa as a
result of aggressive antibiotic treatment and other inten-
sive care medication stresses, such as opioids [11]. Simi-
lar dramatic changes in the microbiome were also
observed in patients undergoing hematopoietic stem cell
transplant, where increased microbial chaos early after
transplant is thought to be a potential risk factor for sub-
sequent graft versus host disease [21, 22]. However,
quantitative measurements of longitudinal microbial
variability among ill patients and an analysis of factors
affecting microbiome temporal stability are lacking
[21, 23–25]. Moreover, despite many report s associat-
ing low microbial diversity with different illnesses ,
most studies associate only one-time microbiome mea-
surements with subsequent clinical outcomes, which
could be potentially problematic in settings of significant
temporal variability [24, 26].
Our group previously reported that a single measure-
ment of baseline stool microbial diversity was associated
with infectious risk for 34 patients during induction
chemotherapy (IC) for acute myelogenous leukemia
(AML) [25]. Simila r to other studies of ill patients, we
observed instances of rapid and profound shifts in the
microbiota in our AML cohort [11, 21, 22]. Thus,
herein, we sought to quantify the overall intra-patient
temporal variability of the oral and stool microbiome of
this cohort expanded to 59 patients. In addition, we
sought to determine the consequences of microbiome
temporal instability on patient outcomes and clinical fac-
tors driving intra-patient temporal variability of the
microbiome during IC. We chose to study such patients
because of the opportunity to characterize the micro-
biome prior to receipt of chemotherapy and intense anti-
biotic exposure (i.e., prior to severe perturbations) and
the capacity to obtain dense longitudinal sampling over
the course of intensive treatment due to the extended
inpatient nature of IC. Moreover, AML patients are at
high risk for infection during IC and such i nfe ctions
are generally derived from the commensal microflora
[21, 23]. We hypothesized that higher microbiome
intra-patient t emporal variability, driven by prolonged
antibiotic exposure, w ould be associated with poorer
clinical outcomes.
Methods
Patient recruitment and specimen collection
Study subjects included 59 newly diagnosed adult AML
patients undergoing IC at MD Anderson Cancer Center
(MDACC) in Houston, TX from September 2013 to
October 2014. AML patients initiating inpatient IC at
MDACC were approached for study inclusion unless
they had systemic infection. AML patients receiving IC
at MDACC are routinely prescribed a prophylactic
fluoroquinolone or cephalosporin prior to the initiation
of therapy. In this study, 100% of patients received rou-
tine prophylaxis, with 64% of baseline stool, and 55% of
baseline oral samples taken after the patient had alre ady
started prophylactics. AML patients over 50 years receiv-
ing IC are treated in a laminar-air flow isolation until
neutrophil counts recover to >500 cells/μL or until day
28. Patients aged under 50 years are admitted for the
duration of the chemotherapy (approximately 4–5 days)
and then followed as an outpatient with clinic visits
three times a week until neutrophil recovery or 28 days.
Buccal and fecal specimens were collected from each
patient at baseline, continued approximately every 96 h
as available, and stopped upon neutrophil recovery.
Baseline samples were considered up to 8 days before
and 24 h following IC initiation. As per availability of
samples, 55 (93%) of the patients had oral samples col-
lected before or at the same time as the initiation of
chemotherapy, while 35 (59%) of the patients had stool
samples collected before or at the same time as the
Galloway-Peña et al. Genome Medicine (2017) 9:21 Page 2 of 14
initiation of chemotherapy. The buccal mucosa of each
individual was swabbed three times on each side using a
Catch-All™ Sample Collection Swab (Epicentre). Patient
stool samples were either collected in a stool hat or
using a BBL™ CultureSwab® (BD Diagnostics). All sam-
ples were placed in sterile 2-mL cryovials and stored im-
mediately at −80 °C until further processing.
16S rRNA sequencing and data processing
Bacterial genomic DNA was extracted from buccal and
stool specimens using the MO BIO PowerSoil DNA
Isolation Kit (MO BIO Laboratories). The 16S rRNA V4
region was PCR amplified and sequenced on the Illumina
MiSeq platform using a 2 × 250-bp paired-end protocol
adapted from the Human Microbiome Project (HMP)
methods [16, 27]. All samples from the same patient and
site were processed and sequenced together to minimize
batching issues. Amplification primers contained adapters
for MiSeq sequencing and single-index barcodes resulting
in PCR products that were pooled and sequenced directly.
Read pairs were de-multiplexed based on barcodes
and merged using USEARCH v7.0.100. 16S rRNA
gene sequences were allocated to spe cific operational
taxonomic units using a UPARSE pipeline and aligned
to the V4 region within the SILVA SSURef_NR99_119
database [28]. Analysis of microbiome communities
was performed in R (R Core Team 2015, version 3.2.2,
http://www.R-project.org), using phyloseq [29] to calculate
α-andβ-diversity metrics. The Shannon Diversity Index
(SDI) was used for α-diversity calculations, and weighted
and unweighted UniFrac for β-diversity distances [30].
The 16S V3–V4 region HMP sequencing reads were ob-
tained from http://hmpdacc.org/HMQCP, trimmed to
match the region amplified by this study, and processed
identically to AML patient samples.
Microbiome community and statistical analyses
Intra-patient temporal variability of microbial diversity
was defined a s the coefficient of va riation (CV) of a
longitudinal collection of α-diversity values , and was
calculated for eac h patient’s set of oral and stool sam-
ples. Higher values were indicative of more variable
microbial diversity. Temporal variability in community
compo sition, or β-diversity, of each patient was deter-
mined for the oral a nd stool by calculating the C V of
the weighted and unweighted UniFrac distances of
longitudinal samples collected from each individual
per site. Again, higher values were indicative of more
variable communities. Pairwise differences in temporal
variability across body sites were made using Mann–
Whitney U test, whereas pairwise differences among
infection or response groups wa s performed using
Student’s t-test with Welch’s correction. Linear correla-
tions between C Vs at different body sites were determined
using Pearson’s r and P values generated in GraphPad
Prism 6.
Heatmaps ana lyzing genera abundance over time
among patients with increasing temporal variability
were generated with the publically available pheatmap
R package version 1.0.8. (http://CRAN.R-project.org/
package=pheatmap), and include correlation metrics
calculated with R’s cor and cor.test stats package func-
tions. P values were corrected for multiple comparisons
using the Benjamini and Hochberg method.
For each body habitat the population was divided into
quartiles based on CV of the weighted UniFrac distance
values or SDI where the first quartile was defined as
stable, second and third as average, and fourth as vari-
able as previously described [15]. To determine signifi-
cant differences in genera abundance between stable,
average, and variable individuals, we tested for differ-
ences between groups using non-parametric Kruskal–
Wallis analysis of variance in R for genera across indi-
viduals, then corrected for the false discovery rate using
the Benjamini and Hochberg method.
Multivariable regression analyses were performed
using base R (R Core Team 2015, version 3.2.2, http://
www.R-project.org ) and included age, antibiotic type,
chemotherapy regimen, and exposure to antibiotics as
covariates. Antibiotic types were subdivided into three
major broad spectrum β-lactam antibiotics received by
this cohort, namely, cefepime, carbapenems (primarily
meropenem), and piperacillin-tazobactam. Chemotherapy
regimens were subdivided into fludarabine-containing reg-
imens, high intensity non-fludarabine-containing regi-
mens, hypomethylators, or other. Fludarabine-containing
regimens included fludarabine in combination with
idarubicin and cytarabine [31], or fludarabine/idarubicin/
cytarabine with G-CSF (FLAG-Ida). High intensity non-
fludarabine-containing regimens were purine analog of
clofarabine or cladrabine in combination with idarubicin
and cytarabine. Hypomethylator-based combinations in-
cluded decitabine and azacytidine [32].
Clinical definitions
Infections were defined as microbiologically defined in-
fections (MDIs) or clinically defined infections as de-
scribed pre viously [25]. Subsequent infectious episo des
were defined as MDIs that occurred within 90 days of
cessation of longitudinal samp ling. Complete remission
(CR) of AML was assessed using standard definitions [33].
Results
AML patients undergoing IC exhibit temporal instability
of the stool and oral microbiome diversity
In order to understand the intra-patient temporal vari-
ability of the microbiome among hospitalized patients
with AML, we performed sequencing of the V4 region
Galloway-Peña et al. Genome Medicine (2017) 9:21 Page 3 of 14
of the 16S rRNA gene via the MiSeq platform (Illumina)
using the 2 × 250-bp protocol [34] on a total of 901 lon-
gitudinal samples collected twice weekly from initiation
of chemotherapy until neutrophil recovery for 59 AML
patients undergoing IC. Of the samples, 848 (84%, n =365
stool and 483 oral) passed sequencing quality control
measures for further analyses. For these samples, we ob-
tained a total of 24,271,698 reads, for an average of 28,622
reads per sample. Patient demographics and clinical meta-
data can be found in Table 1.
Currently, the majority of 16 s rRNA microbiome-based
data are summarized using either numerical or index-
based measurements of species richness and/or evenness
within a habitat (i.e., α-diversity) or characterization of
differences in microbial community composition by meas-
uring the distance or dissimilarity between samples (i.e.,
β-diversity). We first sought to determine intra-patient
temporal variability of α-diversity by calculating the CV of
the SDI for both the oral and the stool samples for each
patient. The coefficient of variation is defined as the ratio
of the standard deviation to the mean; thus, a low CV
would mean an individual had relatively stable species di-
versity over time whereas a high CV would reflect more
variation. We found considerable heterogeneity in the
temporal stability values of both stool (mean SDI CV
0.48 ± 0.25) and oral (mean SDI CV 0.42 ± 0.26) samples
among AML patients during IC (Fig. 1a). There was no
statistically significant difference in CV values between the
two sites (P=0.16). This finding is in contrast to previous
studies performed in healthy individuals where the micro-
biota of oral samples have been shown to be less variable
compared to stool [2, 15]. The intra-patient temporal vari-
ability of other α-diversity metrics, specifically the Chao-1
diversity index and Simpson’s diversity index, were also
analyzed for the oral (mean Chao CV 0.39 ± 0.18, mean
Simpson CV 0.33 ± 0.24) and stool samples (mean Chao
CV 0.48 ± 0.22, mean Simpson CV 0.37 ± 0.27) of the
AML cohort (Additional file 1: Figure S1a, b). Assessment
of the temporal variability of α-diversity also revealed that
the SDI CV of oral and stool samples from the same
patients were statistically moderately correlated (P=0.01,
r = 0.33; Fig. 1b). The relationship between the two sites
leads to the postulation that factors influencing temporal
variability of microbial diversity in treated cancer patients
may be acting on both sites concurrently. Conversely, it
has been reported that the variability of one body site was
not associated with the variability of other body habitats
in healthy cohorts [15].
High intra-patient temporal variability of oral and stool
microbiome among AML patients is associated with
increased pathogenic-associated genera abundance
Next we sought to determin e the temporal variability in
microbiome community structure and membership as
represented by quantitative and qualitative measure-
ments of β-diversity using weighted and unweighted
UniFrac distance measurements, respe ctively. Here, we
Table 1 Clinical features of 59 AML patients
Characteristic Number (%)
Demographics
Median age in years
a
55 (49–68)
Male 31 (52.5)
Female 28 (47.5)
Median days on study 28 (25–35)
Median number of oral samples 8 (6–9)
Median number of stool samples 6 (4–8)
Chemotherapy
Hypomethylators
b
14 (23.7)
Non-fludarabine high intensity
c
19 (32.2)
Fludarabine-containing
d
19 (32.2)
Other
e
7 (11.8)
Chemotherapeutic response
Complete remission after IC 20 (33.8)
Overall response rate
f
43 (72.8)
Infections
g
Microbiologically documented infection 15 (25.4)
Clinically documented infection 14 (23.7)
No infection 30 (50.8)
Antimicrobial administration
Received treatment antibiotics
h
53 (89.8)
Carbapenem >72 h 39 (66.1)
Piperacillin/tazobactam >72 h 14 (23.7)
Cefepime >72 h 26 (44.1)
Received prophylactic antibiotics 59 (100)
Median number of antibiotics administered 6 (4–7)
Median number of days exposed to all antibiotics
i
28 (24–35)
Median number of days exposed to treatment antibiotics 16 (9–24)
Median number of days exposed to prophylactic antibiotics 16 (8–28)
a
All median values in this table have the interquartile range in parentheses
b
These chemotherapies included: 1) vasoroxin in combination with
decitabine; 2) decitabine alone; 3) azacytidine in combination with pracinostat;
4) azacytidine in combination with quidartinib; and 5) SGI-110
c
These chemotherapies included: 1) CIA, 2) CLIA, 3) or CIA + sorafanib
d
These chemotherapies included: 1) FLAG-Ida or 2) FIA regimens
e
Other chemotherapies included:1) omacetaxine in combination with low-dose
cytarabine or 2) Clad + LDAC
f
Includes CR (morp hologic complete remission), CRi (morphologic complete
remission with incomplete bloodcount recovery), and CRp (morphologic
complete remission with incomplete platelet recovery)
g
Specific information on microbiologically and clinically documented
infections can be found in the “Methods”
h
Refers to any antibiotic/antimicrobial-based therapy given for suspected or
proven infection, that is, not included as prophylaxis (cephalosporins or
fluoroquinilones). Denoted are the three most common broad spectrum
antibiotics given in the study. Note that numbers of individual antibiotics add
up to >100% because some patients received more than one of the listed
antimicrobials during IC
i
Includes prophylactic antibiotics
Galloway-Peña et al. Genome Medicine (2017) 9:21 Page 4 of 14
considered the CV of each patient’s samples per site in
order to characterize the dispersion of β-d iversity
metrics. The mean CVs of weighted and unweighted
UniFrac distances for the cohort were 0.24 ± 0.1 and
0.16 ± 0.04 for the o ral samples and 0.32 ± 0.2 and
0.20 ± 0.08 for stool samples, respectively (Fig. 1c).
Contrary to SDI CVs , the temporal variability of the
weighted UniFrac distances between the oral and stool of
patients was not significantly correlated (P=0.10;
Additional file 1: Figure S1d). Reports in healthy persons
have observed associations between diversity and tem-
poral stability, such as individuals with a more diverse
microbiome are likely to have a more stable microbiome
over time [4, 14, 15]. However, we did not find any statis-
tically significant correlations between either the baseline
or median SDI values of patients and their temporal vari-
ability as measured by the CV of the weighted UniFrac
distances of their samples, suggesting microbiome struc-
tural variability does not appear to be affe cted by α-
diversity in treated AML patient s (Additional file 1:
Figure S2).
It is well known that patients in the hospital are at risk
for colonization and intestinal domination by pathogenic
bacteria and that a diverse microbiome provides
colonization resistance against many such organisms
[7, 11, 21]. Thus, to beg in to investigate factors that
might influence temporal instability in our cohort, we
sought to test the hyp othesis that temporal instability
was influenced by increasing relative abundance of
pathogenic-associated genera, such as Enterococcus and
Staphylococcus. In order to visualize this relationship, we
ranked patients and their samples by CV of weighted
UniFrac from smallest to greatest (low variability to high
variability of microbial community structure), and corre-
lated this with the relative abundance of specific genera
(Fig. 2). Consistent with our hypothesis, high weighted
UniFrac CV values were moderately positively correlated
with the relative abundance of pathogenic genera such as
Staphylococcus (P <0.001, r =0.3), Streptococcus (P =0.02,
r = 0.2), and Stenotrophomonas (P =0.01, r =0.2) in the
A
B
C
A ML Ora l
AML Sto ol
0.0
0.5
1.0
1.5
Coefficient of Variation of
the Shannon Diversity Index
P=0.16
Weighted AMLOral
Unweighted AML Oral
W e ighted A ML Stool
U nweighted A ML S tool
0.0
0.5
1.0
1.5
Coefficient of Variation
of the UniFrac Distance
P=0.001
P=0.24
0.0 0.5 1.0
0.0
0.5
1.0
Coefficient of Variation of
the Shannon Diversity
Index for Stool Samples
Coefficient of Variation of
the ShannonDiversity
Index for Oral Samples
P=0.01
r = 0.33
Fig. 1 Intra-patient temporal variability in oral and stool microbiomes
of hospitalized AML patients undergoing IC. a The oral and stool
microbial α -diversity intra-patient temporal variability. Each point
represents the coefficient of variation (CV) of the Shannon diversity
index (SDI) for samples derived from each patient. b The correlation
between the CV of the SDI values originating from oral and stool
samples from the same patient. The Pearson’s correlation (r) value
and P value from correlation analyses also are indicated. c The oral
and stool microbial β-diversity intra-patient temporal variability using
either the CV of the weighted or unweighted UniFrac distances for
samples derived from each patient. In panels a and c, the bars
represent mean ± standard deviation, and P values comparing the
different body sites were calculated using a Mann–Whitney U-test
Galloway-Peña et al. Genome Medicine (2017) 9:21 Page 5 of 14