Dr Jethro Herberg @ MRF's Meningitis & Septicaemia in Children and Adults 2017
1. Distinguishing bacterial and viral
infections by host gene expression
MRF 2017 Conference
Jethro Herberg
Clinical Senior Lecturer, paediatric infectious disease, Imperial College London
2. The
problem
to
be
addressed
Febrile
children:
• Unreliable
clinical
diagnosis
of
bacterial,
viral,
inflammatory
disease
• Lab
tests
slow,
inaccurate
• Many
admitted
for
antibiotics
• Most
leave
without
a
diagnosis
• Most
didn’t
need
antibiotics
or
admission
We
need
better
tests
that
can
discriminate
febrile
children
with
bacterial
infection,
viral
infection
or
inflammation
3. Fatal
SBI
-‐ missed
opportunity
for
early
antibiotic
treatment
Minority
with
severe
illness
requiring
paediatric
intensive
care
Unknown
number
of
children
with
SBI
missed
by
current
clinical
diagnostic
tools
25%
ED
attendances
for
fever
3%
caused
by
SBI
– 20%
receive
antibiotics
Blood
tests,
on
discretion
of
clinician.
20%
have
SBI
-‐ 75%
receive
antibiotics
4-‐6
episodes/year.
90%
have
medical
consultation.
0.1%
have
SBI
– 10%
get
antibiotics
SBI
fatal
Need
PICU
Confirmed
SBI
Febrile
children
presenting
to
hospital
Febrile
children
having
blood
tests
Febrile
children
in
the
community/
primary
care
Scale of
the
problem
St Mary’s Hospital:
26,000 children/year
9,000 with fever
2,300 admitted
1,500 on antibiotics
How many really needed
admission on antibiotics??
4. We
could
give
antibiotics
to
everyone,
just
to
be
safe…
…in
fact
we
do…
(almost)
6. Is
better
pathogen
detection
the
answer?
• Viral
diagnostics
– Molecular
diagnostics
very
sensitive
(80%
of
patients
positive)
– Problems
of
clinical
interpretation
• Culture-‐based
bacterial
diagnostics
– Low
sensitivity
– Specificity
problems
with
commensal
organisms,
especially
for
non-‐sterile
site
samples
• Molecular
bacterial
diagnostics
– PCR
– not
greatly
increased
sensitivity
compared
to
blood
culture
– Still
issues
to
resolve
with
sensitivity
and
specificity
– Antimicrobial
resistance
detection
lags
behind
pathogen
detection
7. Differential stimulation
of host inflammation
How about using host biomarkers?
Pathogens
(bacteria,
viruses)
Multiple levels of
omics data
Viral
Response
Host
Infection
Bacterial
Response
8. Does
CRP
discriminate
bacterial
vs
viral
infection?
CRP values in bacterial, viral or unknown infectionC
onfirm
ed
bacterial
Probable
bacterialC
onfirm
ed
viral
D
on'tknow
1
10
100
1000
Infection group
logCRP
• CRP
separates
confirmed
bacterial
and
viral
well
• CRP
not
much
help
for
“don’t
know”
group
– where
you
need
it
most!
Philipp E Geyer et al. Mol SystBiol2017;;13:942
• Massive
(11
log)
range
in
serum
protein
concentrations
• Wide
range
of
chemistries
– can
only
interrogate
a
tiny
proportion
at
one
time
• Most
discovery
based
on
candidate
molecules
• New
tests
for
bacterial
infection,
eg MeMed – not
yet
for
inflammation
Why
don’t
we
have
better
biomarkers?
Only 5 logs
9. Gliddon, Herberg, Levin, Kaforou Immunology 2017
Transcriptomics:
• Measure
the
expression
level
of
all
genes
in
cells
from
blood
• For
each
gene,
compare
levels
in
patients
to
controls
or
other
groups
• Look
for
gene
expression
biomarkers
of
infection
or
inflammation
10. Immunopathology of Respiratory, Inflammatory and Infectious Disease Study
Child with suspected infection
Apply best available diagnostics
Simple rules to classify patients with bacterial
infection, and those without
Run whole blood samples on Illumina
microarrays
11. Patients
with
a
similar
clinical
presentation
have
distinct
gene
expression
patterns
Comparison
of gene
expression
in RSV and
influenza
-------Flu patients------- -------RSV patients------
Each infection has a unique gene expression signature
Red squares: increased expression
Blue squares: decreased expression
12. Assign a predictive value to each
individual by combining up and
down regulated genes
Simplify the transcript signature into a single
value – Disease Risk Score
Pooled up-
regulated probes
Pooled down-
regulated probes+
Disease Risk
Score
or
=
13. H1N1 RSV
Disease Risk
Score separates
H1N1 and RSV
Pooled up-
regulated probes
Pooled down-
regulated probes+
Disease Risk
Score
or
=
14. Apply this approach to a useful question:
Do patients with bacterial infection have a
common gene expression signature?
(…and can we use this approach to diagnose other
tricky conditions?)
15. Recruitment of febrile patients
Categorisation of patients based on clinical data
INFLAMMATORY
SYNDROMES
No accurate test
VIRAL syndrome
Don’t need
antibiotics
BACTERIAL
syndrome
Need antibiotics
Probable
Bacterial
Unknown
Bacterial
or Viral
Definite
Bacterial
Probable
Viral
Definite
Viral
Inflamm.
diagnosis
Review clinical investigation results
Whole blood transcriptomic profile
Study design
16. Recruitment of febrile patients
Categorisation of patients based on clinical data
INFLAMMATORY
SYNDROMES
No accurate test
VIRAL syndrome
Don’t need
antibiotics
BACTERIAL
syndrome
Need antibiotics
Probable
Bacterial
Unknown
Bacterial
or Viral
Definite
Bacterial
Probable
Viral
Definite
Viral
Inflamm.
diagnosis
Review clinical investigation results
Whole blood transcriptomic profile
Study design
2010
2013
2014
2015
1096 febrile
children
672 samples
viral/bact
438 arrays
Discover
signatures
• 292-case
discovery set
• 130-case
validation set
• Multiple
published
datasets for
external
validation
17. 8565
SDE
transcripts
Compare ‘Definite Bacterial’ patients to ‘Definite Viral’
38
SDE
transcripts
elastic
net
Many transcripts are
highly correlated
Variable selection
algorithm identifies
optimal minimum
discriminatory signature
Log2 fold change
-log10(corrected P value) 80% training set
18. Heat map: clustering based on the bacterial vs. viral 38-transcript signature.
• red - patients with Bacterial
infection
• blue - patients with Viral
infection
• red squares – upregulated genes
• green squares – downregulated genes
19. Using a novel variable selection algorithm: forward
selection – partial least squares (FS-PLS)
Can we shrink signature further?
2 transcripts can identify bacterial vs viral cases
Pool up-regulated
probes
Pool down-
regulated probes
+
Disease Risk
Score
or
=
20. Bacterial vs
JIA, HSP
Bacterial vs
SLE
Bacterial vs
viral
Bacterial vs
viral
Published cohortsTest set (20%)
Berry et alRamiloet alHu et alWright et al
AUC 0.963
(0.874-1.0)
Validation set
AUC 0.974
(0.912-1.0)
Diagnostic Test Accuracy of a 2-Transcript Host RNA Signature
for Discriminating Bacterial vs Viral Infection in Febrile Children
Jethro A. Herberg, PhD; Myrsini Kaforou, PhD; Victoria J. Wright, PhD; Hannah Shailes, BSc; Hariklia Eleftherohorinou, PhD; Clive J. Hoggart, PhD;
Miriam Cebey-López, MSc; Michael J. Carter, MRCPCH; Victoria A. Janes, MD; Stuart Gormley, MRes; Chisato Shimizu, MD; Adriana H. Tremoulet, MD;
Anouk M. Barendregt, BSc; Antonio Salas, PhD; John Kanegaye, MD; Andrew J. Pollard, PhD; Saul N. Faust, PhD; Sanjay Patel, FRCPCH;
Taco Kuijpers, PhD; Federico Martinón-Torres, PhD; Jane C. Burns, MD; Lachlan J. M. Coin, PhD; Michael Levin, FRCPCH; for the IRIS Consortium
Research
JAMA | Preliminary Communication | INNOVATIONS IN HEALTH CARE DELIVERY
21. Give
antibiotics
Uncertain
role for
antibiotics
No need for
antibiotics
Could 2-transcript test be basis of “traffic light” test for antibiotic treatment?
22. Percentage of patients
Definite
Viral
Probable
Viral
Unknown
Bacterial or Viral
Probable
Bacterial
Definite
Bacterial
Comparison of Disease Risk Score values and
antibiotic treatment
Grey bars - proportion of patients receiving antibiotics
Coloured bars - proportion predicted to be bacterial by DRS
23. How
can
we
validate
findings?
• Pilot
data
suggest:
need
for
antibiotics
may
be
identified
by
test
based
on
very
few
transcripts
• BUT
current
data:
– Case
control
design
favours obvious
cases
– Samples
collected
late,
not
at
presentation
– Incomplete
range
of
febrile
phenotypes
– Not
enough
cases
to
compare
severe/mild
phenotypes
What
do
we
do
next?
24. Association of RNA Biosignatures With Bacterial Infections
in Febrile Infants Aged 60 Days or Younger
Prashant Mahajan, MD, MPH, MBA; Nathan Kuppermann, MD, MPH; Asuncion Mejias, MD, PhD; Nicolas Suarez, PhD;
Damien Chaussabel, PhD; T. Charles Casper, PhD; Bennett Smith, BS; Elizabeth R. Alpern, MD, MSCE; Jennifer Anders, MD; Shireen M.
Atabaki, MD, MPH; Jonathan E. Bennett, MD; Stephen Blumberg, MD; Bema Bonsu, MD; Dominic Borgialli, DO, MPH; Anne Brayer, MD;
Lorin Browne, DO; Daniel M. Cohen, MD; Ellen F. Crain, MD, PhD; Andrea T. Cruz, MD, MPH; Peter S. Dayan, MD, MSc; Rajender Gattu, MD;
Richard Greenberg, MD; John D. Hoyle Jr, MD; David M. Jaffe, MD; Deborah A. Levine, MD; Kathleen Lillis, MD; James G. Linakis, MD, PhD;
Jared Muenzer, MD; Lise E. Nigrovic, MD, MPH; Elizabeth C. Powell, MD, MPH; Alexander J. Rogers, MD; Genie Roosevelt, MD;
Richard M. Ruddy, MD; Mary Saunders, MD; Michael G. Tunik, MD; Leah Tzimenatos, MD; Melissa Vitale, MD; J. Michael Dean, MD, MBA;
Octavio Ramilo, MD; for the Pediatric Emergency Care Applied Research Network (PECARN)
IMPORTANCE Young febrile infants are at substantial risk of serious bacterial infections;
however, the current culture-based diagnosis has limitations. Analysis of host expression
patterns (“RNA biosignatures”) in response to infections may provide an alternative
diagnostic approach.
OBJECTIVE To assess whether RNA biosignatures can distinguish febrile infants aged 60 days
or younger with and without serious bacterial infections.
DESIGN, SETTING, AND PARTICIPANTS Prospective observational study involving a
convenience sample of febrile infants 60 days or younger evaluated for fever (temperature
>38° C) in 22 emergency departments from December 2008 to December 2010 who
underwent laboratory evaluations including blood cultures. A random sample of infants with
and without bacterial infections was selected for RNA biosignature analysis. Afebrile healthy
infants served as controls. Blood samples were collected for cultures and RNA biosignatures.
Bioinformatics tools were applied to define RNA biosignatures to classify febrile infants by
infection type.
EXPOSURE RNA biosignatures compared with cultures for discriminating febrile infants with
and without bacterial infections and infants with bacteremia from those without bacterial
infections.
MAIN OUTCOMES AND MEASURES Bacterial infection confirmed by culture. Performance of
RNA biosignatures was compared with routine laboratory screening tests and Yale
Observation Scale (YOS) scores.
RESULTS Of 1883 febrile infants (median age, 37 days; 55.7% boys), RNA biosignatures were
measured in 279 randomly selected infants (89 with bacterial infections—including 32 with
bacteremia and 190 without bacterial infections), and 19 afebrile healthy infants. Sixty-six
classifier genes were identified that distinguished infants with and without bacterial
infections in the test set with 87% (95% CI, 73%-95%) sensitivity and 89% (95% CI,
81%-93%) specificity. Ten classifier genes distinguished infants with bacteremia from those
without bacterial infections in the test set with 94% (95% CI, 70%-100%) sensitivity and
95% (95% CI, 88%-98%) specificity. The incremental C statistic for the RNA biosignatures
over the YOS score was 0.37 (95% CI, 0.30-0.43).
CONCLUSIONS AND RELEVANCE In this preliminary study, RNA biosignatures were defined to
distinguish febrile infants aged 60 days or younger with and without bacterial infections.
Editorial page 824
Related article page 835
Supplemental content
Author Affiliations: Author
affiliations are listed at the end of this
article.
Group Information: Members of the
Pediatric Emergency Care Applied
Research Network (PECARN) are
Research
JAMA | Preliminary Communication | INNOVATIONS IN HEALTH CARE DELIVERY
• Mahajan
et
al:
ED-‐based
study,
22
centres,
diverse
population
• Focusing
on
children
<
60
days
• Active
selection
of
most
difficult
cases
(eg exclude
sepsis,
focal
infection)
• Identification
of
66-‐gene
and
a
10-‐gene
signatures
using
microarrays
How does the 2-gene
signature perform on
these data?
25. Diagnosis of Bacterial Infection Using a 2-Transcript Host RNA Signature in Febrile Infants 60 Days or Younger
Kaforou, Herberg et al JAMA. 2017;317(15):1577-1578. doi:10.1001/jama.2017.1365
Disease Risk Score and ROC Curves Based on the 2-Transcript Signature
(combined IFI44L and FAM89A Expression Values)
26. ?
?
Can we make a simple diagnostic test
using gene expression signatures?
27. Protein
dipsticks
Rapid
sequencing
Quantum dot
nanoparticles
Nanostring
Gold
nanoparticles
Host responsePathogen
Biofire
PCR
CRISPR-based
diagnostic
Culture Modular PCR
….not
forgetting
clinical
features
An
accurate
quick
cheap
test
28. • Horizon
2020
PERFORM
grant:
– 5,000
patient
proteomic
&
transcriptomic
validation
study
– ED,
PICU,
ward
– Full
range
of
febrile
illness,
recruited
at
point
of
presentation
• Platform
development
for
a
diagnostic
test
PERFORMPersonalised managementof
febrile illness
Horizon 2020
29. Acknowledgements
Mike Levin group,
Imperial College (UK)
Myrsini Kaforou, Victoria Wright,
Hannah Shailes, Hariklia
Eleftherohorinou, Clive Hoggart,
Sobia Mustafa, Stuart Gormley
And the clinical teams
Micropathology Ltd (UK)
Colin Fink, Elli Pinnock, Ed
Sumner
Southampton (UK)
Saul Faust, Jenni McCorkill,
Sanjay Patel
Oxford (UK)
Andrew J Pollard
Universitario de Santiago
(Spain)
Miriam Cebey Lopez, Antonio
Salas, Federico Martinón Torres
and GENDRES consortium
University of California San
Diego
John T. Kanegaye, Chisato
Shimizu, Adriana Tremoulet, Jane
Burns
Academic Medical Centre,
Amsterdam
Anouk M Barendregt, Taco
Kuijpers
University of Queensland:
Lachlan JM Coin,
Funding at Imperial: