KITP, July 2014
Single-Cell Variability of Growth in Bacteria
as an Optimization Process
Nathalie Q. Balaban
Racah Institute
The Hebrew University
Jerusalem, Israel
Growth arrest is the prevalent state of
microorganisms in nature
Lennon JT & Jone SE, Nat Rev Micr
(2011)
Typical bacterial growth curve
Jacques Monod, Thesis 1942
Microfluidic devices for following
single-cells
Balaban NQ et al., Science 2004
Growth arrest in microfluidic
devices
Measuring the activity of single growth arrested
cells
Do stationary phase bacteria respond to induction?
Fluorescence induction of growth arrested cells in microfluidic devices
t:
1:25 h
0
2:30 h
4:15 h
5:40 h
7:50 h
9:15 h
GFP Fluorescence [a.u.]
70
60
CASP : Constant Activity
Stationary Phase
50
40
30
20
10
0
1
2
3
4
5
Time from induction [Hours]
6
7
OD (a.u)
CASP: Constant Activity Stationary Phase in batch
cultures
Number of bacteria
10
-1
0
10
20
30
40
50
20
30
40
50
20
30
40
50
mCherry
Fluorescence (a.u)
Time (h)
Gefen O. et al., PNAS (2014)
100
0
0
10
0
10
1
PA (a.u)
At CASP, cells produce
fluorescence at a
constant rate for
several days, despite
the absence of growth
200
Time (h)
0.1
Time (h)
CASP
Precise promoter activity
measurements at CASP
GFP Fluorescence (a.u)
100
10
PA (a.u)
8
50
6
4
2
0
0
5
10
15
20
Time from induction (h)
CASP
25
30
0
10
-6
10
-5
10
-4
IPTG concentration (M)
Fitness cost of protein production at CASP
No detectable cost of protein production at CASP
Typical bacterial growth curve
Jacques Monod, Thesis 1942
Growth arrest during the lag phase
ScanLag: High throughput assay for lag variability
quantification
Scanners
Computer
Relay
0
2261
2336
323
6
Irit Levin-Reisman et al. Nature Methods (2010)
Irit Levin-Reisman et al. Jove (in press)
3806
Time
[minutes]
Appearance distribution can reflect growth arrest
distribution
The area of each colony as a function of the time
-3
x 10
Appearences frequency [minute-1]
Area [pixels]
100
80
60
40
20
0
600
700
800
900
1000 1100 1200 1300 1400 1500
of colonies
Time [minutes]
Colony Appearance distribution
6
4
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
500
1000
1500
2000
2500
Time [minutes]
3000
3500
Persistence to antibiotics
0
10
-1
Survival ratio
10
e
t 1t  (1   )e  2t )
N
(
t
)

N
(

N (t )  N00e
-2
10
-3
10
Balaban Curr Opi Gen & Dev 2011
-4
10
0
5
10
15
20
25
30
35
Time (h)
Bigger, The Lancet 1944
40
45
50
55
Direct observation of the
single cell response to
atibiotics
Persistence is not due to a mutation
High persistence is due to a subpopulation of growth arrested
bacteria
Extended lag variability as a bet-hedging strategy
Growth arrest time
Main Questions
• Molecular level: What controls the duration
of the lag?
• Population level: How can growth arrested
and growing cells co-exist?
• Evolutionary level: Is growth arrest a
strategy selected by evolution or the byproduct of a defective cell-cycle?
Main Questions
• Molecular level: What controls the duration
of the lag?
• Cell level: Is growth arrest at lag similar to
stationary phase arrest?
• Population level: How can growth arrested
and growing cells co-exist?
• Evolutionary level: Is lag extension a
strategy selected by evolution or the byproduct of defective cell-cycle?
Log
Survival
Protocol of evolution under cyclic antibiotic
exposure
Time
Ofer Fridman
The evolved populations survive extensive antibiotic
treatment
Survival fraction after exposure
Resistance
Ancestral
Failure of antibiotic treatments is typically
attributed to resistance
Resistant
Sensitive
Antibiotic
Growth inhibition zone
Evolution of tolerance
Survival fraction after exposure
Resistance
Evolved
Ancestral
No difference in sensitivity to the antibiotic
No difference in growth rate
Tolerance is due to an extension of the lag time
Evolution of tolerance
Ta=duration of intermittent antibiotic treatment
Ta=3h
Ta=5h
Survival fraction after exposure
Resistance
Ancestral
No difference in sensitivity to the antibiotic
No difference in growth rate
Ta=8h
Tolerance is due to an extension of the lag time
Fridman O., et al. Nature 2014
The evolved lag time as an optimization process
-3
tlag
Environment
Frequency
Bacterial
state
Appearences frequency [minute-1]
x 10
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
L
L  
t lag
G
L
G  

t grow
t lag
500
1000
1500
2000
2500
Lag
Time
Time [minutes]
3000
3500
Bacterial
state
Optimal lag duration
The evolved lag time is optimized to the duration of
the antibiotic exposure
Antibiotic treatment
duration
The tolerance-by-lag strains have adapted to the duration of the treatment, not to its precise chemical nature
Strains evolved under cell-wall
targeting antibiotic are also tolerant to
DNA gyrase inhibitors
Tolerance-by-lag genes (tbl)
• Toxin-antitoxin module (vapBC-like)
• Methionyl-tRNA synthetase (metG)
• Ribose-phosphate diphosphokinase (prsA)
Toxin-Antitoxin Modules
on Plasmids
Antitoxin (B)
Toxin (A)
“Addiction” module
Hayes, F. Science 301, p.1492 (2003)
Toxin-Antitoxin Chromosomal Modules
(Aizenman E et al. PNAS 1996)
Nature Reviews Microbiology, 2011
Toxin-antitoxin modules control of lag
time duration
•Mutations in hipA lead to high persistence
(Harris Moyed J Bac, 1983, 1988, 1991,1994;
Korch et al Mol Mic 2003)
The hip (high persistence toxin-antitoxin module)
The level of induction of the toxin
determines the growth arrest duration
0.7
0.6
0. 4
0.4
0.4
0.3
0.2
S36
0.1
0
1
S1
0
41
200
Growth Time [min]
0.2
400
0
01
0
10
1-CDF
Fraction of detectable CFU
0.5
10
34
500
67
Appearance Time [min]
1000
0
0 ng/ml
10
12
14
16
-1
0
500
Time [min]
1000
1500
Eitan Rotem
0
10
12
14
10016
1500
HipA toxin induces a long lag by phosphorylating
GltX, the glutamyl tRNA synthetase
Starvation response
HipB
antitoxin
degradation
Uncharged tRNA
free HipA
toxin
Kaspy I., et al, (Nature Communication, 2013)
Collaboration with Gad Glaser
The HipA toxin extends the lag by mimicking starvation
Threshold amplification of noise
A
A=toxin
+
B
K
A B
B=antitoxin
dA
   k  A  B  k  AB  A
dt
dB
   k  A  B  k  AB   * B
dt
dAB
 k  A  B  k  AB  AB
dt
(Lenz DL et al., 2004; Levine E.
et al., 2007; Buchler NE, 2008)
Stochastic simulations of growth arrest
Adiel Loinger
In collaboration with O. Biham
From molecular noise to phenotypic variability of growth at the
single-cell level
00:00
00:45
01:22
02:15
03:45
Rotem E et al, PNAS 2010
The level of induction of the toxin
determines the growth arrest duration
0.7
0.5
0.4
0.4
0.3
0.2
S36
0.1
0
1
persistent
0.2
S1
0
41
200
Growth Time [min]
400
tolerant
0
01
0
10
1-CDF
Fraction of detectable CFU
0.6
sensitive
0. 4
10
34
500
67
Appearance Time [min]
1000
0
0 ng/ml
10
12
14
16
-1
0
500
Time [min]
1000
1500
Eitan Rotem
0
10
12
14
10016
1500
Difference between large numbers leads to amplification of noise
Time (min)
MIC and MDK: a global framework to define
resistance, tolerance and persistence
MIC and MDK
Summary
• CASP: Constant Activity Stationary
Phase enables quantitative
measurements on non growing cells
• Toxin-antitoxin modules are ideally
suited to control tolerance and
persistence by mimicking starvation
• Growth arrest can evolve to match
the time scale of the stress
– Understand tbl mutations
– Search for similar effects in patients
treated with cyclic exposure to
antibiotics
100
50
0
0
5
10
15
20
25
30
Acknowledgments
Hebrew University
Broad Institute
Ofer Biham
Adiel Loinger
Noam Shoresh
Gad Glaser
Ilana Kaspy
Balaban lab
Orit Gefen
Eitan Rotem
Sivan Pearl
Ira Ronin
Ofer Fridman
Irit LevinReisman
Eliq Oster
Noga Weiss
Amir Goldberg
Funding