Tip:
Highlight text to annotate it
X
In the last talk, we discussed some of the problems
confronting development of new tools to prevent and to treat tuberculosis.
In this next segment, I really want to focus on the problem of antibiotic therapy
and delve into this issue of why the antibiotics we currently use to treat tuberculosis
are not more effective than they are.
I'll remind you that we discussed in the last talk
the "Mitchison hypothesis" that's been with us for some time, proposing
that TB chemotherapy is problematic because the bacteria
inhabit different tissue compartments
and therefore adopt different physiologies that make them
refractory to one or more anti-TB drugs.
This idea, which has been around for quite a while now
might explain why the kinetics of killing with the antibiotics that we use
against tuberculosis are, in all cases, biphasic or multi-phasic.
In this case, this is an experiment in which chronically infected mice
were treated with one of our front line drugs, isoniazid
via mono-therapy. What you can see,
looking on the y-axis at the number of viable bacteria
still within the lungs (that's a log scale)
versus the days on chemotherapy
is that killing of bacteria by isoniazid (or INH as I'll call it)
is not a log-linear process.
So, initially, there's a period of rapid cell killing --
rapid being a relative term here.
It's actually very slow compared to the action of most antibiotics against most bacteria.
But, the remarkable thing is that after about 4 weeks of chemotherapy,
the killing essentially stops and the bacterial population stabilizes at a level about
0.1 to 1% of the initial population.
So there's a very high frequency of survivors in the face of chemotherapy.
We call these the "persisters."
We call them persisters to distinguish them from
the resisters, which are bacteria that have evolved through mutation
resistance to the antibiotics
Because, in fact, this is a phenotypic, metastable state
that is not inherited by the progeny of these cells
So, if you were to take these bacteria out of the tissues,
re-grow them in culture, re-infect animals,
and treat them again, you would get exactly the same curve back.
They are not stably, heritably resistant mutants.
This is a purely phenotypic phenomenon that we know very little about.
But, clearly, if we could convert this biphasic curve to a simple
log-linear kill curve, that would be a major advance
towards accelerating the clearance of infection
with chemotherapy.
Now, we can postulate that the reason for these biphasic kinetics is due
to the existence, as Mitchison proposed, of a dormant
subpopulation of bacteria that is killed with slow kinetics
compared to the bulk bacterial population.
If we perform a simple mathematical model in silico for this process
on the basis of differential killing dynamics
of a "fast" population (that is a population that's killed with rapid kinetics
that represents the bulk of the bacteria -- maybe 99.9% of the population)
versus a "slow" population (the "persisters" that's killed with very slow kinetics),
and then calculate the composite of these two curves, as shown by the blue line here,
what we get is an almost perfect match to the experimental data.
Now, of course, the fact that a model matches the experimental data
does not prove that it is true, but it at least
says that the idea is plausible.
So, according to the Mitchison hypothesis then,
the persistence of TB in the face of chemotherapy
is due to a set of environmental inputs leading to a phenotypic switch
(in this case a state of dormancy in a subset of organisms),
which causes them to become these drug-tolerant persisters.
So, there's a simple flow of information from the environment
to the physiology of the bacteria,
and this is what's responsible for the recalcitrance
of these bacteria to killing with antibiotics.
It's a very attractive idea, and as I indicated in the last talk,
there's actually some clinical evidence in favor of this idea,
but there are some awkward bits about this model as well.
One of the most obvious and glaring being
that the persister phenomenon is not peculiar to the in vivo environment.
In fact, even if we grow a clonal population of bacteria in an axenic culture
-- just in a tissue culture flask in vitro --
and then subject them to a drug like isoniazid,
we see that the killing curve is at least biphasic, if not multiphasic,
just as it is in vivo. Now, the kinetics of killing
are quite different in vitro, as compared to in vivo.
I don't want to say that they're the same.
But, the biphasic shape of the kill curve in vitro is essentially the same as what we see in vivo,
suggesting that the persister phenomenon, although it may be
modulated by the tissue environment, is not absolutely dependent,
perhaps, on the tissue environment.
In a way, this is good news, because it means that we can study this
phenomenon in a test tube, rather than in an animal,
and that, of course, greatly accelerates the research on understanding
this persister phenomenon.
So, the fact that there are no obvious environmental inputs
in this axenic culture for generating a heterogeneity of populations
suggests that, instead, purely stochastic factors...
random fluctuation from cell to cell in gene expression, inter-division time, and so on
could play an important role in generating the persister phenomenon.
So, we would like to postulate that stochastic processes may, in fact, underlie
the fundamental phenotypic switch to the drug-tolerant persister fate.
Now, this does not, of course, preclude a role for environmental factors.
In fact, environmental inputs could have a modulatory effect --
a stimulatory effect on the rate of switching -- even if the underlying switch mechanism
is essentially stochastic. And there are excellent examples of this from model organisms
that have been well studied.
Probably the best studied of which is the switch
that occurs at the lac operon in Escherichia coli.
So, I'm sure all of you have studied the lac operon at one time or another
in your undergraduate studies. I'll just remind you of
the components of the system: the lac operon encodes a beta-lactamase,
encoded by the LacZ gene, a permease (LacY)
that brings lactose into the cell,
and the beta-galactosidase can then act on that lactose that's brought into the cell
to convert it in part to allolactose, which is the true
inducer of the system.
Allolactose then binds to the LacI repressor
and causes it to fall off the DNA, thereby derepressing
expression of the lac operon.
So, again, the sequence of events is lactose enters
through the LacY permease, that lactose is then converted to allolactose
by beta-galactosidase, that then causes derepression by binding to the LacI repressor
and causing it to fall off the DNA, leading to derepression of the operon.
So, it's a very simple, lovely architecture, but it obviously can't work
as I've drawn it here, because in an uninduced cell
where the lac operon is repressed, there's no LacY.
There's no permease to bring the lactose into the cell.
And even if the lactose could enter through, say, simple diffusion,
there's no beta-galactosidase to make the true inducer, which is allolactose.
The reason why this system works is that, in fact,
it's not perfect. It's leaky. In a subset of cells, there's always some
expression of the lac operon, and this is simply because repression
depends on the continued interaction of the LacI protein and the LacO operator
on the DNA. And like any intermolecular interaction,
it breathes a bit. Sometimes the LacI just falls off randomly,
and when this happens, there's some transcription of the lac operon,
leading to some production of LacZ and LacY.
And that subset of cells in which these events occurred are now primed
to respond if lactose happens to be added to the culture at this time.
So there's always a subset of cells, even in a clonally homogeneous population --
a genetically homogeneous population that are primed, if you like, purely by accident
to respond to lactose if it should happen to enter the environment.
The key, as I said, is LacI and the interaction of LacI with its operator.
Now, it's important to note that, although this is essentially a stochastic switch
that occurs when LacI happens to fall off the operator,
the rate at which actual induction can occur and switching to the Lac on state
is modulated by environmental factors, most notably by the presence
of other sugars like glucose in the environment.
Glucose is a preferred sugar over lactose, so when that is present
in the environment, even with addition of lactose
the rate of switching is modulated because glucose blocks
the uptake of lactose, and it also represses the CRP transcription factor,
which is required for expression of the lac operon.
So, here we have an example of what is fundamentally a stochastic switch
to a state that's primed for responsiveness to lactose,
but where that phenotypic switch can be modulated
by the presence of other environmental factors like glucose.
What this kind of architecture then leads to
is a bi-stability of the response of the bacterial population to an inducer like lactose.
When lactose is added those cells that are primed to respond,
respond completely and rapidly,
and they induce the lac operon to its full state.
In those cells in which LacZ and LacY are still absent when lactose is added,
they're essentially blind to it because they can't take up the lactose,
and they can't convert it to allolactose,
and so they stay in the uninduced state until, by random chance,
LacI falls off of its operator and allows those cells to become primed.
Because this is essentially a positive feedback loop, once cells have entered a state
in which they're capable of being induced, they are very rapidly induced to the full state.
So, in a situation like this that is shown graphically here,
if you look at a population of cells after addition of an inducer like lactose,
where the cells have been engineered to express GFP once the lac operon is turned on
(that's the green fluorescent protein, a marker that can be used on a single cell basis),
what you find is that the population of induced cells consists of two discrete subpopulations.
These white cells are cells that were not primed to respond to lactose
at the time this inducer was added, and therefore they have not.
So, the lac operon is still off in these cells.
In contrast, these green cells were primed to respond, and they have
by completely inducing the lac operon.
There's no intermediate state.
So, in fact, cells are on or they are off -- they are not partially on or off.
So this is a bistable phenomenon,
rather reminiscent perhaps of the persister switch leading to a subpopulation of cells
that are refractory to chemotherapy,
although in that case, we do not understand the underlying mechanisms.
And this bi-stability of the population that I've shown you,
where simultaneously in a clonal population, two very different cell types can coexist,
I think underscores the inadequacy of the traditional approaches
that many of us have been taking to study phenomena in microbiology,
cell biology, and so on,
based on averaging the behavior of cells, using what I call batch culture methods.
So that would typically be to take a population of cells growing
in a shaker flask, and then to pool the cells, to grind them up,
extract (say) an enzyme, and measure enzyme activity,
or the amount of DNA, or this sort of parameter.
These approaches are very good for giving you an average value,
which I'll call alpha, of some parameter in the population.
But, these approaches reveal nothing whatsoever about the underlying population structure,
which can be quite heterogeneous, and which, for the same value of alpha
can represent very different underlying population structures.
And that's illustrated in a simplified form graphically here,
where I have schematized on the y-axis number of cells
occupying a particular phenotypic state
versus on the x-axis some particular cell parameter.
That could be expression of a particular enzyme, a particular cell growth time,
interdivision time... any parameter than can be quantified.
So, my point here is that in these four different populations that I've illustrated,
the average value (alpha) is exactly the same for these four populations, and yet,
clearly, these are very different populations of cells.
So, this average value alpha could represent a simple, narrow Gaussian distribution,
as illustrated in this panel, or a broad Gaussian distribution, as indicated here.
It could even, at one extreme, represent a bimodal distribution of cells,
in which the average value, in fact, is not represented by any individual cell
within the population whatsoever.
So that average value is extremely misleading vis-a-vis the actual phenotype
of individual cells within the population.
This bimodal or multimodal distribution can also occur with the
interesting phenotype, if you like, being represented by a small minority of the population.
This would be representative, for example, of the persister phenomenon,
in which the cells in this small bump here are the ones which we're actually interested in,
and yet, if we were to take a batch culture approach to studying this phenomenon,
this is the signal we would get --
the signal from the cells that we're not interested in --
and we would learn nothing about the behavior of this subpopulation of cells,
which is in fact the target that we want to study.
So, we and many others now are taking new approaches to
the study of bacterial behavior at the single cell level --
that is, approaches that will allow us to capture the full complexity of cellular behavior
and the variation between cells and behavior of particular characteristics.
One of the approaches in my lab that we're taking is the use a combination
of microfluidic culture of bacteria
and time-lapse microscopy to study the behavior of large numbers of
individual cells before, during, and after imposition of a stress,
such as antibiotics.
We do this, as I said, using microfluidic devices
that we fashion using soft lithography,
and which are illustrated schematically here.
They consist, in this case, for example,
of a simple glass coverslip on which the bacteria are seeded,
which is the overlaid with a semipermeable membrane,
so that the bacteria are growing right underneath the membrane,
sandwiched between the membrane and the coverslip,
We this overlay this with a block of material that's been fashioned by soft lithography
with an input port for media to flow in, an output port for media to flow out,
and flow channels cut through it for the media to flow across the surface of the membrane.
So this is essentially a sandwich, through which we can pump media
that feeds the bacteria. They grow happily on the coverslip,
fed by the material diffusing through this membrane.
And, using a fully motorized and computer-controlled microscope
that can visit as many as a hundred different points on a coverslip
and remember where it's been
(so it can keep coming back to the same points time and time again),
we can collect, over weeks of observation, even months of observation,
in fact, the detailed behavior of a bacterial population at single-cell resolution.
So, first and most importantly, we find that
when we grow bacteria in these microfluidic devices
and then expose them to a drug like isoniazid,
the kill curve that we see is clearly biphasic.
There's a period of rapid killing, followed by a period in which killing essentially stops.
So, the phenomenon that we're trying to study,
this biphasic killing response to the drug isoniazid
can readily be studied at single-cell resolution using these microfluidic devices.
So the beauty of this approach is that it allows us to study the genealogy, if you like,
of individual cells. We can follow them by recording their behavior
through multiple cell divisions
before we impose a stress, such as the drug isoniazid.
We then expose them to the drug for whatever period of time we desire,
which of course leads to the death of most of the cells
with the exception of the rare persisters.
But then we can readily identify these persisters that failed to lyse
and then recover growth after we wash out the drug,
and now we have a complete record through time of
the behavior of these cells and their ancestors.
So, we can now ask, is there anything that we can identify that's different
about these cells, either before, during, or after the exposure of the drug
that causes them to be persisters when 99.9% of their siblings are not.
So, using this type of approach, it was possible
for the first time to test a hypothesis that was proposed many years ago
(in fact, more than 60 years ago now)
by Joseph Bigger in a classic paper, where he looked at the response
of staphylococcus to treatment with penicillin
in a test tube, and for the first time observed this
persister phenomenon and coined the term persister.
He suggested -- it was purely hypothetical --
it was quite a brilliant insight, in fact,
that the persisters might be insensitive to the penicillin
because they might be temporarily in a dormant or non-dividing phase,
similar to bacteria that have exited the cell cycle due to, say, nutrient starvation.
And, this would make sense, because penicillin kills bacteria only when
they are undergoing active cell growth and division.
So that was Bigger's hypothesis back in the 1940s.
It had to wait 60 years, in fact, for experimental validation using newly developed tools --
for time-lapse microscopy and microfluidics.
And this was reported in a paper published in the year 2004
by Balaban and colleagues
(there's the reference for those who are interested),
in which they used microfluidic cultivation of Escherichia coli
to study, at single-cell resolution,
the response of these bacteria to the drug ampicillin, a beta-lactam.
So, the persister in this case is circled here in red.
So, I'll ask you, when the movie plays,
to focus on that cell.
As you can see, the cells around it are actively growing and dividing,
but that pair of cells is not growing.
The ampicillin is added, and the cells die,
but when it's washed out, these cells are capable of resuming growth.
And, in fact, if you follow them long enough, they eventually resume rapid growth
that is indistinguishable from the growth rate of their siblings
who died in response to the drug.
So, using this type of approach, Balaban and colleagues
were able to demonstrate that Bigger's hypothesis was, in fact, correct --
that the persisters comprise a pre-existing [subpopulation], and that's the important point.
This is not an adaptation to the drug. It's a pre-existing subpopulation of
bacteria that are temporarily in a slow or non-dividing state,
and are therefore protected from the killing action of this drug.
The actual mechanisms that are responsible for generating these persisters
are likely to be stochastic, but they have not been identified.
Likewise, for reactivation of these persisters after washout of the drug.
So that's a major challenge for the future is to try to understand the molecular processes
that allow these cells to become persisters.
But, at least at the cellular level, in the case of E. coli,
it seems to be very clear that Bigger, in fact, was correct.
Persisters are a pre-existing subpopulation of non-growing cells.
So, we began to collaborate with this group to study,
using these same approaches, the persister phenomenon in Mycobacteria,
focusing at first on the drug, isoniazid.
We had a rather strong bias at the outset that the persister phenomenon in Mycobacteria
was going to be very similar for the reasons I've already described,
and which Mitchison proposed -- mainly that
drugs like isoniazid kill bacteria only when they are actively growing and dividing,
and therefore, it would make sense if the persisters would be these cells,
that are in a temporarily non-growing state.
So, that was our assumption when we began these experiments...
or our hypothesis, rather,
when we began these experiments.
But, the beauty of doing science is that our ideas about the world
are never as interesting and complex as the world itself.
And so, we found, in fact, very quickly
that this idea was wrong, and that what had been discovered in E. coli, in fact,
apparently does not apply to Mycobacteria.
That's shown here -- a typical experiment in which we have grown
Mycobacteria, in this case in a microfluidic device.
The bacteria have been engineered to express GFP, as in the
previous E. coli experiments.
That's why they glow green.
So, what you're going to be looking at is bacteria grown in 7H9
initially, which is just basic broth with no drug added.
We'll then switch to isoniazid-containing medium, and that will be shown by
the INH appearing in the corner there
and look at the response of cells to the drug.
We will then wash out the drug.
7H9 will reappear.
And we will look at the regrowth of the rare persisters that survived exposure to the drug.
So, if we can roll the film, you see
cells are actively growing and dividing in the absence of drug.
As soon as we add the drug, they cease growth
within minutes of addition of the drug.
It's really an extremely rapid response.
And then the cells undergo lysis.
So, they stop growing very rapidly, then there's a delay of about a day or so
where cells began undergoing lysis.
This particular experiment involved recording the bacteria for a period of about 1 week.
So, this is a very compressed time course here.
Now, remarkably, after prolonged exposure to the drug,
the subset of surviving cells starts to grow and divide again --
at slower rates than in the absence of drug --
but, very appreciable rates.
Some of the progeny of those divisions die, some survive.
Those that survive long enough
can then, when the drug is washed out,
reactivate growth, and in fact, they do so immediately,
and they resume very rapid cell growth
within just a few minutes of washout of the drug.
So, the response of the bacteria, in this case, is very different
than what was seen by Balaban and colleagues in the case of E. coli
responding to the beta-lactam, ampicillin.
Now, using this type of approach, we can, as I said,
study the responsive bacteria, at single-cell resolution.
And this has made it very clear to us that the behavior of these cells
is even more heterogeneous than we might have thought.
One example of this is in the movie shown here
where you're going to see a typical cell,
but one that we think is behaving in an extremely interesting way
that we think is likely to be very informative about how the drug
is actually acting on these cells.
So, we can roll the film.
What we see again is that cells are growing and dividing in 7H9.
We add the drug, they arrest
very rapidly. They swell a bit, and then
they begin to undergo lysis
and disappear from the frame.
But, focus on the cell over here.
It's going to do something rather extraordinary.
It arrests, but then, in the continued presence of the drug,
it eventually resumes very rapid cell growth.
In fact, it's growing as fast now, as if there's no drug present at all.
But clearly, this was not a drug-resistant cell
because it did eventually die.
In fact, it appears that division is inhibited in the cell
even though cell growth is not.
So, my point in showing this sort of rare occurrence
is that there's an enormous amount of heterogeneity in the behavior
of individual cells within a bacterial population
that simply isn't captured by the kinds of batch culture methods we typically use
to study bacterial behavior.
My personal feeling is that the future of microbiology is
going to be to focus on the individuality of bacterial behavior,
rather than to stick with the averaging methods we have used in the past.
I think this is going to be a particularly important approach for us to take
when we want to study the behavior of important subpopulations
like the persisters.
So, there are a number of points here that I want to make,
in terms of comparing and contrasting between E. coli.
The first and most obvious one
is that, in contrast to E. coli, where cells
that turn out to be persisters after drug exposure
are already in a very slow or non-dividing state
prior to the addition of drug, this simply isn't true in Mycobacteria.
So, in fact, if we quantify the growth rate prior to drug addition
of cells that die, over here, versus cells that survive, over here,
in a particular experiment
(these are data from one experiment),
we find that there's absolutely no difference in the average growth rate
or in the spread of growth rates
between these two populations.
So, Bigger's hypothesis apparently is true for E. coli
responding to ampicillin, but it is not true for Mycobacteria responding to isoniazid.
We do not, at this point, know whether this is a difference
in the species that we're looking at or a difference in the drugs that we're looking at.
This remains a challenge for the future.
So, that's the first take home lesson from this sort of compare and contrast...
that in E. coli, persisters are rare, slow-growing cells
and they exist prior to antibiotic exposure,
whereas in Mycobacterium, the persisters are growing and dividing
at completely normal rates prior to antibiotic exposure,
so it's a very different phenomenon.
The second point of contrast is that in E. coli,
the non-persisters (those cells that actually die in the presence of the drug)
continue to grow at nearly normal rates after exposure to the antibiotic,
so the antibiotic doesn't stop them at all.
They keep on elongating until they lyse.
And that cell lysis occurs actually quite rapidly after exposure
to the drug.
In contrast, in Mycobacteria
the non-persisters -- that is the bulk of the population --
arrest growth really within minutes after exposure to this drug, isoniazid.
So, it's a very rapid response to the drug.
But then, cell lysis begins only after quite a prolonged delay of about 24 hours or so
after initial exposure to the drug.
So, it's a completely different, really opposite response
to drug exposure.
In E. coli, after washout of the drug in a microfluidic device...
In the continued presence of the drug,
the persisters do not resume growth
as long as we continue to pump it through the microfluidic device.
In fact, they do not resume growth until after washout of the drug.
Whereas in Mycobacterium, the persisters, as you saw,
resume growth and division after prolonged (that is, many days)
of antibiotic exposure, although some of the progeny
of those divisions, in fact, do not survive.
And when we washout the drug in a microfluidic device,
we find that E. coli persisters reactivate with slow and stochastic kinetics
after antibiotic washout.
So, it's almost as though it's a process of radioactive decay determining
whether a cell can reactive at a given time or not
after drug washout.
Whereas, in Mycobacteria, those persisters reactivate almost immediately
or very rapidly after drug washout and with uniform kinetics.
So, it seems very clear that the drug is actually keeping in
this non-dividing state. As soon as that pressure is removed,
the bacteria start to grow again.