Abstract:
The discovery of antibiotics revolutionised healthcare, greatly reducing the incidence
of bacterial infections. Unfortunately, however, bacteria are under strong selective
pressure to evolve resistance to antibiotics, and this, along with the over- and mis-use
of antibiotics and the dearth of development of new antibiotics, mean the world is
now facing an antimicrobial resistance crisis.
Antimicrobial peptides (AMPs) are a type of antimicrobial agent that often act
by disrupting the bacterial cell membrane, which allows less potential for resistance.
Most classes of AMPs exhibit regular secondary structure, and these are also the
most well understood. Some, however, in particular non-ribosomally synthesised
AMPs, do not form secondary structure, and may include non-natural amino acids,
be lipidated, or form cycles. This thesis focuses on laboratory synthesised variants of
two types of non-ribosomally synthesised AMPs, battacins and polymyxins.
The overarching goal is to understand which chemical features of AMPs are most
important for their activity both to rationalise the differences in activity of existing
variants as well as enable better prediction of even more active variants. Molecular
dynamics (MD) simulations are used to probe the peptide-membrane interactions of
battacin (chapter 2 and 3) and polymyxin (chapter 4) analogues at an atomic level of
detail, and, in chapter 5, Markov state models (MSMs) are used to characterise the
major structural states of two battacin analogues and quantify the transition rates
between them in order to further understand their behaviour.
Chapter 1 contains an introduction to antimicrobial resistance and AMPs, including
their different types, modes of action, and clinical usages. It then briefly introduces
microbial cell membranes, before seguing into a detailed description of how MD
simulations work. Finally, there is a brief contextual summary of MSMs, which are
described in further detail in chapter 5. Chapter 2 comprises a research article published in the journal ACS Omega
(Chakraborty et al., 2021) describing the use of MD simulations to investigate the
mode of action of two novel linear battacin analogues, an octapeptide and a pentapeptide.
These two AMPs are unusual in that they have activity against both Gram negative
and Gram positive model species (De Zoysa et al., 2015). The simulations of single
copies of these peptides in the presence of model E. coli and S. aureus membranes
showed that, in agreement with studies of other AMPs, the positively charged 2,4-
diaminobutyric (Dab) residues were important for binding to the membrane surface,
with the importance of the positive charge confirmed by simulation of uncharged
NH2-Dab analogues. Insertion of the hydrophobic portions of the lipid moieties and,
to a lesser extent, hydrophobic amino acid side chains, into the membrane core also
made an important contribution to the peptide-membrane interaction energy, with
such insertion also having been observed for other AMPs.
Chapter 3 investigates four additional linearised battacin analogues, again in
the presence of both E. coli and S. aureus membrane models due to experimental
evidence of their activity against both Gram negative and Gram positive bacteria
(Yim et al., 2020). Here, in addition to simulating single peptide molecules with the
membrane models, multiple peptide molecules in solution and in the presence of the
membrane models were also simulated. The two analogues with linear acyl chains
were found to aggregate in solution, which appeared to slow their interaction with
the membrane due to some molecules in each aggregate being occluded from the
membrane surface. Overall, however, the results were similar to those observed
in chapter 2, with positively charged residues being important for interactions
with the membrane surface, and penetration of the hydrophobic portion of the
lipid(s) and, to a lesser extent, hydrophobic amino acid side chains, providing an
energetically favourable anchor. The tert-butyl benzoate alkyl group of one analogue
had substantially more favourable interactions with the membrane compared to the
linear acyl chains, although it is possible that this was in part due to the lack of
inter-peptide interactions that might provide alternative binding partners for the
linear chains.
Chapter 4 describes the use of MD simulations to investigate polymyxins, another
type of AMP. As with the peptides investigated in Chapter 3, a host of analogues of the
naturally occurring polymyxin B3 were synthesised and tested experimentally. Both
single and multiple molecules of a select subset of these were simulated in solution
and in the presence of the model E. coli IM. A similar pattern was observed for the
polymyxins as for the battacin analogues described in chapters 2 and 3: positively
charged residues such as Dab are important for binding to the membrane surface,
and lipid groups are required for anchoring and membrane disruption. Too many
lipid tails is not necessarily helpful though: aggregation in solution was observed
for the triply-lipidated analogue, providing an explanation for its lack of activity
experimentally.
Chapter 5 contains a detailed background to MSMs, followed by their use to
characterise the dynamics of two of the battacin analogues from chapter 3, the
negative control, Ba-S, and the highly active Ba-t4. A set of features derived from the
internal coordinates of the peptides during 1500 ns of MD simulation were extracted
and subjected to time-lagged independent component analysis (tICA) to identify
the features related to the slowest time scale motions. These were then discretised
using the Common nearest neighbour (CNN) clustering algorithm in a hierarchical,
iterative manner. Lastly, the core-sets of features identified by clustering were used
to build MSMs. Unfortunately, a quality check revealed the core-sets and thus the
MSMs to be of poor quality, due to either insufficient sampling in the MD simulations
and/or poor discretisation. The MSMs were therefore not able to be interpreted with
respect to differences in behaviour resulting from the different chemical structures of
the two peptides or their environment.
Altogether, the work presented in this thesis has confirmed the importance of
two factors known to be key to AMP activity: the presence of positively charged
moieties that form attractive Coulombic interactions with the negatively charged
surface of bacterial cell membranes, and the requirement for one or more lipid groups.
For the non-ribosomally synthesised AMPs studied here, the positively charge is
typically provided by non-natural amino acids such as Dab. Unfortunately, these are
also the cause of the nephrotoxicity of the non-ribosomally synthesised AMPs that
have been used clinically, polymyxins B and E, so there is a need for analogues that
reproduce the favourable membrane interactions while reducing toxicity. The lipids
can be hydrocarbon chains, acyl groups, or geometrically more complex alkyl groups.
More than two such groups can be detrimental to AMP activity, however, as they can
stimulate aggregation of the AMPs in solution that can then reduce their propensity
to interact with the membrane. Long hydrocarbon chains can have the same effect.
Hydrophobic amino acid side chains, such as the benzyl group of Phe, insert to a
lesser degree compared with the attached lipids, and appear to be insufficient for
high levels of activity.