Successful NIHR grant

Last year I coordinated a bid to the NIHR for capital to improve our research capacity to study antimicrobial resistance (AMR) at the Oxford Biomedical Research Centre. We were successful and were awarded £1.8 million to fund several different activities, including developing vaccines to prevent the spread of AMR.


Previously in the John Radcliffe hospital Clinical Microbiology had one small second-generation genetic sequencer; now as a result of the grant we have a second, but more crucially, two very high-throughput third-generation genetic sequencers. These are GridIONs from Oxford Nanopore and sequence DNA in a completely different way that could revolutionise the use of genetics in Clinical Microbiology.


Grants like this all too-often often focus on the experimental equipment at the expense of the compute and storage you need to analyse and store the data. We were fortunate to secure funds to provide a small processing cluster in the room next to the sequencing facility in addition to much larger storage and compute at the Big Data Institute, part of which my group will be able to use to continue to develop methods for de novo prediction of the effect of individual protein mutations on the actions of antibiotics.



New Publication: Predicting whether mutations confer resistance to an antibiotic

Access the recommendation on F1000Prime

Due to the rise of antibiotic resistance, it is increasingly important that your clinician knows which antibiotics will work (and which will not). Traditionally, this is done in hospital microbiology labs by growing a sample taken from the infection site, and then testing how a range of antibiotics affect its growth, or, ideally, kill it. Despite automation, this is a expensive, open to interpretation and can take days or weeks, depending on the organising.

Given the dramatic reduction in cost of whole genome sequencing, it is comparatively cheap and straightforward to simply sequence the genome of the infecting pathogen and then, by comparing to a reference genome, mutations in genes known to confer resistance to antibiotics can be identified. If the mutations have been seen before, and so their effect on the antibiotic is known, this information can be returned rapidly to the clinician. The rub is if the mutation is novel, or just has not been seen enough times. Hence methods that can predict the effect of individual protein mutations on the action of an antibiotic are needed.

This paper demonstrates one such method. The key hypothesis is that bacterial mutations that confer resistance do so by reducing how well the antibiotic binds (but at the same time not affecting the binding of the natural substrate of the protein, otherwise the bacterium would die). Hence to predict if a mutation confers resistance, you have to calculate how the mutation affects the binding free energy of the antibiotic (and maybe also the natural substrate).

Here we validate this approach on trimethoprim in S. aureus. Trimethoprim binds and inhibits the DHFR protein, which is encoded by the dfrB gene. We use alchemical free energy methods to calculate the effect of seven different protein mutations on the binding of the trimethoprim and the natural substrate, dihydrofolic acid. We are able to not only distinguish the three mutations known to confer resistance from the four that do not, but are also able to predict the size of the effect.

The application of genetic sequencing in clinical microbiology is not science fiction; it is now routine in England for new cases of Tuberculosis, hence methods like this are urgently needed. In future work we will investigate (i) optimising the method and (ii) applying it to some of the proteins in M. tuberculosis responsible for antibiotic resistance.

You can view the paper, either on Cell Chemical Biology’s website (which may require a subscription), or you can get a version from my website. If you want to reproduce some of the calculations, you can download the GROMACS input files from this GitHub repository.


Automated detection of bacterial growth on 96-well plates (AMyGDA)

I am involved in an international collaboration, the Comprehensive Resistance Prediction for Tuberculosis: an International Consortium (CRyPTIC), that is collecting 30-50,000 clinical samples from patients with tuberculosis (TB). Although often viewed as a historical disease, TB kills more people globally than any other infectious disease, with 1.7 million people dying from it in 2016. The ultimate goal of CRyPTIC is identify as many of the genetic determinants of antibiotic resistance  in TB as possible, thereby enabling the rapid and accurate diagnosis of individual TB cases by examining the genome of the pathogen. Crucially, the clinician is provided with a list of which antibiotics are likely to be effective and which will not. Each sample collected by CRyPTIC therefore has the genome of the infection M. tuberculosis (MTB) pathogen sequenced and its susceptibility to 14 different antibiotics determined using a bespoke 96-well AST plate manufactured by Thermo Fisher. Each plate is inoculated for two weeks and then each well is examined by a laboratory scientist to determine if MTB has grown or not. Since each drug is present at a range of concentrations, the minimum concentration that kills the bacterium can be determined. A photograph of each plate is taken at the point of reading. For more detail, please see my Research page.

A potential weakness in this approach is that assessing each plate for growth is a subjective task; often it is straightforward, but some plates are difficult to “read”, usually because MTB has not grown well. To allow the project to objectively compare the results of different laboratories I developed some software, called the Automated Mycobacterial Growth Detection Algorithm (AMyGDA), that first processes the photograph to improve contrast, then identifies the location of each well and finally assesses whether MTB is growing in each well. According to our preliminary study, a preprint of which is free to download from the biorXiv, AMyGDA is sufficiently reproducible and agrees well enough with the human readings that we could use to supplement measurement by laboratory scientists in the CRyPTIC project.

You can download the AMyGDA software here. It is a python module and instructions on how to install the prerequisites are included, as is a short tutorial and a number of test images.

I will be shortly submitting the manuscript to a peer-reviewed journal and I will update this post when it is accepted and published.



New Publication: Protein crowding affects the organisation of ion channels

Protein crowding and lipid complexity influence the nanoscale dynamic organization of ion channels in cell membranes

It is difficult to look at the dynamic spatial organisation of ion channels in cell membranes, but this is something coarse-grained molecular dynamics simulations can offer insights. This work, led by Anna Duncan, shows how altering the lipid composition of the membrane changes the large-scale organisation of the Kir2.2 channels. Building on some previous work, we also show how the membrane properties, such as stiffness, are also altered. The latter relies on some python code that you can download from GitHub. The paper is free to download.


New Publication: Effect of SAO mutation on Band 3

There is a lovely story behind this paper just published earlier this week in Biochemistry. Reinhart Reithmeier came to visit Mark Sansom in Oxford whilst on sabbatical back in 2002. Now Reinhart, if you don’t know, is a world-expert on Band 3 which is the transmembrane protein in the membrane of red blood cells that mediates the exchange of bicarbonate ions (and hence, effectively, carbon dioxide). In particular he was interested in the effect that an inherited nine-residue deletion has on the first transmembrane helix (TM1). These deletion is known as South Asian Ovalocytosis (SAO) and has serious consequences for the individual but also offers some protection against the parasite that causes malaria.

Since there were some NMR structures of wild type and SAO TM1 available at the time, Reinhart persuaded a few people in Mark’s lab to run some simulations on a couple of the NMR structures from the NMR ensemble. In total 22 ns of molecular dynamics simulation was run, much of it in octane rather than a lipid bilayer. A manuscript was prepared and sent to the Biophysical Journal and the reviewers were interested but did not believe that the simulations were representative of the dynamics of TM1. Witness this comment

The 10 ns WT-NMR-PC simulation is unacceptable. This simulation has gone wrong as a consequence of poorly selecting their starting structure.

At the time this study was pushing the boundaries on how long transmembrane helices could be simulated before, yet this was not enough. Unfortunately by the time the comments were returned, the lead author had left the lab and science and so the project lost steam.

Let’s fast forward to 2013, eleven years later. Reinhart visits Mark’s lab again, clutching the original manuscript and reviewers’ comments. I get interested, partly through my earlier work on the roles proline residues can play in transmembrane proteins and decide that the only way to avoid another comment from a reviewer like the one above is to simulate ALL the structures in the NMR ensembles. Not only that, but let’s repeat each one three times and also try starting from an ideal helix as well (and repeat that fifty times).

Of course, this type of high-throughput simulation was now possible; computer speeds had increased, GPUs had been introduced and GROMACS continually optimised. We could now also use coarse-grained molecular dynamics simulations to embed each structure in a lipid bilayer and, just as importantly, run high-throughput analysis using MDAnalysis, a Python module able to read and analyse large numbers of molecular dynamics trajectories efficiently and quickly.

In total I ran 4,460 ns of molecular dynamics for this study, an increase of nearly 200x over the study a bit over a decade ago. Note that Moore’s Law alone is responsible for 50-150x so my main advantage was simply starting the simulations a decade later. This is a lovely illustration of how the gradual but relentless increases in computer speeds (along with other advances) have allowed us to push the boundaries of simulating the behaviour of biological molecules.

New Publication: Lipids can form anti-registered phases

When we think of lipids phase separating in a cell membrane we usually think of this process occurring symmetrically, i.e. with like on top of like (this is described as a registered phase). If we consider the simplest case of two lipids, one saturated (A), one unsaturated (B), then if their lengths are sufficiently different (i.e. a high degree of hydrophobic mismatch), then theory suggests that A/B and B/A is energetically more favourable than A/A and B/B. Such an asymmetric arrangement is described as an anti-registered phase. The theoretical paper demonstrated the effect using a very simple model that was a long way from biology.

In collaboration with two condensed matter physicists, John Williamson and Peter Olmsted, we ran a large number of coarse-grained simulations using the MARTINI forcefield using a mixture of 3 lipids that has been previously shown to phase separate. By varying the number of beads in the tail of the saturated lipid, we were able to increase or decrease the period of time the system spent anti-registered. The theory also predicts that decreasing the size of the periodic box will favour anti-registration, which we were also able to confirm.

This is important, because it is usually assumed that lipids in a cell membrane phase separate in a registered manner, leading to local regions enriched in cholesterol that are usually called ‘lipid rafts’. This study, when combined with our work showing that the cytoskeleton can lead to membrane compartmentalisation suggests that there could be small, dynamic patches of anti-registered lipids forming in the corals produced by the crowding of membrane proteins and the effects of the cytoskeleton.

What is really nice about this study is how it came about; I was at the Biophysics meeting in Baltimore in early 2015 tweeting away and bumped into John Williamson. We went for a coffee with Peter Olmsted and they told me how they’d noticed a tiny uptick in the percentage of anti-registration at the start of one of simulations in this paper and that this might agree nicely with their new theory. The saturated lipid in those simulations had 4 beads in each tail, so I agreed to try increasing it to 5 beads per tail and seeing if that led to a prolonged period of anti-registration.

Sure enough, it did, hence the paper.

This is the third and last in a set of three papers that bring my research on cell signalling and membranes in the SBCB group within the Department of Biochemistry to a close and is available to download here for free.

New Publication: Proteins Alter the Stiffness of Membranes

Although there have been many studies of proteins whose primary function is to ‘sculpt’ the surface of membranes e.g. BAR domains, there have been very few investigations of what effect regular membrane proteins have on the stiffness of membranes. Here we show via very large simulations, using the MARTINI coarse-grained forcefield, that ‘regular’ integral membrane proteins, such as an ion channel or a beta-barrel, reduce the stiffness of the membrane, leading to larger fluctuations. The systems studied push the boundaries of what is currently achievable with biomolecular simulation, containing around 50,000 lipids and 100 proteins. We had access to the French supercomputer CURIE, through the EU PRACE network, for this work.

This is the second in a set of three papers that bring my research on cell signalling and membranes in the SBCB group within the Department of Biochemistry to a close and is available to download here.

New Publication: Membrane Compartmentalization Reduces the Mobility of Lipids.

Lipids are not free to diffuse around the cell membrane. Rather they are constrained not just by all the embedded proteins but also by the cytoskeleton, which, it has been suggested, corral the lipids. In this paper, we show by very large coarse-grained simulations of a realistic model of the plasma membrane how compartmentalisation leads to reduced, anomalous diffusion of both lipids and proteins.

This is the first in a set of three papers that bring my research on cell signalling and membranes in the SBCB group within the Department of Biochemistry to a close and is available to download here.

New Publication: Predicting affinities for peptide transporters

PepT1 is a nutrient transporter found in the cells that line your small intestine. It is not only responsible for the uptake of di- and tai-peptides, and therefore much of your dietary proteins, but also the uptake of most β-lactam antibiotics. This serendipity ensures that we can take (many of) these important drugs orally.

Our ultimate goal is to develop the capability to predict modifications to drug scaffolds that will improve or enable their uptake by PepT1, thereby improving their oral bioavailability.

In this paper, just published online in the new journal Cell Chemical Biology (and free to download, thanks to the Wellcome Trust), we show that it is possible to predict how well a series of di- and tai-peptides bind to a bacterial homologue of PepT1 using a hierarchical approach that combines an end-point free energy method with thermodynamic integration. Since there is no structure of PepT1, we then tried our method on a homology model we have published in 2015. We found that method lost its predictive power. By studying a range of homology models of intermediate quality, we showed that it is highly likely an experimental structure of hPepT1 will be required for in silico accurate predictions of transport.

This is the second paper that Firdaus Samsudin has published as part of his DPhil here in Oxford.

New Publication: The Extra-Cellular Domain of PepT1 and PepT2

PepT1 is a nutrient transporter found in the cells that line your small intestine. It is not only responsible for the uptake of di- and tai-peptides, and therefore much of your dietary proteins, but also the uptake of most β-lactam antibiotics. This serendipity ensures that we can take (many of) these important drugs orally.

Our ultimate goal is to develop the capability to predict modifications to drug scaffolds that will improve or enable their uptake by PepT1, thereby improving their oral bioavailability.

In Structure we report the structures of the extra-cellular domains (ECDs) of PepT1 and PepT2. This is an important milestone on the road to elucidating a structure of PepT1 and allows us to propose the first full-length structural model of PepT1 (see above). Intriguingly, the data also suggests that the ECD also interacts with trypsin, thereby increasing the local concentration of peptides around the transporter, improving its efficiency.