Bacteria That Cause Bloodstream Infections in Africa are Resistant to Nearly All Drugs
Machine learning is helping us discover high-risk diseases faster
Summary
- Researchers have uncovered a new strain of salmonella resistant to all but one antibiotic available in the Democratic Republic of the Congo
- This is part of an ongoing process of evolution, which may be spawning more dangerous strains over time
- This evolution can be detected by a recently-developed machine learning algorithm, allowing faster detection and containment of high-risk strains
Most salmonella infections result in symptoms associated with food poisoning. While unpleasant, these symptoms are usually not life-threatening.
But in sub-Saharan Africa, salmonella can evolve to cause bloodstream infections, becoming what are known as invasive non-typhoidal Salmonella (iNTS). Every year, iNTS infections are estimated to affect 3.4 million people and result in almost 700,000 deaths globally. It’s one of only two causes of child mortality (along with AIDS) that result in a higher rate of death today than in 1990.
Diagnosing bloodstream infections is time-consuming, costly and challenging, but crucial for designing strategies to reduce the rates of these infections and deciding how to treat them effectively. The Democratic Republic of the Congo, a site with a high incidence of childhood bloodstream infections, is a challenging environment for providing healthcare. Patients pay for their own care, and it’s estimated that only 30% of people have access to a regular healthcare system, while 40% self-medicate.
To help understand what causes bloodstream infections in the DRC, and how to control the incidence of disease, researchers in the country set up a program for surveillance of these bloodstream infections. Blood samples from people infected with salmonella that show antibiotic resistance are collected by hospitals and health centres. They are then sent overseas for DNA sequencing, which allows us to identify known markers of antibiotic resistance and now, with the development of a new machine-learning algorithm, assess the risk of bloodstream infection posed by each sample.
Discovery of a new strain
Through DNA sequencing, we found a new strain of salmonella in blood samples taken from children under five presenting with bloodstream infections. It made up >10% of the samples that were collected in the project and was identified in three different sites across the DRC, suggesting it is relatively widespread. By analysing the DNA of the bacteria, we can estimate that this new strain emerged in 2004, and started to take off around 2012.
Levels of antibiotic resistance appear to have increased over time as this strain has evolved, making these infections progressively harder to treat. Access to antibiotics in the DRC is limited by both price and method of administering the treatment. People can’t receive IV antibiotics like they could in a wealthier country, because it isn’t practical to keep patients in the hospital long-term. Currently, the recommended treatment for these infections is one of two remaining drugs, ciprofloxacin or ceftriaxone. But this new strain is now resistant to ceftriaxone, making this the first “extensively drug-resistant” iNTS we’ve seen. One sample we looked at also showed an increased tolerance of ciprofloxacin, the last available drug in the DRC to treat these infections. If this resistance continues to rise, these infections won’t be treatable within the current DRC healthcare context.
What’s particularly concerning is the fact that the genes that give this strain its antibiotic resistance are carried on a plasmid, a piece of DNA that can move between different bacteria. This means that if this strain becomes widespread, there’s a risk it could pass its antibiotic resistance on. More work would be needed to confirm that this plasmid can spread, and if so, how closely related the bacteria would have to be for this transfer to be successful.
Where do iNTS come from?
iNTS has been around for a long time, but the growing problem of these invasive infections has only gained recognition in the last decade. The disease usually affects children under five suffering from malnutrition and malaria, and adults with HIV.
Since the emergence of AIDS in the 1980s, a change has been occurring in this common stomach bug, and the mortality rate associated with infections has increased. It’s thought that the introduction of HIV, which has a profound impact on the way the human body interacts with and responds to bacteria, has created a unique niche. It is a niche in which bacteria that would normally be met with an intense immune response can now stay in the human body for longer and become better adapted to infecting it.
The timing and pattern of movement of HIV and iNTS suggest that invasive salmonella may be following the spread of vulnerable populations across Africa. But these bacteria have been able to take advantage of a young, immunocompromised host population for a long time, so why the sudden rise in disease now? It’s thought that the rise of the HIV epidemic created a new, internationally mobile adult population of infected people, which likely drove the spread of these invasive strains.
ST313 is thought to be moving along a spectrum of infection styles seen in salmonella, changing from a strain that can move between animals, food and humans and cause relatively mild disease, to one that’s specialised to a specific host and can cause life-threatening illness.
For the first time in a real outbreak scenario, I got the opportunity to test a machine learning model I’ve been developing for the last four years. We used the algorithm to look for characteristic patterns in the DNA of salmonella that cause invasive infections. The algorithm outputs an “invasiveness index”, which gives an indication of whether strains are shifting along this spectrum over time.
We see an upwards trend in invasiveness index with each new strain of invasive salmonella that has evolved over the last few decades, mirrored by a shift in the way these bacteria behave in the lab. An example is the strains’ ability to form complex communities of bacteria that are better at surviving tough conditions like those faced when salmonella spread through contaminated food and water. In the image above, we can see the appearance of these communities of bacteria appears to be changing with each successful new lineage, to look more like the bacteria that cause typhoid and less like their close relatives that cause food poisoning. These lines of evidence suggest that the bacteria are behaving more like invasive strains.
Identifying these new invasive strains can be tricky. It’s hard to tell how invasive a strain is, especially in humans. The ability of a machine-learning algorithm to flag new strains as more invasive gives us a promising indication that it could be used to identify other dangerous new strains as they appear.
We’ve seen major innovations occurring in these bacteria on a timescale of decades, that have allowed them to become more dangerous. Being able to recognise a new strain as it appears means we can mount a response to contain its spread. It also helps us understand and target the upstream factors that lead to their appearance and success. DNA sequencing is becoming a cheaper option for public health, and work like this illustrates the insight we gain into a complex infectious disease by examining its DNA.
Learn more
To see more about what I do, visit my website or our team website; to learn more about what we’re doing to track the global spread of antibiotic resistance, see the Global Health Research Unit.
