From millions to just four

How Strand’s NGS team was able to find the needle in a Sepsis haystack

A Tuesday, a few months ago brought bad news. An ex-colleague, struggling with nausea the previous evening, had slept in past noon. Fearing something amiss, his family rushed him to the hospital, where, in the ICU, his blood pressure dangerously low, he was diagnosed with multi-organ failure and sepsis. A blitz of common-sense treatment, antibiotics, antivirals, blood pressure meds, followed, and, once his vitals were stable, dialysis to clear the kidneys. A bleak prognosis after the fact turned to a more cautiously positive, but still fundamentally puzzled, one.

What led to the sepsis? And what could really help treat it?

 Sepsis happens when the body can’t fight off an invading microbe. Usually, when microbes enter a body, they’re immediately neutralized by first-responder immune cells. But sometimes, either because immunity is too weak or the microbe too strong, the first response can’t clear the infection, leading to a storm of “cytokines” – molecules that attack not just the invader but other organs and tissues. A cytokine storm gives the body two problems to solve: getting rid of the originating infection, and silencing its own destructive messengers. Sepsis is the immune system walking this tightrope, usually with no success. The fatality rate in cases like our ex-colleague’s, with multi organ failure, is up to 80%.

If he was to recover at all, we needed to know what was wrong – what microbe, fungus, bacteria, virus, was the invader? The tests, the blood culture and CT scans, said mainly that there was inflammation in the bladder and the intestines. Though consistent with sepsis, these findings were otherwise unremarkable.

On Friday, three days on, while he was still in the ICU, we turned to circulating fragmentary DNA (cfDNA)  as a possible answer. Dead molecules of the microbe causing his sepsis shed fragments of their DNA into his bloodstream. If we could sequence the circulating DNA in his blood, we could compare it to the DNA sequences of known pathogens, thereby identifying it and identifying in turn a specific treatment, rather than the intuitive but generic treatments he was receiving.

At least that was the theory. In practice, both biology and the clock were against us. Microbial cfDNA is tough to make out against the detritus of background host DNA, most of which is ejected from cells dying natural deaths. In our ex-colleague’s case, what made this detection harder was, ironically, the unknown effectiveness of his current set of treatments. cfDNA has a half life of a few hours. His treatment had been going on for a few days. What if there was no cfDNA left to detect?

On Friday evening, we obtained consent and drew blood from our ex-colleague as well as healthy “controls”, including that of his wife. Over the weekend, while our informatics and software teams set up the technology infrastructure that would map and identify the pathogen, our lab team worked to isolate the cfDNA from the blood, to amplify it, and to sequence it.

We found that 99% of the cfDNA was harmless, human in origin, as expected. The 1% that remained was processed through publicly available informatics tools (Kraken2 and Centrifuge, for those curious), and then subtracted off from the controls to make sure we weren’t spending our time detecting benign microbes. All in all, we expected 0.05% of the total cfDNA to map to microbial databases, and some hopefully perceptible fraction of that to be our culprit.

But our data review found nothing. While the sequenced data and the steps in the processing pipeline were unimpeachable, the output revealed only 0.005% of the cfDNA was microbial in origin. Not only was this an order of magnitude lower than our expectations, it also, more to the point, made our task of detecting the offending pathogen that much harder.

We turned to Strand NGS, our in-house tool for visualizing sequencing data, and began the task of carefully comparing the cfDNA fragments against 304 assembled sequences of pathogens reported in patients with severe sepsis.

On the Friday exactly a week after we started this journey, we had our Aha!! moment. 4 cfDNA fragments (out of tens of thousands!) matched a bacterium with an arcane name. Salmonella enterica subsp enterica serovar Typhimurium str. LT2, or S. Typhimurium, for short. Two things had made, in retrospect, this needle stand out in the haystack of background DNA. One, the visualization tool we used, Strand NGS, helped us see a DNA fragment where earlier we only saw a string of A, G, C, Ts. Second, the fact that our engineers know and have built Strand NGS helped us tweak its inner workings in a way that would be tough to do for the open-source tools with which we started this exercise, and especially tough when, literally, a life is on the line.

S.Typhimurium is not, in general, a fatal or even remarkable disease. A non typhoidal strain (NTS), symptoms of S. Typhimurium infections are commonplace — fever, headache, nausea — and are usually self-limiting. In 5% of the cases, the bacteria, originating in a tissue, makes its way into the blood, in what is known as bacteremia. But even this is typically fought off by the body, unless it’s immuno-compromised, either chronically or due, for ex., to malnourishment.

So what led to the sepsis? We think it all started the Monday before our ex-colleague was taken ill, when he ingested some frozen meat. NTS infections are known to be transmitted by contaminated animal-derived food. It probably wasn’t because he was immuno-compromised, but more because he ingested a large quantity of S. Typhimurium. In other words, it was because the pathogen was strong, not because the body was weak.

Things moved fast after the detection. S. Typhimurium infections respond to specific classes of antibiotics, like fluoroquinolones and expanded-spectrum cephalosporins like ceftriaxone. In his first few days in the ICU, our ex-colleague was on ceftriaxone but had been taken off. A week and a half later, on the basis of our cfDNA detective work, he was put back on ceftriaxone. His recovery was swift, again showing that his immune system was fundamentally healthy and just needed a boost, perhaps from the right treatment. Daily dialysis was needed for a time, until it thankfully wasn’t. Fending off a likely nosocomial attack of Covid that followed without incident, he is now, several months on, as healthy as he ever was.

A story about remission, or one about a small triumph of an emerging tool in precision medicine,  this incident also reminded us about the particular friendship and community of small and tight-knit companies like our own. Our ex-colleague worked for many years at Strand, on software similar to Strand NGS. We’re grateful that we were given the opportunity and had the persistence and technical chops to be able to help him when he most needed it. That’s a type of fulfillment hard to write about — it ripples across teams and is felt long after the event has passed.

For an in-depth look at the biology and informatics of this incident, read our CEO’s note here. For more details, you can reach out to us on