Which bug could it be?

A phone call late Tuesday evening broke shocking news that a colleague, A, had taken ill all of a sudden. He had complained of nausea Monday evening and had a fitful night’s sleep that left him exhausted. He slept again at 6am on Tuesday and didn’t wake up even past noon. His wife suspected something was amiss and called for help. A few hours later, at the ICU in a prominent Bangalore hospital, the doctors found A’s blood pressure plummeting. They noted severe sepsis leading to multi-organ failure and offered a bleak prognosis.

A was intubated and put on a ventilator to support his lungs. His blood pressure was raised artificially with a high dose of medication. His high fever was treated with both antibiotics (ceftriaxone) and antivirals (acyclovir), so as to cover all bases. His kidneys were barely functioning as indicated by his meagre urine output and the build up of creatinine in his blood. Dialysis would be needed to detoxify his blood and buy time for the kidneys and other organs to recover, but that would only be possible once his blood pressure came back up. His doctors waited through Wednesday for that to happen.

Fortunately, things improved on Thursday. A’s blood pressure started rising. The lungs got better. The kidneys remained ineffective, but dialysis became a possible option. Some signs of neurological activity became visible. This was enough for the doctors to revise their bleak prognosis on Tuesday night to a more cautiously optimistic prospect of recovery, albeit with the caveat that an assessment of long term damage to the organs if any was yet to be made and that continued recovery would need more time.

The root cause of the sudden turn of events on Monday and Tuesday remained a mystery though. Blood and urine samples had been collected late Tuesday and sent for culturing to identify the presence of any pathogenic bacteria. The results came much later, the following Monday, but failed to pinpoint anything relevant (in fact, culture tests fail to pinpoint the source of infection in 60% of such cases). Meanwhile, treatment had to proceed empirically based on the assumption that there was an infection from an unknown agent and that the antibiotic/antivirals chosen would be appropriate. CT scans showed inflammation in the intestines and in the bladder, indicating a possible infection, but nothing that seemed sufficiently strong to explain A’s multi-organ failure, let alone explain the speed at which he had gone from healthy to critical. COVID tests were of course negative.

So, on Friday afternoon, in a brainstorming chat, we started to explore the use of DNA sequencing to pinpoint the cause—presumably an offending bacterium or virus. This was by no means a standard process yet, but a few proofs of concepts had been published. The idea was to take A’s blood and trap the free floating DNA fragments therein. These so-called cell-free DNA (cfDNA) fragments are released into the blood when cells anywhere in the body die. For the most part, these would come from A’s own human cells. But bacterial and viral cells might also die under the onslaught of the antibiotics and antivirals that A was put on, and DNA fragments from these cells would also find their way into his blood. If we could trap many of the cfDNA fragments in his blood, sequence these fragments with modern DNA sequencing technology, and compare these fragments to viral and bacterial DNA sequence databases, we might be able to pinpoint the offending microbe. But there were many unknowns to contend with.

First, A had been on antibiotics/antivirals for two or even three days by now. Would the offending microbe still be present in A’s body and would the DNA from this microbe still be present in A’s blood? After all, cfDNA has a half-life of only a few hours in the blood and gets cleared rapidly. It would have been ideal to get a sample of A’s blood on Tuesday or Wednesday. Was Friday too late?

Second, once we collected A’s blood, we would have to extract cfDNA fragments from it. The overwhelming majority of fragments would be from A’s own cells. Microbial fragments would be in an extreme minority, like the proverbial needle in a haystack. DNA from A’s cells would itself be of two types: fragments longer than 200 characters, and shorter fragments. Microbial fragments were likely to be of the shorter variety. By ridding the extracted cfDNA of long fragments, we could potentially enrich microbial fragments. Nevertheless, the proportion of microbial fragments was likely to be no more than 0.05%, and that too had we collected the sample on Tuesday. That proportion was perhaps much lower now. We worried that we might just not get a detectable signal at all.

Third, we would have to really scramble to make arrangements: get consents, brief the doctors on the technology and likely outcomes, order the relevant reagent kits, and set up the required data transfer and computing infrastructure. Fortunately, our lab had the supplies readily available and some brainstorming with the lab and computational teams indicated that all pieces could be brought together with some concentrated effort over the weekend so we could get to an answer early in the following week.

By early evening on Friday, we took the leap, knowing we may well be in for a wildgoose chase. A’s blood was drawn late on Friday after taking consent from his family and doctors. The next morning, blood was sampled from A’s wife, and three other healthy volunteers, to serve as controls. The lab team worked through the weekend to extract cfDNA from all the blood samples, eliminate the longer fragments, and prepare the rest for sequencing. The prepared fragments entered the DNA sequencer late on Sunday. Meanwhile the computational team assembled the data analysis pipeline so data would move to the cloud right after it was generated and well established off-the-shelf software tools would compare each fragment sequence to a database of microbial DNA sequences and provide a list of matching hits.

The sequencer ran for about 27 hours and the sequence data for the various fragments became available late on Monday. Roughly 25 million fragments were sequenced from A, and 19.5 million from his wife. The other three healthy control samples had to be combined into one pool because the total amount cfDNA extracted from each individually was low; 42 million fragments were sequenced from this combined pool. The combined data was over 50GB.

The data transfer to the cloud was complete by 11pm on Monday. Once there, the analysis pipelines we had set up over the weekend kicked in and ran through the night, first converting the data to a standard format and then matching the various fragment sequences to the human genome sequence. Over 99% of the fragments matched the human genome sequence as expected and were filtered out, leaving the remaining 1% as potential microbial candidates. These were then processed by popular off-the-shelf tools (Kraken2 and Centrifuge) to obtain hits to microbial genome databases. By 9am on Tuesday, close to a week since A had entered the ICU, we were ready to review these hits.

We looked for DNA fragments present in A but not in the other healthy controls and arising from microbial species known to be pathogenic (our body of course has many good bacteria as well; those were not the subject of our study). To our sheer disappointment, no such fragments stood out. Much of Tuesday and Wednesday went in a careful review of the data and the results, so we could spot any mistakes that might have happened. The raw data quality was excellent. Each step of the process seemed to hold up. Yet the final number of fragments matching microbial genome databases was just 0.005%, an order of magnitude lower than the 0.05% we had expected. Perhaps the fact that we had drawn A’s blood on Friday rather than the previous Tuesday or Wednesday made the difference; microbial counts had perhaps plummeted under the influence of antibiotics and antivirals.

On Thursday, we were on the brink of giving up. It had been 6 days from when we had drawn A’s sample, and 10 days from the time A had entered the ICU. There was a nagging doubt in our mind: had the off-the-shelf tools we were using (Kraken2 and Centrifuge) missed something? These tools were tailor-made for microbial DNA analysis but we couldn’t examine their innards quite as critically as we could for the software we built ourselves. So, we turned to Strand NGS, a software platform we had built and refined over 10 years. Strand NGS would allow us to carefully track and visualize the tens of millions of fragments we were dealing with, so we could critically review if we were missing anything at all. We rapidly assembled DNA sequences of 304 pathogenic bacteria that had been reported in patients with severe sepsis and used Strand NGS to compare the DNA fragments we had sequenced from A and from the healthy controls with these 304 sequences. We experimented carefully with various analysis thresholds so we did not miss near-hits while visually verifying these near-hits to ensure they weren’t spurious.

Our persistence was rewarded with an aha-moment early Friday morning. Four DNA fragments from A matched a bacterium strain with an arcane name: Salmonella enterica subsp enterica serovar Typhimurium str. LT2 (S. Typhimurium, for short). The matches were very clean and high quality and stood up well to cross-verification using other popular tools. None of the healthy controls had any such fragments, in spite of tweaking various thresholds to coax out such hits. There was little doubt that A’s blood had distinct traces of DNA from S. Typhimurium while the healthy controls, including A’s wife, showed no such traces. Could this indeed be the culprit? Some frenzied reading of the medical literature followed next.

S. Typhimurium is a cousin of the better known S. Typhi that causes typhoid fever. 12.5 million cases of typhoid fever are reported worldwide annually, and most commonly in India and Africa. Spread happens by food or water contaminated with the faeces of an infected person. Symptoms typically begin 6 to 30 days after exposure with gradual onset of fever over several days, accompanied by nausea, constipation and headache. Notably, diarrhea appears less common, and is at least not a hallmark symptom. A’s disease onset was far more sudden and he had marked diarrhea, both at variance with the typical course for S. Typhi, but more consistent with S. Typhimurium’s course of progression, described next.

S. Typhimurium and other related strains of Salmonella (like S. Enteritidis) are called non-typhoidal strains (NTS). In contrast to typhoid fever, NTS related diseases occur worldwide and not just in the developing world, with over 93 million cases annually. The incubation periods tend to be much shorter, of the order of hours, and inflammatory responses tend to be stronger (ref). Patients present with acute diarrhea, abdominal pain, fever, and vomiting, similar to what A presented with. Most cases are self-limiting though and stop well short of the multi-organ failure that A had experienced. In fact, antibiotics are not even recommended unless the disease spreads beyond the intestine, which it rarely does. However, in about 5% of the cases (ref), viable bacteria find their way into the bloodstream in large numbers and carry the infection to other parts of the body; a phenomenon called bacteremia. Interestingly, S. Typhimurium is among the more likely NTS to do so (ref).

In many cases of bacteremia caused by NTS, the immune system manages to control the infection gracefully. However, occasionally, the disease takes a more severe turn; the bacteria in the bloodstream invite a systemic inflammatory over-reaction from the body, called sepsis (in fact, sepsis can also happen in cases where no bacteria are explicitly detected in the blood). In sepsis, molecules called cytokines are released all over the body to fight the infection and also contain the damage caused by the infection. While beneficial to an extent, overproduction of these cytokines damages tiny blood vessels all over the body and also dilates these vessels, causing dangerously low blood pressure; taken together, this causes the organs such as the heart, lungs and the kidney to fail, a condition called septic shock, which A experienced.

Such a sudden septic shock reaction from NTS is unusual though. Most often, such extreme reactions happen in young, malnourished children, and in those with a compromised immune system (ref). A was a typical 50 year old and was unlikely to be severely immunocompromised. However, there are case reports of septic shock and multi-organ failure on account of NTS in individuals without any immunocompromise (ref). Perhaps this happens when a large dose of the bacteria is ingested; indeed experiments suggest higher doses lead to greater severity (ref). So, perhaps A was inadvertently exposed to a severely high dose of S. Typhimurium. Where could that have come from?

NTS transmission usually results from contaminated animal-derived food products (meat, poultry, eggs, and dairy products), or from person to person contact, or from contact with pets such as cats, dogs, rodents, reptiles, or amphibians, or even occasionally from contaminated plant produce (ref). Contaminated animal-derived food was the most likely source of exposure for A; in fact, he had bought and consumed some frozen meat on that fateful Monday, and that perhaps was heavily infected. All in all, the presence of the S. Typhimurium fragments in A’s blood seem to correlate well with this exposure and to his subsequent clinical course, though the breakneck speed with which the catastrophe unfolded remains unusual. How can this knowledge help doctors treat A?

S. Typhimurium infections respond more effectively to some classes of antibiotics, like fluoroquinolones and expanded-spectrum cephalosporins (e.g., ceftriaxone or cefixime), with other classes like trimethoprim-sulfamethoxazole (TMP-SMZ) and ampicillin being used less often on account of commonly reported resistance (ref). A was on Ceftriaxone for the first few days in the ICU but was taken off thereafter. The results of this analysis were communicated to his doctors on Saturday, about 11 days after he entered the ICU, and a week after his blood sample was drawn. His fever had still not subsided and his doctors restarted Ceftriaxone.

Meanwhile, A’s recovery has been as strong as the onset of his disease had been sudden. All his organs seem to have recovered, except for his kidneys. Daily dialysis is keeping things going until they do so. His mental and physical faculties seem normal and unaffected by the systemic assault he experienced. Hopefully, his fever will recede soon and his kidneys will complete their recovery shortly, and any other systemic damage that might have taken place will heal in due course as well. He may continue to excrete S. Typhimurium for a month or two even after the infection subsides completely (ref). When all is well again, he may look back at his years of writing software at Strand and feel some fulfilment that his efforts helped solve the mystery of his sudden affliction—those four DNA fragments amongst the several tens of millions—remnants of the scourge that almost brought him to the brink.

.