Skip to main content Skip to secondary navigation
Main content start

The future of infectious disease immunology

An expert in computational immunology details how his field is tapping into the secrets of the immune system to identify pathogens faster and serious illness sooner.
Illustration of immune system defense concept
The immune system has been developing and refining for millions of years, and we may be underutilizing it. | iStock/

When we’re sick, the time between onset and diagnosis is critical, sometimes life-saving.

It turns out the human immune system is pretty good at knowing what’s making us sick. In fact, it’s telling us all the time, but only now is science tuning in to what nature has to say, explains Purvesh Khatri. The immune system is a “perfect diagnostic,” he tells host Russ Altman in this episode of Stanford Engineering’s The Future of Everything podcast.

Embed Code

Listen on your favorite podcast platform:


Russ Altman (00:03): This is Stanford Engineering's The Future of Everything. I'm your host, Russ Altman. Today, Professor Purvesh Khatri will tell us how diagnosing and treating infectious disease can be improved by asking our immune system what's going on. The immune system has evolved and is very good at sensing what's happening in our bodies, and a new approach to understanding infectious disease is to just ask the immune system what it's detecting. It's the future of infectious disease immunology.

Russ Altman (00:34): A huge fraction of deaths worldwide occur because of infectious disease, bacteria and viruses. Covid and influenza are viruses. Strep throat and urinary tract infections are from bacteria, for example. We treat them differently. We use antibiotics for bacteria, but we have to use antivirals for viruses. Sometimes doctors don't know what they're dealing with. Well, Professor Purvesh Khatri from Stanford University figured out that you can ask the immune system what it's dealing with.

Russ Altman (01:05): After millions of years of evolution, the immune system has figured out when it's dealing with a bacteria and when it's dealing with a virus, and we just have to ask it. He'll tell us that he's developed new ways to diagnose both infection by bacteria and by viruses. He can even figure out when you're in trouble and this infection is going to be bad, and when you're going to clear the infection with no problem.

Russ Altman (01:30): Purvesh, your work focuses on diagnostics that use the immune system as a marker for what might be going on in the body, either autoimmune disease or infection. Tell me why this approach is different and why it has so much promise now, after many years of doing diagnostics in very different ways.

Prof. Khatri (01:49): Sure. If you think about out immune system, it had one job, or one of the most important jobs during evolution was find what is dangerous to the host, respond to it by proliferating, and when that danger signal is gone, shut down that immune response. As an electronics engineer, that's the perfect diagnostic, right? It's a distributed sensor ... the immune system continuously surveils ... with a built-in amplifier, so it amplifies the signal ... when there is a danger, the immune cells proliferate ... and then when the pathogen or whatever is gone, it shuts down. That is why it's so attractive.

Prof. Khatri (02:38): The second reason that I really think the immune system is going to play a crucial role in diagnostics is until now, we've been looking for a pathogen, but now it's very well established that when you look for a pathogen, the immune system actually learns how not to let the pathogen get throughout your entire body, because then it becomes sepsis and you die. If you are looking for a pathogen, people have compared it to a needle in a haystack kind of a situation.

Prof. Khatri (03:10): Here is the problem. You are not looking for a needle in a haystack. There is no needle in the haystack, because the immune system has already sequestered the pathogen. You could deep-sequence the blood until the end of time, but the immune system made sure that it doesn't get there, and that's the second reason that I really think that the immune system has a true potential to be a fantastic diagnostic.

Russ Altman (03:38): Let's step back for a second. What does a regular person need to understand about the immune system? It seems complicated, and I know you're querying it and using it as you just described. Tell me, can you give us a little primer on how the immune system works, and what are the challenges of using it for diagnostics?

Prof. Khatri (03:56): Yeah. Imagine the immune system is a network of sensors in your body, and just like any sensor, it has multiple different parts. There is a set of cells that their job is to continuously look for a danger signal, and when there is a danger signal found, isolate it and present it to rest of the immune system. Then there is a bunch of other immune cells, different types ... you can imagine like B cells, T cells, we've learned about this during the pandemic ... and these are the cells whose job is to identify what danger signal it is and generate a corresponding response. That's it. There are different components that talk to each other, and they have very specialized functions that work together.

Russ Altman (05:01): I know that your application of these ideas to disease is very broad and very interesting, and I'm hoping to get into a lot of it. I know that one of the first things that you did ... and I think many parents can relate to this ... is you did work trying to tell the difference between a viral infection and a bacterial infection. As you know very well and many people know, bacteria are the ones that will respond to antibiotics, whereas viruses, until recently, we really didn't have any treatments. Now we have some antivirals, but even for physicians it can be very difficult to tell is this patient infected by a bacteria or a virus, and yet that has a huge difference in how they're going to manage it.

Russ Altman (05:40): The reason I mentioned parents is when your kid has an ear infection and you take them to the doctor, the thing that the doctor's trying to figure out is does this child need antibiotics, because this looks bacterial, or is this going to be a virus, it'll be self-limited, it'll get better. Using that as an example, tell me how. In the old days, I know we would just try to culture the bacteria out of the ear and say what's growing. You have a different approach. Can you take us through that?

Prof. Khatri (06:06): For differentiating bacteria and viruses, right?

Russ Altman (06:09): Yes.

Prof. Khatri (06:13): I'm simplifying here dramatically.

Russ Altman (06:17): That's what we love. That's what we love.

Prof. Khatri (06:18): There might be some inaccuracies, but they are purposeful inaccuracies. Imagine bacteria and virus. Usually, bacteria are not inside the cell. Usually they're outside the cell, and viruses usually go inside the cell. The reason for this, evolutionarily speaking, is a bacteria has all the machinery it needs to replicate itself, whereas a virus doesn't. It uses the host machinery, our cell replication machinery, to replicate itself. This is a gross simplification, but let's put that on the side.

Prof. Khatri (06:55): The immune system, over the evolution, learned that there are two types of danger signal, one that is outside the cell and one that is inside the cell. Depending on whether it's inside or outside, it learned what host response to generate to get rid of that danger. We analyzed a large amount of data where we had different bacterial and viral infections, and we figured out what is it that the immune system did not change during the evolution, that no matter what bacteria or virus it was, it always generated that response. That's what we are now starting to read.

Russ Altman (07:43): I see. Let me ask, how hard is it to distinguish these two? Now that you've done a bunch of work in this area, are there indeed signals that are very reliable?

Prof. Khatri (07:53): They are extremely reliable. We've now shown that what country or continent you were born and raised in, what your race and ethnic background is, what strain of a pathogen you are infected with, whether you have some comorbidities or not like the obesity, autoimmune diseases, immunosuppression, irrespective of these ... or you have even both bacterial and viral infections together, because that also happens ... you can always see that there are these conserved immune responses to these types of pathogen, and you can target them.

Russ Altman (08:34): This is very exciting. Let me ask the obvious question, is how close is this to routine use, A, and are doctors excited about this or does this disrupt them in some way?

Prof. Khatri (08:47): I'll answer the second question. Doctors are definitely excited. I've heard repeatedly, and I still hear it, that they can't wait to try this out in their clinics, so that's really exciting. That wasn't the case about seven, eight years ago, but now it is definitely. The reason that they're excited is because we are, I want to say, less than a year into this actually becoming clinically available, with a point-of-care machine that you could use as a single blood draw, less than 40-minute turnaround time from sample to answer.

Russ Altman (09:28): That's very exciting, and that will happen definitely even in the lifetime of those of us who are getting older rapidly. Okay, so that's great. I think I know the answer. I believe you had to combine both development of devices, as you just described, and also computational analysis. Tell me a little bit about the computational analysis. Is this using a fancy AI? Is this old-fashioned analysis? How do these algorithms work that try to pick up these signals?

Prof. Khatri (10:03): It's not fancy AI. To be really honest, the stuff that we are using now was developed sometime in the mid '70s. There are two differences. One, we have so much data available that we now make better use of the methods that were developed years ago, decades ago. Then the second thing is we changed a few things in this 50-year-old algorithm, that what was the fundamental guideline for using it, we are ignoring it. Instead of making sure that the data sets are comparable to each other, we make sure that they are as disjointed from each other as possible.

Prof. Khatri (10:46): What that has allowed us to do, and the reason for doing that, was we wanted to find a solution that is not biased. In today's AI, that's a continuous discussion, of avoiding biases. If you want to truly help patients, you need to make sure that there are no hidden biases, and your solution has to be broadly applicable to large number of patient populations across the globe.

Prof. Khatri (11:17): What we've been doing, the methods that we developed, are to account for these biases implicitly. We are not accounting for them explicitly, but implicitly we are making sure that there is biological, clinical and technical heterogeneity that we can represent in our data, and that allows us to make sure that when we go to a real-world patient population, be it in Africa or Europe or Asia or North America, it's the single solution that continues to work across the globe.

Russ Altman (11:58): That really does make sense. In fact, earlier in your comments you talked about all the different disease states that people might be in where it still works, all the different geographies where it still works. Clearly, your ability to make those statements comes from this approach. In fact, I know that on the internet I can find talks that you've given about dirty data, and I'm suspecting that this is what you mean. Not that it's bad data, but that by having it be real-world and from multiple sources, you can really check to make sure that this technology that is promising in the lab remains promising when it's deployed into the real world.

Prof. Khatri (12:35): Exactly. Let's take a very simple example of any discovery that goes to a patient. The first time it is described in a paper through lab experiments or whatever, we know that it does not immediately go to patients, because the first thing we get asked is, "Can you show that it works in another cohort?" What we are implicitly saying is, "I know your patient population was not representative of the real-world patient population. Show me that, if you were to now go out and do the same thing in a different patient population or a more heterogeneous patient population, it would still work."

Prof. Khatri (13:11): That's where most things die. We wanted to avoid that pitfall. We wanted to make sure that our a priori odds of success in the real world were better than 70, 80%. That's why, instead of starting with a homogeneous, one cohort, we decided to start with a heterogeneous, multiple cohort. You asked me earlier about AI. As we ask more sophisticated questions using this, we do realize that we are now starting to include more neural network-based approaches, because now from the same set of variables, we are trying to ask multiple different questions. I'll give you an example.

Prof. Khatri (13:58): One of the things that we started out with was just trying to differentiate whether the patient has a bacterial infection or a viral infection. There is even a bigger problem, which is many times we don't even know whether the patient is infected or not before we can ask a question, is this a bacterial or a viral infection. Now we need to decide whether there is a presence of infection. If yes, what's the type of the infection? Then the third question that comes right after that is, if you show up at 2:00 in the morning in an emergency department, it's should I let this patient go home, or it's going to turn into sepsis and I need to admit them to ICU? There are three questions that we need to answer, so presence, type, and severity of infection. We need to be able to do it in under an hour, let's say, but much earlier if we could.

Prof. Khatri (14:47): Now, with the small number of variables that we have identified that generalize the rest of the entire world's population, we need to come up with better classification models. That's where we are starting to use multilayer perceptrons, support vector machines, neural networks, depending on the question, so that, using a very small number of variables that you can measure in under an hour, you can answer multiple different questions using a single test.

Russ Altman (15:19): I'm very interested in your third category. They're all very interesting, but that third category of severity gets to the next topic that I wanted to talk about, which is now you're not just diagnosing a disease, but you're also looking into the future a little bit. I wanted to ask the degree to which that immune system that you're studying is giving you good indicators not only of what's happening now, but what's going to happen. It sounds like it is giving signal.

Prof. Khatri (15:46): Exactly, and this goes back to the immune system being a distributed sensor with a built-in amplifier. Now the immune system, you can imagine, is the trigger for the symptoms that you are going to see an hour later, a day later or a week later, depending on the kind of disease you have. If you are reading the immune system correctly, you are getting an earlier indication of what's going to happen, and that's where the advantage comes in. I'll give you a very specific example. Then the other thing is you can't fool the immune system, because the immune system always tells you, "Well, I've been looking at this for a few million years." Here is a specific example.

Prof. Khatri (16:28): We were looking at this bacterial/viral infection diagnosis, and we came across a study where there were healthy volunteers, more like younger than me, 20- to 30-year-olds, who were asked to inhale live influenza virus and then go in quarantine for seven days, and they would take a blood sample every eight hours, do a nasal RT-PCR to see whether they were infected or not. Usually, about 50% of the people get infected if you are exposed to live virus. We could show that our diagnostic signature was actually identifying who has influenza infection about 24 hours before their symptoms would show up, on average.

Prof. Khatri (17:17): What was really remarkable was there were two people, one our diagnostic said is not infected but showed symptoms every single day, and another that showed no symptoms but the immune system said you are infected. We asked the original study author why, what's going on here, and they said the one that showed no symptom but you said is infected, that was an asymptomatically infected person. She was shedding virus every day, didn't show a symptom, but the immune system was actually saying, "There is a danger signal. I'm responding to it." Then the other one is what we would call a hypochondriac. Never shed a virus, never had any infection, but showed every single symptom because he thought he actually had inhaled the live virus. The immune system was actually picking up what's going on.

Russ Altman (18:04): Don't argue with an immune system that has been developing for millions of years. Well, this is The Future of Everything with Russ Altman. More with Purvesh Khatri next. Welcome back to The Future of Everything. I'm Russ Altman, and I'm speaking with Professor Purvesh Khatri from Stanford University.

Russ Altman (18:19): In the last segment, Purvesh described how he can use the immune system to understand whether a bacterial infection is happening or a viral infection. He's basically listening in to the immune system, and letting it tell him what's going on. In this segment, he'll tell us about exciting new advances in the detection of tuberculosis, which is the number-one all-time killer of humans by infectious disease. He'll also tell us how signals from the immune system give us clues about how best to treat patients using precision medicine.

Russ Altman (18:54): Purvesh, I know you've been working on tuberculosis, and some of it is progressing to the clinic pretty rapidly. What are the challenges with tuberculosis? Many of us hear about it. We all get our little test at the doctor every few years. Sometimes we have to take medications. Where are we with tuberculosis, and what's the promise here?

Prof. Khatri (19:13): Tuberculosis, just to set the stage, it has killed 1 billion people in the history of mankind. That is every pandemic in known history combined, plus all the world wars combined, times two, right? Nobody, nothing, has killed as many people as tuberculosis. In the 21st century, as of 2020, we are still not able to diagnose 40% of the patients across the world, so that's the severity of the disease. It still kills. It used to be among the top 10. It is back in the top five reasons for mortality worldwide.

Prof. Khatri (19:55): WHO has set this goal for ending TB by 2035. What they've asked is they want a non-sputum-based test, because the current test requires that patient cough up a sputum and then culture it, which takes anywhere from one to three weeks. They want something that is fast and is as good as the current standard of care.

Russ Altman (20:23): It's also very unpleasant to get sputum up, and some patients just can't do it.

Prof. Khatri (20:27): Yes, especially children and patients with HIV coinfection. They can't even generate sputum. We set out to do this about 2016, so about six years ago, and we found that there is a three-gene signature in peripheral blood that would now diagnose whether you have active tuberculosis disease or not, compared to healthy controls, those who have latent infection, or any other lung disease. That was 2016.

Russ Altman (20:59): Wait a minute. When you say a three-gene signature, help decode. What does that mean to somebody who's not a scientist?

Prof. Khatri (21:08): Every cell has a certain number of genes that come from our genome. Depending on what immune response you have, different genes are expressed. Not all genes are present in a cell at every time. It depends on what condition you are in. These three genes are only present in immune cells when you actually have tuberculosis. That's what we found. Then a company in Sunnyvale licensed this from us, from Stanford, created a cartridge, and two years ago a group of scientists across two continents in four countries showed that they could now measure this three-gene signature using a fingerstick. Looks like Theranos, right, but what they showed is ...

Russ Altman (22:03): Yes, but it works.

Prof. Khatri (22:05): Yes, but they could show that in a truly prospective trial at point of care in clinic across four countries, this three-gene signature, measured on a cartridge, diagnosed TB in 45 minutes. Before the sputum left the clinic for culture, they knew what the patients had, and they could send them home with the treatment. The problem is getting it adopted broadly.

Prof. Khatri (22:40): As you said in the beginning, this is an entirely new kind of diagnostic tool. We haven't used this. We've been using host responses for almost 150 years. It started with lactate, but they've been protein-based biomarkers, not gene expression-based biomarkers. We haven't had many of those, or actually none, in the clinic. Raising awareness, showing that they are as good if not better than the current diagnostics that we have, that's the challenge that we are going to work on.

Russ Altman (23:15): Yeah. You raise an important point, because I know that lots of the tuberculosis is occurring in the developing world where they're very resource-constrained. Is there an effort to try to make this super robust? A lot of these tests have to be conducted in tents or temporary hospitals that also have other services. Is that one of the pushes of the work, is to try to figure out how to get this to be robust and cheap?

Prof. Khatri (23:43): Exactly. I would speak for the robust part. The cheap part is where, given that the technology itself is new and there isn't a market, one could argue that the cost would be high, but as it gets more adopted, production scales up, prices are going to come down dramatically. Twenty years ago, it took a few million dollars to sequence a genome, and now we are sequencing it at a hundred, two hundred dollars. The same thing's going to happen, but I think it's going to happen faster than the 20 years as the usage increases.

Russ Altman (24:19): Yes. That seems to be how things can go when you show initial that it works, then all of a sudden that creates a whole ecosystem for different people who are good at engineering and reduction of technology, to figure out how do we get this to cost pennies so we can deploy it broadly in places that really need it.

Prof. Khatri (24:38): Exactly. My goal has been to demonstrate repeatedly, over and over, in different patient populations in different countries, that this works, it is clinically useful, it can be translated in clinic, and then that would bring more people in and just move the field forward.

Russ Altman (24:57): Great. In the last few minutes, I know you were also doing work in precision medicine, that is to say predicting what drugs would be best for patients. Could you set that up? What's the challenge here, and how are you addressing it?

Prof. Khatri (25:10): Yeah, so there are two things we could imagine. I'll speak for one, and that is most drugs don't work across all patients. Only a subset of the patients respond. Given that I've been very interested in infectious diseases, that particular problem is very evident in sepsis. Just to put some numbers on the table, every sepsis patient in the State of California costs one day. One day in ICU is $19,000 a day, and that's the median price. It could be much higher. The problem has been, if you look in the literature, about 150 drugs have been tried and they've all failed. There is no single drug that works. There was Xigris that was approved and then post-marketing it failed, because it didn't work for everybody.

Russ Altman (26:11): This is for sepsis. This is very bad infections in the ICU?

Prof. Khatri (26:13): Right. It's about 45% mortality in ICU, is because of sepsis. What we have started to now learn is the immune system itself is very heterogeneous, and the responses within sepsis patients are also heterogeneous. As we were talking about earlier, the immune system can tell us what's going to happen to the patient at presentation. That basically helped us hypothesize that maybe there are different groups of patients within sepsis as a bucket, and that if we could identify these patients, then maybe we could start treating them better.

Prof. Khatri (26:56): What we've been now focused on is identifying subgroups of patients. We call them endotypes of patients, so endotyping sepsis patients into different buckets, understanding what the underlying immune response is that puts them in that endotype, and then identifying either existing drugs that could treat those patients or identifying drug targets that could be used to develop new drugs for those patients. Then the hope is, in the next 10 years or so, we would now change sepsis from mostly one-size-fits-all to understanding the endotypes and treating them accordingly.

Russ Altman (27:38): I'm fascinated by your answer, because never once did you mention the actual bacteria that the patients have. It shows me that there's been a shift at least in your thinking, that it's all about how the immune system is responding to this. I'm sure it matters what the bacteria is, and I would never say that that doesn't matter, but the comments that you just made highlighted that for you, it's the response of the body to the infection that might be the key to treatment, even more than what it is that's infecting the patient.

Prof. Khatri (28:09): Exactly. It's the host's inability to respond appropriately and sufficiently that allows the bacteria to keep growing, and the patient doesn't recover. I think of treatment as supporting the host in responding to the bacteria.

Russ Altman (28:27): Thanks to Purvesh Khatri. That was the future of infectious disease immunology. You have been listening to The Future of Everything with Russ Altman. You can follow me on Twitter @Rbaltman, and follow Stanford engineering @StanfordEng.