Question and Answer Sessions

At the close of some talks, questions were posed by audience members.  To find out about these questions, click on the hyperlinks below.

Limitations of Current Dose-Response Data and Models: Information Needs for Microbiological Risk Assessors 
Margaret Coleman

Use of Epidemiological Data in Dose-Response Models 
Kirk Smith, DVM, PhD

Enumeration of Salmonella enteritidis in an Outbreak Associated with Ice Cream 
Sita Tatini, PhD

Suitability of Small Human Clinical Studies to Measure Pathogenesis of Foodborne Pathogens 
David Tribble, MD

Why Relate Numbers of Foodborne Pathogens to Human Illness 
James D. Wilson, PhD

Open Discussion 
Dennis Lang, PhD

Limitations of Current Dose-Response Data and Models: Information Needs for Microbiological Risk Assessors

Margaret Coleman

DR. GAYLOR: Thank you, Peg. You certainly raised a lot of issues that we could spend the rest of the day discussing, whether it's thresholds or nonthresholds, linear, nonlinear, healthy versus susceptible, subpopulations and so on.

We have a couple of minutes here if anybody has a question they'd like to address now. Would you come to the mike and identify yourself?

MS. PUTZRATH: Resha Putzrath, Georgetown Risk Group.

Even though you just mentioned it in passing, Peg, since you did mention toxicity equivalency factors I have to make a comment, since I've spent the last several years working on it, which is that I urge caution if you want to go in that direction. In particular, if there are a lot of restrictions on the model and limitations for its use, one quick caution is if you believe in thresholds and do dose-response modeling with thresholds in them, you will often find that the dose-response curves cross, thus changing not only the magnitude of the proportional difference but the rank order.

MS. COLEMAN: Thanks, Resha.

MR. POWELL: I have a question for both Peg and Jim.

Given the evidence, the analysis that Peg and Harry have done regarding the magnitude of inter-individual variability that can be hypothesized, some three or four orders of magnitude greater than is conventionally used to account for inter-individual variability in chemical safety assessment and risk assessment, do you think that this might hasten, delay or have no effect on the transition from safety assessment to predictive consequence risk analysis in this arena?

MS. COLEMAN: I guess my general comment is that, for example, my title is still chemist within my agency, even though I've not done chemistry in years, so I came into the paradigm from the chemical risk assessment field and I've tended to shy away from the safety factors and assuming tenfold is sufficient across the board.

I think I'll take that one step further and agree with Jim that we ought to take our time in evaluating the evidence and I think case by case because personally, my opinion is that we don't have enough knowledge to really set defaults that could apply across the board to all pathogens.

I guess one other comment is maybe the time is now. Mark has raised issues about concerns from the international front for our agency of risk assessment that the U.K. is developing on hormones in beef, and that's going to be falling into our lap soon.

So I think even though this wasn't a direct answer to Mark's question, I think we do need to explore it.

DR. WILSON: The unidentified questioner was Mark Powell of the Food Safety Inspection Service.

First of all, Mark, even though our mutual friend Adam Finkel talks about assuming a tenfold span of human susceptibility, he's as ignorant as a post. In fact, when you expand the traditional process--the factor of 10 does not assume from the starting point as the median responsiveness or the median susceptibility of the population but something like 4 percent, okay?

So, in other words, the starting point of that safety factor is not the average susceptibility but a no effect or an observation where there is no observed effect, which is, at most, according to the work of our friend Dr. Gaylor, something around 4 or 5 percent in a typical experiment, okay?

So in the standard practice the inferred distribution of susceptibility is A, log normal or approximately log normal and B, and I'm not sure how you characterize it, say the difference between the first and the 99th percentiles is somewhere between three and four orders of magnitude.

Now, if you do back-of-the-envelope with a soft pencil eyeballing of the data that Peg presented, the span in the microbes, the susceptibility span is probably wider. That is to say, those response curves are shallower.

What was the other part of the question? Oh, the other thing you asked about is will we transition from safety assessment to risk assessment. My observation is that in microbial practice, they're not to safety assessment yet, let alone risk assessment.

Use of Epidemiological Data in Dose-Response Models

Kirk Smith, DMV, PhD

DR. SMITH: Do we have questions? 

DR. WILSON: Jim Wilson. I got the impression from what you were saying that the various chocolate samples were rather heterogeneous, that there was not, at least in the one that I heard you characterize in a little more detail, there were not organisms recovered from all the samples and there was only a small fraction of those from which organisms were recovered where the number could be quantified.

If that is characteristic, that suggests that A, you're sampling from the tail of the exposure median. You know you're sampling from the tail of the distribution of people affected because, of course, you don't ever find those who are not affected. And that makes me worry a little bit about using the means or even the medians of those numbers to draw conclusions about how many organisms were required to cause the infection.

I mean, there's certainly some reasonable probability that all those affected got a lot more because they got the hot candies.

DR. SMITH: Maybe somebody could speak to the distribution of Salmonella contamination in chocolate. There's probably somebody here that knows a lot more about it than I do. In the one outbreak I know they did look at 100 10-gram chocolate balls, and six were contaminated.

DR. WILSON: Have you estimated the number of subjects that would be required to get a reliable estimate of an ID-1?

DR. SMITH: I know it's a whole lot. 

DR. WILSON: You can imagine doing a volunteer trial with 1,000 volunteers.

MS. SILBERMAN: I'm Jennifer Silberman with BNA. I wanted to know how do Salmonella strains get into cheddar cheese and chocolate in the first place?

DR. SMITH: I don't so much know the chocolate. I can speak to the cheddar cheese. The cheddar cheese outbreaks are usually due to some defect in pasteurization of the milk that goes into the cheese.

The chocolate, I think, is somehow contaminated at its source, the cocoa beans.

MS. COLEMAN: I'm Peg Coleman, USDA.

I haven't thought very much about cheese and if you have a one-pound block of cheese, are you likely to have organisms all throughout or are they mostly on the surface?

DR. SMITH: I Think I'm going to have to defer to Dr. Tatini on that. I think probably they're not all on the surface, that they're probably throughout--

DR. TATINI: [Inaudible.]

DR. SMITH: So there is heterogeneous distribution, and organisms are not just on the surface.

QUESTIONER: You are discussing other host factors. Does that mean that the data simply are not available yet?

DR. SMITH: Yes, I think that is the case. Usually we'll get sex and age data and I think predisposing conditions. Antibiotic use is certainly being looked at now but really hasn't in the past, and exactly how much people ate. I think those kinds of detailed questions really aren’t being routinely asked. You investigate, find out what the product is and get it pulled.

But I think certainly we're headed in that direction and can do a lot more of that.

DR. LEVINE: Kirk, both in the volunteer studies and in epidemiologic investigations where you go into an outbreak situation and maybe culturing contacts who are not clinically ill, one could be talking about infectious dose, meaning including subclinical infection, or clinical infectious disease dose.

In your last bullet, what are you referring to? Clinical ID or literally infectious dose?

DR. SMITH: I would think maybe a clinical ID is more useful for our purposes but again that's going to take even more people, obviously.

Enumeration of Salmonella enteritidis in an Outbreak Associated with Ice Cream

Sita Tatini, PhD

QUESTIONER: Does clustering affect virulence?

DR. TATINI: Well, I can't answer that question. But if there is a cluster and you don't know it, you think you're introducing one organism and that's going to affect the severity of the illness, would it not? If you think you've had one cell but indeed it is a cluster of 100,000 organisms. So I'm not sure I can answer your question.

I don't think the virulence would be affected. Virulence is a property of a single organism.

Suitability of Small Human Clinical Studies to Measure Pathogenesis of Foodborne Pathogens

David Tribble, MD

QUESTIONER: [Inaudible.] What kind of factors determine dose selection?

DR. TRIBBLE: I'll try to repeat. Let me just make sure I have the correct question. Your question was what were we using in our study to determine the dose selection in the Campylobacter study?

We primarily used the pre-existing information from the CVD study. In this study there was a lack of a illness dose-response at doses up to 109 inoculum in the same strain that we were using. The CVD investigators observed around 40 to 50 percent overall attack rate for diarrhea or fever.

That was the reason we chose as high a dose as 109 as one of the initial doses. The reason we chose the 105 dose as our lowest dose was based partly on the results from the C. jejuni A3249 strain. With this strain, a 46% attack rate was demonstrated in the 104 dose however, the attack rates were very inconsistent across the dose range. Using the bicarbonate buffer delivery modification, it was conceivable that a more consistent attack rate could be demonstrated at this lower dose. Another modification from the CVD study was the use of screening Campylobacter serology with enrollment of only seronegative volunteers. In the CVD study, retrospective analysis of baseline serology detected a decreased risk of illness post-inoculation in volunteers with elevated baseline serology. So those were the reasons we used that range. Limited number of inpatient beds, ward time availability, and limited number of volunteers were also considerations. Logistical concerns do weigh heavily in the design of these trials. 

Why Relate Numbers of Foodborne Pathogens to Human Illness

James D. Wilson, PhD

DR. GAYLOR: So now I'll ask if there are any burning questions. I don't want a long discussion but does somebody have a short question or two for Jim?

I guess we're going to stay on schedule.

Thank you, Jim. Actually I agree with you. I'm a born again statistician. I've spent my lifetime doing dose-response modeling and quantitative risk estimates but I believe we're moving away from quantification and certainly for a regulatory standpoint, just trying to establish what we think is a relatively safe dose and not trying to attach some number of 10-6 or 10-5 or whatever it is that we think we're within a factor of 100 or 1,000 of. So I certainly agree with your comments and you got us off to a great start here.

Open Discussion

Open Discussion

DR. LONG: Welcome back for our final discussion this afternoon. I will remind you one more time that you have these yellow cards in your folders and Dr. Bradlaw and myself will be sort of walking the aisles and looking for people who want to pass cards to us if you'd rather put a card forward than stand up and ask a question.

This afternoon's panel discussion will be moderated by Dr. Dennis Lang. Dr. Lang is the Enteric Disease Program officer at the National Institute of Allergy and Infectious Disease of NIH. He's charged with the development, coordination and direction of the extramural grants and contracts program in enteric diseases, which includes the development of vaccines and therapeutic agents against enteric pathogens.

Dennis is a member of the Risk Assessment Consortium and a member of the Dose Response Work Group of the Risk Assessment Consortium and he has helped us out a lot and he's helping us out again this afternoon. Thanks, Dennis.

DR. LANG: I thought I would introduce this session by reviewing for myself, as well as yourselves, how I got involved in this process. I was asked to participate in the Risk Assessment Consortium as an NIH representative by virtue of the fact that I direct the Enteric Diseases program at NIAID. The organisms studied by our investigators and supported by NIAID are the ones that are of most concern in foodborne disease.

When I went to the first meeting I sat around the table with statisticians and epidemiologists and wondered what I was doing there. It quickly became apparent during discussions that there was a real lack of information to estimate the risk of the human population to foodborne disease.

The research that NIH supports is both basic and clinical research on the genetics, microbiology, pathogenesis, and virulence of these organisms and ways to control them. Mike Levine has been one of the pioneers in developing human challenge studies for the purpose of studying microbial pathogenesis and for measuring vaccine efficacy. Dr. Tribble alluded to a lot of the work that Mike pioneered, and to which others have contributed.

In performing such challenge studies we seldom work at low doses. The goal of such studies is to approach 100% attack rates in a small number of volunteers. We do not typically study a small number of organisms where it might take 1,000 people to be exposed in order to see an illness.

There is a real disconnect, in a way, between the real life human experience and available human dosing data. These questions are going to become important in defining the human risk of foodborne diseases which, for the most part, result from low dose exposures.

As part of the recommendations from the RAC committee, the FDA has published a request for applications calling for research to try to make the correlation between human exposure at low doses and infection and disease. Parallel work will also be done in animal models, with the goal of developing correlative dose-response curves for some enteric organisms.

They have received applications and I'm not going to say any more about them because that process is still on going. But suffice it to say, that there is going to be an investment by FDA in trying to make some of these extrapolations and measurements.

Therefore, the question to the panelists and to the people in the audience is what kind of recommendations can we make regarding the design of those studies? Which organisms should we study? What animal models are the most appropriate to use for comparison to the human data that we get? What kind of human data should we be looking for?

If we're not going to be seeing full-blown diarrheal disease, are there other end points such as shedding or immune correlates that would be a more sensitive assay that would be quantifiable and meaningful?

These are questions I'm almost certain we are not going to answer today. Part of the reason for having this meeting was to start a dialogue and get the people who, like Dr. Slauch, study pathogenesis in animals to talk with the people who are doing human challenge studies and come up with recommendations that are compatible, with an approach that makes sense.

I will just toss that out first to the panelists, whoever would like to comment about what they see as of common themes that have emerged after today's presentations about pathogenesis, and where we should be going.

Would anyone like to volunteer? Peg, why don't you start? Peg is our risk assessment person and probably as familiar with the trials and tribulations and limitations of risk assessment models.

MS. COLEMAN: Thank you, Dennis.

I'm actually very excited to see this kind of interest in dose-response modeling and to be hearing from the people who can generate data and help us draw inferences that really are plausible in our modeling.

So I agree, Dennis, even if we haven't generated a list of solutions today, we have started the dialogue. It has been helpful for me to hear from some of the basic researchers about what kinds of interests and their own research work might bear on dose-response modeling.

DR. LANG: I have a question, actually, that was generated in the audience that I'd like some response from the panelists. Is threshold dose determination useful for regulation, for regulatory agencies? Let us assume that threshold in this case is a disease-causing threshold and not a threshold of infection.

DR. WILSON: Let me ask, does this refer to some--is there some specific technique that leads to something called a threshold dose? Does that refer to some--is that a term of art in the microbiological business?

DR. LONG: I don't think so and maybe you should define threshold dose for us, for the audience and for the panel, as you interpret it.

DR. WILSON: Mathematically it's a discontinuity in the dose-response curve at the point at which the response as a function of exposure goes from zero to some value greater than zero. I don't think that defines dose in any way.

Within toxicology we talk about no observed effect level of no observed adverse effect doses, but as Dr. Gaylor would tell me if I didn't say this, all that means is that the response is less than the observational uncertainty.

So those two concepts of threshold and dose don't necessarily go together, as I understand them.

DR. LANG: Perhaps the person who asked the question might like to elaborate.

DR. GAYLOR: Actually I asked the question. I heard a lot of discussion today about what's the minimum number of organisms that'll cause illness or infection or maybe illness, and that hasn't served us very well in chemical safety assessment--how many micrograms or milligrams will cause an adverse effect and below that we're okay.

This is pretty illusive and it's coming up with a noninfective dose or a dose that doesn't cause illness--illness to whom and when and under what conditions?

I think it's worthwhile knowing whether 1,000 organisms or 100 organisms or 10 organisms, sort of an order of magnitude. I think that's worth knowing but whether three, four or five E. coli in hamburger are safe or dangerous, I don't think it's worth our effort trying to answer that question. Even if we figure out that three are safe for everybody, say; what about the person that's already exposed to three coli from eating lettuce? Where does that leave us with hamburger?

So there's a lot of regulatory problems even if we could answer the scientific question.

MS. COLEMAN: Can I add a little bit to that discussion? I agree with you. In fact, your language really implies that there is a distribution of thresholds for the population and there's not just one threshold, that each individual may have a threshold.

But the sense in which we've used threshold in our work was how many bacteria does it really take to cause symptoms? If you have one cell infected in your GI tract, will that cause you to show gastroenteritis? Or is it an effect of an accumulation of damage to cells that is necessary to cause symptoms?

So that's the sense that we've used it. And the example of a threshold of four was just meant to illustrate that even low thresholds can have tremendous effects in dose-response modeling.

So we weren't trying to presume that four cells were safe and three were not but that assuming that one cell can cause illness is quite an assumption. So that's the sense that I was raising that point. Should we really look a little closer at the mechanisms of pathogenicity to address that kind of issue?

DR. LANG: Any other comments on that question from the audience?

DR. TATINI: I would like to comment something on that. If theoretically one cell has the potential to cause infection and it's a question of whether this single cell present in the food can reach the target site within the intestinal tract, for example, maybe all problems in the threshold are related to how many should be there in the food such that at each of these barriers, as it goes through, how many of them survive? And if one goes through, that would be adequate if it has the potential to establish in the gut and cause the illness.

So maybe the threshold is related to what happens to these organisms in food. Can they reach the targets? If one reaches the [inaudible] illness, depending upon what [inaudible] the other conditions that exist.

DR. LANG: Another question. Where are the data describing the susceptibility of the human population with respect to the genetic components and their acquired resistance? There is probably very little data available in terms of the human genetic components involved in any of this, but perhaps with the human genome project, in another 10 years we'll be in a better position to assess that component.

But I'll ask Mike Levine to respond initially to that question. Mike?

DR. LEVINE: I brought a couple of slides--it's old technology apparently but we'll give it a try--that addresses this question. It addresses it in the following way.

Let's consider vibrio cholerae as a paradigm of looking at an organism that's incriminated as causing diarrheal disease in humans and let's just ask the questions of what's involved on the host side and on the bacteria side in leading to clinical response.

Let me give as a bit of background there was a guy named Robert Koch a bit more than a century ago who discovered vibrio cholerae in Egypt in association with a large outbreak of cholera, which he called the vibrio comma. He received great publicity with this observation.

There was a more senior individual in Germany at the same time named Max Pettenkoffer and he looked upon the publication of Koch and he decided that in his view, this organism by itself couldn't explain the epidemiologic and clinical features. There was something missing. This is the famous XYZ theory of Pettenkoffer.

And to make a long story short, Pettenkoffer, who was quite an elderly gentleman at the time, got into this very, very acrimonious debate with the younger Koch and sought to prove that Koch was wrong by drinking a pure culture of vibrio cholerae. He did it and he didn't get sick, which raised a few questions. And I'll show you some data that, looking back a century later, we now know why that was so.

About two decades ago we were asked by the NIH to set up a model of El Tor vibrio cholerae 01 biotype El Tor to demonstrate or investigate whether that biotype was a cause of diarrheal disease.

It had been demonstrated some years earlier that if you took vibrio cholerae 01, the classical biotype, and gave it to volunteers without buffer, just gave it to fasting volunteers, if you gave a million organisms or 10 million organisms, nobody got infected and nobody got sick. You had to go up to 100 billion organisms and really enormous amounts to get a healthy, fasting young adult ill. But it was found that if you buffered the gastric acid, suddenly a 106 infectious dose was a 90 percent clinical attack rate dose.

So we looked at an El Tor strain called El Tor Naba N1691 and for the first time with buffer, carried out a dose response in which we went down, went down to 103 of the three. And even at doses of 104 and 103, the attack rate remained high--80 at 104, 67 percent at 103. But if you look in the far right column, the severity of illness in these small numbers went down as the dose went down.

These data were important because this was the first time that there was a link between volunteer studies and the epidemiologic data, which suggested that in the field in endemic areas like Mafla Bazaar in the Kamilla district of Bangladesh, the infecting dose of vibrio cholerae in nature is probably 102, maybe 103organisms. Next slide, please.

Now the next thing we did was to ask, with this dose bicarb that causes a 90 percent clinical infectious disease attack rate, what happens if we give the same inoculum with plain water--in this instance 300 milliliters of water in which the inoculum is suspended, given to a fasting volunteer--or if we give the same inoculum, 6 logs with food. The food is a sort of quasi-Bangladeshi meal of fish, rice, custard, a little skim milk.

And with buffer, 90 percent attack rate. With water, just water, no buffering, nobody got infected; nobody got sick.

Now, this is with distilled water--not distilled--with sterile water in which we put the inoculum. In nature there's a lot of incrimination epidemiologically of water but in those instances, sort of river water, for example, or tank water in Bangladesh, that water has zooplankton. It has entities to which vibrios can attach and it's possible that those zooplankton suspended in the water get the organisms through. But just water by itself will not do the job.

And a meal, food, gives you the same clinical attack rate and the same degree of severity as buffer. Next slide, please.

Now, the next slide, which is probably unreadable, just makes the point--it's a listing of a bunch of different strains. This is one of the most important points that I would make. We have tested, over the past two decades in volunteer model probably 10 different wild-type strains, maybe more. And what is quite interesting is that there is no in vitro assay, there's no animal model that predicts what will happen in humans.

And what you see is that some strains are red hot in volunteers, cause very severe illness. Other strains that do the same thing in rabbit RITARD model will cause moderate illness and others will be virtually nonpathogenic or will cause a mild illness that we would not clinically call cholera. Next slide, please. So that's some food for thought.

Selection of the strain going into volunteers is very important and I would say if you want to look, if the question is is such a genus and species a disease-causing pathogen, if you want to answer that for humans, you can't do that with a single strain. You have to look at multiple strains.

That was the pathogen side. Now we have the host side and here, this is also very important in terms of genetic susceptibilities and other susceptibilities.

We know, for example, that gastric acid very much determines severity of cholera. We also know today that blood group O is the single most important host factor that determines severity.

Nutrition status plays a role in developing areas. The more malnourished, the more the severe illness. And background immunity also plays a role.

So all of these, as asked of the question, all of these play a role. Next slide, please.

I want to show you some data with blood group to give an example. Actually, before getting to blood group, hypochlorhydria or low gastric acid can come from several different sources.

One of the things we found back in the 1970s when our volunteers were University of Maryland students, there were a lot of what we used to call grasshoppers. Dave Nalen, who was involved in these clinical studies, first came upon the possible association. In rounding with these volunteers he came up with the suggestion that it appeared that the heavy grass smokers--gave a history of heavy grass smoking--had more severe illness.

So over ensuing studies in a prospective way, he looked at that and in this Lancet publication clearly showed that severity of diarrhea in that model was related to the proclivity or the history of how much grass you smoked, which was, in turn, related to gastric acid because marijuana diminishes gastric acid secretion, and David showed that quantitatively. Next slide, please.

Let's go to a more biological host factor. This may not portray very well so I'll summarize what it shows.

In 1991 cholera came to the Western Hemisphere, returning after an absence of about a century. We knew by this time, from studies in Asia and in volunteers, that blood group O is a critical risk factor, making an individual much more prone to developing cholera gravis, severe cholera.

The lowest prevalence of blood group O in the world, as you might guess, is in the ancestral home of cholera, in Bangladesh, lowest prevalence in the world. The highest prevalence of blood group O in the world, unfortunately for 1991, is on the west coast of South America, where the native South American Indians have a blood group O prevalence of 85 or 90 percent.

So the cholera appeared to be quite severe, appeared to spread and we took some of those strains early on and fed those to volunteers, North American volunteers, and asked the question of whether there was a relationship in North Americans between blood group O and severity.

And in these studies, which summarize results with a couple of different strains from South America, the attack rate in persons of blood group O was 93 percent; the attack rate in non-O was 44 percent. The mean diarrheal stool volume in the blood group O was 5.23 liters, more than five liters. Five liters is an adult human blood volume. This is cholera gravis.

The mean stool volume in the non-O was less than two liters. The number of individuals--of the 15 individuals who were challenged who are blood group O, one-third of them developed, by our definition, cholera gravis, a five-liter purge, and of the non-O it was zero of nine.

So in the North American volunteer, as well, we see the genetic susceptibility playing a critical role in the clinical response. I just thought I'd show some of these data as a response to whoever asked that question and I'd be happy to expand with respect to any other organism they might be interested in.

DR. WILSON: Let me follow up with that and perhaps if people don't know, perhaps you could speculate.

It looked as though the distribution of susceptibilities, at least as measured by severity, was not log normal, was not a simple distribution. It was at