Predicting human dose-response relationships from multiple biological models: Discussion Transcript
September 28, 2000
USDA Center at Riverside
Riverdale, Maryland
Introduction | Agenda | Speaker list and presentations | Meeting handout PDF
Summary of 4 panel questions and conclusions:
1) How useful are different biological models as a source
of data for modeling human dose-response relationships?
Are there biological models or sources of data that are not useful?
- All models are useful; different models for different endpoints.
- No unsuccessful models for C. parvum.
- Model usefulness is easier to assess for microbial exposure than chemical exposure.
- The usefulness of healthy volunteer studies is limited without correlated studies in susceptible animal subpopulations.
- Human dose-response trials do not simulate the randomness of real-world exposures; computer models can add this component.
- Outbreak exposure data, microbial ecology (food matrix) data, low-dose data, quality-controlled data.
- More data.
- The relative safety of C. parvum in human trials lessens its relevance to other organisms.
- C. parvum dose-response studies are more advanced than studies for other organisms, despite less knowledge of C. parvum molecular biology and genetics.
PANEL DISCUSSION TRANSCRIPT
MS. COLEMAN: I'll go ahead and
give just a brief introduction for the panelists who haven't spoken to
us already. Second from the left is Jack Colford. He received his
Ph.D. in epidemiology from the University of California at Berkeley, and
he's also a graduate of John Hopkins School of Medicine. He has
a joint appointment as an attending physician at UCSF San Francisco, VA
Medical Center, Infectious Diseases Clinic. Thank you for coming,
Jack.
And we also have with us Dr. Charles Haas, and
he's third from the right. Chuck received his B.S. and M. S. at
Illinois Institute of Technology and his Ph.D. from the University of
Illinois at Urbana-Champaign. He's currently the L.D. Betz Professor
of Environmental Engineering at Drexel University, and he's been one
of
the founders, really, of dose-response modeling for microbial pathogens,
and we welcome Chuck to our panel also.
And sitting to Chuck's right is Mary Alice Smith.
Mary Alice received her B.S. from Auburn University, an M.A.T. and an
M.S. from Emory University, and her Ph.D. from the University of Arkansas.
Mary Alice is Associate Professor in the Department of Environmental of
Health Sciences at the University of Georgia. So thank you to all
the panelists and speakers for joining us today.
In our conference call, we had asked the panelists
if they might be willing to take five minutes or so to give us their impressions
of the talks, the technical presentations that we just heard this afternoon.
So would one of the three of you like to begin? Any volunteers? Chuck
volunteers, great.
DR. HAAS: I just want to focus on one issue,
and that is what is the purpose of a biological model? And I've
just bolded three items, and I think it's important for us to keep in
mind what we're asking a model to tell us. It might be one purpose
of a biological model to ask of it to generate an entire dose-response
curve that we could then use in and of itself to project human risk.
Another possible use would be--and I'll adapt
a term, I guess, that you came up with, Peg--to "anchor" a dose-response
curve that may have been generated in a human study, but to anchor it,
to shift it or somehow or other modify it to correct for factors that
were not subject to tests in the human study, such as difference in host
characteristics, difference in vehicle, difference in strain.
And the third possibly is to explore mechanistic
features of the biology of the pathogenesis of the microbe host interaction.
And I think we've seen some examples throughout the day of biological
models put forth, or being used for any one or mixtures of all three of
those purposes. And the only point I want to leave it at is to
say that while a biological model may be used for one purpose, that doesn't
necessarily mean that it's useful for all purposes.
And so I think we need to keep the purpose and
the model separate in mind, with respect to what we're asking it to provide
us with in terms of information.
MS. COLEMAN: Mary Alice?
DR. SMITH: Thank you. I first of
all want to congratulate all the speakers today, and the organizers of
this, because I think everyone brought excellent points to the table about
risk assessment, particularly for microbial pathogens. I have a
few points that I just wanted to make about what I think are very important
considerations in microbial risk assessment. They were all, I think,
illustrated today in one way or another.
First of all, I think quality control is an important
issue in any kind of risk assessment, and it's particularly true in microbial
risk assessment. And I think there were several times it was mentioned
today and some very good illustrations of where we have to have some good
quality control. But I think that it has not been addressed in sort
of an organized way yet, and I think we're going to have to do that if
we're going to really compare between laboratories, if we're going to
compare between serotypes or genotypes or whatever we're looking at in
terms of the organism, and maybe even more importantly, if we're going
to compare between pathogens. So I think quality control, quality
assurance issues still have to be addressed.
The second thing, and I think Peg's demonstration
did a really nice job of showing uncertainty. I think we have to
make sure that whatever the uncertainty is in our risk assessments, that
we're very clear about stating those uncertainties and making sure that
people understand that risk assessment is not just a formula. You
know, that there are uncertainties that go with those numbers and they
have to be expressed.
The other thing I think that's really important
is the risk assessments need to be predictive. And here, what I'm
referring to is, so many times when we do experiments, whether they're
human or whether they're animal or whether they're in vitro, it's very
difficult to get a handle on those low-dose exposures and, in fact, that's
what we have to have when we're talking about doing risk assessments for
humans. So not only are we asking our dose-response models to be
predictive in the range where we have data, we need them to be predictive
in the range where we have no data, and that's a really tough thing.
I think the other points I have are questions
that we're going to be addressing, but those were three that I think are
very important in terms of microbial risk assessment in general.
DR. COLFORD: I made a number of notes today,
too, same kind of themes that went through a number of the speakers.
These are in no particular order, but the first is this, from a clinicians
point of view, the very important distinction between infection and disease,
or whether or not you care about that in your response. Some people
made that explicit, some people didn't in their presentations.
Echoing the comments just made about point estimates versus confidence
intervals. For instance, in the model I think that you just presented,
Peg, I couldn't tell whether the uncertainty was also taking into account
the uncertainty in each of the point estimates or just the uncertainty
of the overall model.
MS. COLEMAN: Both.
DR. COLFORD: Okay. That's good.
Another issue is immunocompetent subjects versus immunocompromised subjects,
and when we have designed some epidemiological studies with the CDC, for
example, it was felt that by doing it, doing a study in an immunocompromised
population, if no effect, for example, was found there, then perhaps one
could declare the immunocompetent population safe, as well. But
I think making explicit the population being studied is critical.
Another issue. I'm not a risk assessment
expert by any means. And I think one of the things I struggle with
is how to assess different risk models and compare them even to each other.
With two mathematical models, how do we evaluate which is a better model
when there's little empirical data with which to conduct that evaluation? Are
there statistical means by which we can make that sort of comparison.
And then finally, I think there needs to be more
dialogue between epidemiologists and risk assessment modelers in both
sharing available data and the scientific thinking and the approaches
that are used.
MS. COLEMAN: Thank you. In fact,
I think we'll move right to the first question. So our first question--and
the procedure we'd like to use for these six questions, if we can actually
get through all of them before the next hour is up--we'll have any of
the panelists and invited speakers who'd like to address a question off
of their comments, and then immediately following that, anyone from the
audience who would like to chime in is welcome to do so.
So our first question: Are there biological models
or sources of data that are not useful for modeling the human dose-response
relationships? Anyone on the panel like to address that question?
And I guess I should tell you in advance that we have discussed all these
questions, so the panelists have been thinking about this, and this might
be the easiest question to answer. So who'd like to tackle it.
DR. CHAPPELL: I'll tell you from my point
of view, not being a risk assessment modeler, when I see a question like
that, it bothers me a little bit, because saying something is not useful
doesn't really take into account the kinds of points that Doctor Haas
has made. Different models are going to be useful for different
points. You have different endpoints, you're measuring different
things. So I don't think I would be willing at this point to say
that any model would be not useful.
However, I do think that there are some that
will be less useful than others. When Doctor Smith, Doctor Haas
and I were at another workshop, a risk assessment workshop in the Netherlands
this summer, an interesting concept was shared with me that had four different
diamonds. One above the other and two on the sides. And the
top one was human studies. And one on each side were animal in
vitro studies or animal in vivo studies or human in vitro studies.
And you have to decide which, for the particular question that you are
asking, is sufficient or is more accurate or better to address that.
The diamond at the bottom that was the farthest removed was animal in
vitro studies. And so, because you have a lot of correlations that
you have to be concerned about. And so as you move away from in
vivo to in vitro in humans into animals, you're always going to encounter
some problems and have to really define the differences between the two
models. So I think most of our issues are with those side diamonds.
Is it going to be better to look at the various endpoints in a tissue
culture model? And if you're looking for infectivity, that certainly
may be the best way to do it.
If you're looking for illness, tissue culture's
not going to tell you anything, or it's unlikely to tell you anything. Perhaps
at some point we will get so sophisticated that we can understand the
sequences and the polymorphisms of all the virulence genes that Cryptosporidium has, and when we have that, we'll have perhaps the perfect model of looking
from the laboratory and being able to predict what we see out in the population.
So I guess I would say that I wouldn't eliminate
any model but would be very careful in what models we use and how we
evaluate those models.
MS. COLEMAN: Thank you.
DR. TZIPORI: I have an issue with the tissue
culture, actually. I think it doesn't distinguish between viability
and infectivity. Am I splitting hairs here? Because there
is a difference there. You can have a viable, say, a sporozoite
is viable infectious organism, but you give that orally to an animal
and
it wouldn't make it, but it would infect tissue culture.
So sometimes you have an innoculum which would
be viable but not infectious, in the sense that it would not really infect
an animal. Tissue culture won't distinguish between that.
It really can tell you whether these are viable or not, but not necessarily
whether they're infectious or not. That's number one.
The second issue is with excystation. Sometimes
you grow tissue culture and you get an infection or you don't, and sometimes
you take oocysts that do not infect tissue culture but they will infect
an animal. Somehow, the process of digestion seems to facilitate
excystation when you get an infection in animals that you would not get
in tissue culture. When you're going to start taking samples from
the environment when you don't really know whether, a) they are viable
and, b) whether they are infectious, and you try then to compare tissue
culture with infectivity or viability with animals, there are going to
be differences which would be very difficult to interpret based on those
two issues. Viability and oocyst excystation. They just don't
work in the same way between the in vitro and in vivo.
MS. COLEMAN: Thank you. Other comments?
DR. DELEON: Probably a little bit in response
to Doctor Tzipori. I think it's very important to know what the
endpoint is and what is being measured and how it is actually being determined
whether you have infectivity or you don't have infectivity. Just
to illustrate, it's the difference between looking for messenger RNA after
an infection, versus looking for DNA after oocysts have been inoculated
in a monolayer. You do a DNA PCR, that is, in the absence of looking
for the RNA first. You get false positives.
You'd have oocysts that are just simply resting
on the monolayer that a direct PCR system is going to score as positive.
Whereas if you look for the RNA, that's not the same story. So
I think it's very critical in looking for these in vitro systems,
that the right assay is being conducted for the right application. And
they're not necessarily applicable to all cases.
But yet with an environmental sample, you're
not going to have the liberty for all environmental samples to go into
an animal model or a human volunteer model. It's just not going
to be practical. So it gives you some advantages and it has disadvantages,
and it's a question of using it where it is most appropriate to be used. I
envision it really more as a monitoring tool than anything else.
And for studies, for example, in comparative
disinfections, we're finding that it is very expensive to use animal models.
You couldn't even use, for example--I mean how expensive would it be to
conduct use in genotype 1 studies on the comparative disinfectants using
your piglets. I mean you'd run quite a bill. Whereas you could
probably get some fairly good responses to the different disinfectants
using cell culture without having to incur that kind of cost. There's
places where it's the right tool for the right type of job.
DR. TZIPORI: For the method that you're
describing, you don't need tissue culture. I mean you could run
a DNA assay or an RNA assay on the innoculum itself.
MS. COLEMAN: I think we can live with disagreement,
but really what we were trying to do with the panel was get you to offer
your opinions of this question, and we can work out the rest. Are
there others who would like to offer opinions? Anyone from the audience
want to tackle this one? Tom Oscar.
DR. OSCAR: Yeah, well, one of the things
that, you know, if you look through all the talks today, one of the pedestals
of the disease, frankly, was not being addressed in any of the models,
and that's the food matrix. I guess my question for the panel is,
in your opinion, are there any models that you can envision that can
be
developed that can address the food matrix question and/or modifications
of the current models to be able to address the whole issue of food matrix
and what effect it would have on the infection dose.
DR. SLIFKO: I have done a couple of different
experiments and studies with using different types of food. One
study, I was looking at water activity and the effects of water activity
on viability. We were suspending these organisms in apple juice
and bread dough, like a simulated bread dough and pancake syrup.
And at two different temperatures and looking at the survival over time. The
water activity did have a great effect, especially at the room temperatures.
But my point is that we used a very unusual type
of matrix to see the infectivity. Unfortunately, that was a long
time ago and I didn't have the method quantified at the time, so I don't
know. I just knew that we had a lot of infectivity or very little. And
so recently, we used apple juice and orange juice looking at ultra high
pressure effects on Cryptosporidium.
We didn't strain out all the pulp that was there,
because we wanted to have the most realistic type of circumstances as
could be, and then put it on subculture to see what the effects of this
pulp would have, especially with the antibody assay. And I had my
controls with it, the PVS, and I did not see any reduced infectivity.
And at the time, it was quantified. And as a matter of fact, the
pH enhanced the infection in some of the tests and that was based on
modifying
the buffers without the juice.
So it has been done for survival studies.
We haven't done it for monitoring. I assume that with some of the
other methods that are available, like INS and things like this, this
can be combined as a useful tool for monitoring, and subculture isn't,
like Rick said, isn't as expensive to run as animal assays. And
it's convenient. And it's a conservative model, no doubt, but it
tells you something about whether or not it's potentially infectious.
MS. COLEMAN: Chuck wants to address that
question, too.
DR. HAAS: If you look at any of the dose-response
models, they all contain parameters that can be interpreted as a survival
probability with the host, which encompasses the food matrix that the
organisms ingested. I think the way that the food matrix might
be approachable would be to start to do more detailed physiological investigations
of the factors that cause organisms to decay from the point of ingestion
until wherever they attach and colonize.
I'm not aware of any work on Cryptosporidium,
but in the Netherlands at the workshop that was referred to earlier, the
group at RIVM presented some very interesting data on E. coli 0157,
where they're actually starting to look at effects of pH in stomach resonance
time, for example, on the survivability of 0157 in animal model GI tracts.
And I think that's probably the road down which
we need to go, is to understand in detail what happens within an organism
to cause attenuation and then back that out into the parameters of the
dose-response models that we've been using.
MS. COLEMAN: Any other questions from the
audience? Ralph Kodell.
DR. KODELL: I think he had already pointed
it out but I'll simply state the obvious. I think the dose-response data
you showed right before this panel discussion were not very useful because
of all the 100 percent response to the top, as you pointed out.
Of course, that probably wasn't planned, and
I'm sure the experimenter expected to have a less of a dose-response,
none of those levels would have been used. But obviously, as I say,
to state the obvious, it would have been better to have those high-dose
subjects at lower doses, more of them down there. And of course
more subjects, if possible.
It might actually be better to drop off the top two of those, I
think, once you're at 100 percent for the second time because, in effect,
they do have some leverage. The leverage they have might be to make
the dose-response shallower than it ought to be. I would just say
those kind of data probably aren't very useful, though the intention
was
good, I'm sure.
MS. COLEMAN: Other questions? Or
are there comments from the panel on that question? Audrey?
DR. ICHIDA: I remember several years ago
in California as a graduate student, I heard a talk about a company that
was using a lettuce model. They were testing various microbes by growing
them on lettuce and then using this as a model for infection. It was done
specifically to decrease cost. So the idea was that you could sort
of do this dose-response. You'd use lettuce first and then save yourself
a lot of money because you'd have your starting points for mice.
I think in that situation, no one was making
any kind of argument that lettuce was going to replace mice. I mean
that was definitely not the thought. The thought was to get a ballpark
range using something that's way cheaper than mice, and then move into
an animal model where you really need to conserve more.
I'm thinking about your points that you were
making about trying to get things in tissue culture and then moving to
the more expensive animal model. Really to not lose sight that when you
see something in one model and another model, it's not necessarily a direct
contradiction. A lot of the time you have people thinking it's my
model against your model, whereas I don't think that's really the case. It's
what can we learn and how are the models different.
MS. COLEMAN: Can we move on to the next
question. How useful are different biological models as a source
of data for modeling human dose-response relationships? Anyone from
the panel like to tackle that one? Thanks, Mary Alice.
DR. SMITH: Yeah, this is not an easy question.
If it were easy, we wouldn't be here, I think. So I think that's
part of it. But I think biological models offer something.
You know, pretty much all biological models offer some piece of evidence
to us that we need for humans. Most of the time, we're not going
to be able to do dose-response data in humans that are fully applicable
to the situation.
And I guess if I can use Cynthia's cases.
Her study is excellent, and offers us so much, but the obvious next question
is, what would it be in the sensitive subpopulations? And that's
a question we won't be able to answer with a human volunteer study.
So I think the biological models offer a possibility for really answering
some of those questions. We certainly have to be careful how we
interpret the data, but we're sort of limited in having to go to other
models.
DR. CHAPPELL: I would fully agree with
you and pick up on what Doctor Tzipori said, that each model has its
own strengths and its own weaknesses, and as long as you define those and
you know what you're dealing with, then you should be able to get some
useful information from them.
DR. COLEMAN: As panel chair, I thought
I'd just put my two cents worth in, too. And I was going to go
back to what Chuck Haas told us, and that was that mechanistic modeling
could
really add a lot, even though we'll never have a human trial in susceptible
individuals, if we ask more targeted questions mechanistically, we might
be able to reason from other sources of data.
DR POSNICK: Just putting those two questions
in perspective. In all the work that's been done with Cryptosporidium,
have there been any models that you've tried to work with, models or animal
models, and in certain types of models with humans that have really not
proved to be useful, that have proved to be a blind alley, from which
data, data from those models just hasn't been very useful?
DR. HAAS: You know, I'm not aware of anything.
On the other hand, I suspect you have publication bias operating here.
You know, negative results tend not to be disseminated. I suspect
my experimental colleagues in this audience have certainly tried a number
of different approaches that never worked, and nobody, really, these
days
writes up a paper saying, well, we tried it in the system and it didn't
work.
MS. COLEMAN: Maybe Doctor Tzipori would
want to comment on that. I sense from Lauren's question, if you
can't use type 1 strains in certain classes of models, I mean obviously
no one's going to set up those experiments and run them. So this
sense of choosing a surrogate model when you can't use a human--is that
what you were getting at?
DR POSNICK: Well, I think that one of the
questions, one of the reasons that those questions arose was when we were
way back discussing this conference, was there are a lot of different
systems out there which may be basing dose-response modeling on.
Well, okay, you know, at least start by saying let's throw out some that
haven't been useful, which eventually you get your data from.
DR. TZIPORI: I think you've got to use
the best model, the cheapest, the one that represents as close to what
you are hoping to achieve. And again we come to the issues, exactly
the questions that are being asked. Back to your comment with regard
to type 1. If the pig is the only available model, then there isn't
much else even worth trying, unless somebody comes up with some sort of
conditioning that would make it possible to use the mouse. But
that's progress.
I mean when we started with the mouse, with rodents
back in the early 80s, the only rodents we were able to infect were neonates
until all the immunodeficient and immunosuppressed mice came about. And
that took, you know, some time to produce.
Of course, we threw all the ones that we used
to use earlier on that became less and less useful, and that's really
part of the progress. Hopefully, sometime in the near future, there
will be some other cheaper, more controlled models for type 1.
AUDIENCE PARTICIPANT: One of the big differences
between chemical risk assessment and microbial risk assessment is that
microbial risk assessment is, at least in theory, much more validatable.
It's very hard to validate chemical endpoints such as cancer and attribute
it back to a specific agent. But, at least in theory, we can do
that with microbial risk assessment, and so I think at least one measure
of utility that we've been discussing is whether we're able to predict
human disease.
And so, theoretically, this question may be answered
by our ability to develop particular models and then test them out against
actual observed disease rates. And having said that, the state
of our surveillance systems right now is pretty pathetic for that purpose,
and so I think, as an interagency group, while we're focusing on maybe
creating dose-response models, we should also be focusing on improving
our surveillance so that we can start testing and validating those dose-response
models.
MS. COLEMAN: Could I ask John to expand
on that? What do you think is the weakness of the current system?
AUDIENCE PARTICIPANT: Since I'm not an
expert on this, maybe Jack wants to do this and will amplify what I had
to say. I mean there's a lot of weakness, ranging from the structure
of our governing system that provides disincentives for providing specimens.
Chuck and I were talking about this in comparison to the U.K., to the
kind of dissemination of techniques for detecting cryptosporidiosis to
the amount of money I think that's going into collecting this data from
everyone today, to providing laboratories. I just don't think there's
that much of a structure out there to collect and place this data in
a
central location.
MS. COLEMAN: Okay.
DR. COLFORD: Yeah, I would like to second
that. Surveillance is very unsexy, and yet everyone wants those
numbers, so you go to every talk and there's a CDC slide up there that
everyone uses over and over and it's based on sort of one observation. In
California, what we finally came to in the Bay area, eight water utilities
have, after we got through the funding issues, worked out a way to support
a group we work with to do enhanced Cryptosporidium surveillance
in the Bay area through active visits to the labs and trying to really
get a good count.
And what can you do with data like that?
Well, what it tells us is that we have an upper bound of what's going
on in the Bay area, so even if foodborne disease was causing 100 percent
of the cases in the Bay area, the number wouldn't be much more than what
we're estimating here. Several of us in the room were at a recent
exposure workshop here in Washington, and it was interesting.
As each speaker spoke on their particular exposure,
whether it was fomites, water, food, once you added up the number of cases,
it turned out there were 17 million cases a day of Cryptosporidium going
on, because, you know, each person's exposure was the key issue. And
so parceling out all these attributable risks is really important and
a big issue.
DR. COLEMAN: Well, I'd like to throw in
another comment about the FoodNet in particular, and that is that for
listeriosis especially, we are very unlikely to ever know what food caused
any one particular illness. And again, apportioning out what cases
are attributable to what exposures, that is a real weakness, and I'm not
really sure how to get over that. But the idea of looking regionally
and funding smaller surveys, that that might be seen as a great supplement
to what FoodNet does that is valuable.
AUDIENCE PARTICIPANT: [Off mic.] With listeriosis
you may not know what type of food it was from. On the water side,
the vehicle is more or less the same every time and it's really the organism.
AUDIENCE PARTICIPANT: I don't think so,
because there are certain inherent problems in epidemiology, especially
having to do with these chronic diseases in terms of what the exposure
was. And you have number of foods. Unless you send a team
in there to find out what's really going on, you're going to have a lot
of trouble figuring it out.
DR. COLFORD: Yeah, that's the team that's
sent in. Right.
MS. COLEMAN: Are you saying that you don't
think that epidemiology can provide the validation?
AUDIENCE PARTICIPANT: It can't do it.
I don't think so. I think what you're going to have to do is your
human studies, controlled human studies. The kind of stuff Jack
was talking about, but expanding those studies, and getting some legs
under the data as she's doing, and then use that data on the population.
We went through this with chemical risk assessment a long time ago. The
idea was that you could have a serum program, you could have all these
cancer programs, and then you try to do the toxic materials and you just
can't do it.
DR. COLFORD: You'll be surprised that I
agree with you. I want to merge two of your comments here, because
you're leaning towards cancer there. And in fact, with infections,
what's quite different is the incubations are much shorter and you can
measure infections. So in Iowa, we're about to begin this month
a study of 1200 people, where we're doing something to their water and
measuring how their illness changes over a year, so you can do experiments.
AUDIENCE PARTICIPANT: Exposure.
DR. COLFORD: Certainly.
AUDIENCE PARTICIPANT: Because a lot of
stuff's going on there in the exposure scenarios. You need to do a really
good --?
DR. COLFORD: Well, it's a randomized trial. I
can explain the experimental design.
MS. COLEMAN: We're not here to solve the
problems. We're just brainstorming, and let the ideas flow. Go
ahead, Chuck?
DR. HAAS: Yeah, I would just, speak up
as a risk assessor that strongly supports the need to get better epidemiology,
as John pointed out, and I think you've hit the nail on the head.
You know, weakness in many of the outbreak epidemiological studies is
lack of attention paid to exposure estimation. And I would just
site two outbreaks in the Cryptosporidium area to illustrate this.
You know, Milwaukee, I think, did a very good
job at estimating what the attack rate was and a very good job at pinpointing
the source, but did a very poor job of getting a baseline exposure estimate,
and we had to go through all sorts of hoops to get, really, a very crude
and very rough estimate of ingested dose. On the other hand, there
was an outbreak in the U.K., I think Bradford, a year or two before or
after Milwaukee, much smaller, maybe a couple of hundred to 1000 people,
where they did an excellent job on both sides of the equation, on getting
the attack rate, as well as doing a fairly intensive sampling program
to attempt to do dose reconciliation.
So I think the idea of--and I don't know whether
I'm properly expressing--it getting the epidemiologist more attuned to
the need to do exposure estimation.
DR. COLFORD: We would word it the other
way.
DR. HAAS: Okay. Fine. Fine. You
know, that's what needs to be done.
MS. COLEMAN: Shall we move on to the next
question? How well do dose-response models based on data from human
trials predict human dose-response relationships during outbreaks? Anyone
like to comment on that?
DR. CHAPPELL: I'll comment on it.
MS. COLEMAN: Go ahead.
DR. CHAPPELL: I think it's been brought
up two or three times. We have a very particular subset of the population
that we're looking at. These are the healthiest people we can find
and, of course, when you have an outbreak, you have people who are not
healthy and people who are old and people who are young. And so
it only provides you one picture of the whole. It's one small part. So
I don't think it's, in and of itself, the entire answer.
DR. COLFORD: And I wonder, too, if you
might not have a situation in which in the outbreak, in a sense, you're
going to see rising to the top the most susceptible, versus you're on
the other end of the spectrum in the human volunteer studies, for obvious
reasons, the least susceptible, in a sense, the healthiest people. Because
you've excluded in your inclusion criteria all those people who have
any reason to be more susceptible.
DR. CHAPPELL: It's a little bit like the
tissue culture model. It's the most conservative that you can be.
DR. COLFORD: Right.
MS. COLEMAN: Other comments?
DR. HAAS: We may have a difference here,
though, with water versus food, because if I think of the water-borne
outbreaks we've had in the U.S. anyway, they tend to be fairly broad population
exposures. So it's not just the most vulnerable that rise to the
surface, but it's really, maybe not an average of the population but closer
to the average of the population. My suspicion--and now I'm going
to show my ignorance on the food side, perhaps--is in a foodborne exposure,
perhaps more frequently it is the more sensitive, because the exposures
tend to be more focused for the most part. I'm thinking of your
food abuse, restaurant type of situation here.
MS. COLEMAN: And the idea that you can
have growth in a food matrix and unlikely in water? That that's
your thinking?
DR. HAAS: Well, no. The thesis is
that the number of people who are exposed tend to, for the most part,
be smaller in most of the foodborne outbreaks, although there are exceptions. And,
therefore, to see the outbreak at all, you see it because the more sensitive
subpopulations amongst that small denominator come to the surface.
MS. COLEMAN: Isabel?
DR. WALLS: Yes. One of the speakers
this morning used a factor of three to account for those individuals who
had HIV, and I wonder where that factor came from and whether anybody
had used any other factors in looking from animal models to humans. I've
never seen this.
AUDIENCE PARTICIPANT: In terms of the process
by which it was determined, it came from a paper successfully done in
the state of New York by Bird, et al. I can't remember exactly
where it first came from.
DR. HAAS: Columbia.
AUDIENCE PARTICIPANT: Thank you.
It was little more than a back of the envelope estimation looking at differential
attack rates, and as I mentioned--and as the attack rates merged together,
the probability of infection and the probability of illness, and so teasing
out those two is difficult, and it was justified by saying somewhere between,
you know, nothing--somewhere between one and the difference that we see
the ratio attack rates. So it's not well-substantiated.
AUDIENCE PARTICIPANT: Well, you're never
going to do human dose-responses in susceptible individuals. You
can do dose-responses in susceptible animals. And I think the very
encouraging message that I got from Cynthia's talk is the very good,
excellent,
in fact, correlation that you're seeing between at least these three
strains of Cryptosporidium among the response to the infection,
the shedding response in the mouse models that you're looking at. Even
to a certain extent in the cell culture models, correlating with the
degree of infection
and severity of the infection in humans.
And I think it's the understanding that maybe
a few more data points generated along those comparisons that would give
you confidence, at least, putting a parameter on how the shapes of the
curves look in an animal versus a healthy human, and then being able to
hopefully use mouse models that are deficient in one way or another to
see how it affects the curve in that animal model, and then make the projections,
intelligent projections, of what it might do in a similar human. And
I think that is really the power of this comparison.
DR. CHAPPELL: In fact, we saw the best
correlation with the tissue culture, and a good correlation with the mouse
model. And I think listening to the other talks today, one of the
reasons that we got the data that we did was that we chose a human cell
line from the region of the gut that we think is the most susceptible
to this infection. I mean that's as close as you can get to the
natural receptors, perhaps, as we're going to get. But I agree
with you.
I think, then, you need to move to the animal
model in order to study the organism in a different way, asking different
questions and knowing that you have different factors being very important
in those models that you didn't have in the tissue culture.
AUDIENCE PARTICIPANT: And it's also another
way to approach the end question, is at the very low infectivity doses. And
in fact, Cryptosporidium is probably not so significant, because
it's very infective in one, but when you're talking about a Salmonella where, you'd have to expose thousands of human volunteers at 10 organisms
or 100 organisms to get an infection, you might very well be able to do
that experiment in a mouse model.
AUDIENCE PARTICIPANT: Yeah, to follow up
on that point, again, one aspect that needs to be considered in preparing
the animal models or humans, that the mechanisms of pathogenesis that
are going on in our animal models that we're using need to bear some
relationship to at least what we think is going on in human models, or human
disease,
as well, or you may get a very false picture of that extrapolation.
DR. CHAPPELL: I would agree with you, and
I think that's one advantage of the pig model, because the pig is the
closest physiologically to the human GI tract, as the pig, as well as
the mouse model, at least the neonatal mouse models, has another problem,
in that you're looking at very immature animals, immature immune systems.
So as I said before, there are pluses and minuses for each model. You
just have to define what you're saying. You have to know what you're
dealing with and factor that in.
AUDIENCE PARTICIPANT: And in particular
the immunocompromised mouse model that was being referred to, it's kind
of important to relate those immunocompromised characteristics of those
mouse models to what we think are important in the human situation.
DR. COLFORD: Nobody mentioned primate models. Are
they out of favor or not well developed?
MS. COLEMAN: We have one amongst us, a
primate model.
DR. COLFORD: Please rise.
MS. SMITH: I think she's talking about
me, but I don't know if I want to admit it. Yeah, we're using a
primate model, but for Listeria, not for Cryptosporidium,
so I think, again, it depends on what you're looking at, it depends on
the question you want to ask.
DR. COLFORD: I mean specifically thinking
of his point, though, like gut mucosal immunity and other factors to
deal with pathogenesis.
MS. SMITH: Obviously we think the primate's
a very good model, but for what we're doing, which is looking at listeriosis
in pregnancy, we don't think there's much else that you can do except
go to something like a primate model to try to get a good idea of what's
going on in the human situation. So, you know, I think primates
are a very good model for humans in most situations, but I think, of
course,
the downside is it's very expensive.
DR. COLFORD: But it can't be more expensive
than the humans, I wouldn't think. Is it? Oh, because of
the cages.
MS. SMITH: It depends on what question
you're asking.
MS. COLEMAN: Shall we, okay, one more comment.
DR. TZIPORI: We use it mostly as an HIV
model, because if you want to look at the relationship between infection
and the immunodeficient hosts as it relates to AIDS, you know, there just
isn't anything better. And we use it to look at immune parameters,
like I mentioned when I was talking, and some therapies combining infection
of SIV and Cryptosporidium.
MS. COLEMAN: One last question, Tom.
AUDIENCE PARTICIPANT: One other aspect
regarding human dose-response models that are currently available that
may not be very predictive of disease outbreaks is that they don't take
into consideration the fact that response--the coming together of the
exposure and the response is a random event and that foodborne pathogens
are heterogeneously distributed in food and that, one of the things that's
not being taken into account in current dose-response models is this random
coming together of the exposure and the infection dose.
And one of the ways that can be done is by using
simulation and doing sort of a two-staged dose-response model, where your
first stage is a probability distribution that describes the distribution
of infection dose, which is the result of the disease triangle, and then
what the Monte Carlo simulation would do is take the exposure from serving
one through 10,000 and match it up with a random sampling of the probability
distribution for infection dose to bring the two together in a random
fashion. And the current dose-response models don't do that.
MS. COLEMAN: Unless you'd like an exposure
model. But let's go ahead and move on to the next question.
Thanks, everyone, for your comments so far. Are the currently available
data or model systems adequate for modeling human dose-response issues
and, if not, what new data would be more useful? Chuck?
DR. HAAS: Well, I'll repeat what I said
in a conference call. My gut reaction to this question is that
it's really a risk management question, and I think to some degree, the
people
that are intending to make use of the output of risk assessment need
to be more explicit in terms of the level of uncertainty they're willing
to tolerate before they make a decision.
MS. COLEMAN: Chuck has also pointed to
me in the past that in water safety, there has been a bright line set
and that that's been useful in moving the field along. There aren't
really bright lines for us who are dealing with microbial pathogens in
foods, and so educating our risk managers about how would we actually
make a decision about what's acceptable risk, I mean that goes back to
analytical deliberative process, I think. So I agree with Chuck. Other
comments?
AUDIENCE PARTICIPANT: Yeah, I guess I'd
like to add that I think it would be worthwhile to have an "outbreak swat
team," and next time there's a outbreak, there would be tools available
for people from Houston to Boston in collecting exposure data knows that
data.
MS. COLEMAN: All right, Wes, did you plant
that question?
DR. LONG: No, I didn't, but I'll respond
to it. One of the activities at the Risk Assessment Consortium has been
the consideration of the development of a similar approach, and we have
a study that's been underway at Chicago Public Health Department to try
to identify what we're calling "triggers," which are essentially conditions
that are taking place as you see an outbreak evolving that would convince
you to put more resources towards that outbreak because of the likelihood
of getting good data and information on dose-response.
And it's been very challenging for Chicago Public
Health, and what they've found is that the triggers that need to be there,
even in a city like Chicago that has a lot of outbreaks, it's a pretty
rare event. And to date, they have not been able to satisfy all
the triggers that they developed in their models.
But the next step in the project is to take it
to New York state and see if they adapt that model developed in Chicago
to New York state, and then we're going to try to take it, if it's workable,
we want to take it nationwide and see if we can get the local--it's at
the local level this has to happen with food, because they're the ones
who are there when it happens, and they're the ones who just need to be
educated to look for these opportunities. And I think that they'll
have no trouble getting help once they've identified, oh, here's an outbreak
that seems to meet the criteria that might make this a scientific study
and not just a--not that it's not very important--but not just a response
to an outbreak.
DR. SMITH: Can I comment to that?
I want to play a devil's advocate just a little bit to that, because I'm
a toxicologist, so I'm coming from a little bit different viewpoint, I
know, and bias also, but I think it's very difficult, and it goes back
to what Angelo said. It's very difficult to get good dose-response
date from those kinds of situations, whether they're outbreaks or even
surveillance.
And I'm trying not to talk too much about Listeria,
because I know our purpose here today is not Listeria, but in the
case of pregnancy and exposure to Listeria, you're three weeks
away from the exposure to the outcome is generally about three weeks. You're
not going to get very much dose-response information, I think, in that
length of time.
And I think that the other really big issue in
that kind of surveillance or outbreak data is going to be quality control.
If you're going to use that kind of data and it's going to be useful in
dose-response modeling, you have to have some kind of confidence that
laboratories you send those samples to are really sampling the same thing
and that they are coming up with numbers that are similar. So I
think I would encourage a really good system of quality control if you're
going to try to move to something like that.
DR. LONG: I agree. And the approach
we took in the risk assessment is to use outbreaks as just for comparative
purposes to see, does the outbreak seem to match the calculated estimate.
DR. SMITH: And I think maybe the bottom
line is that's going to depend to a great deal on the organism and the
situation.
MS. COLEMAN: I just thought I'd throw in
that the Risk Assessment Consortium has been discussing the multidisciplinary
nature of this kind of process and how infrequently we really involve
someone who deals with microbial ecology of foods in an outbreak investigation.
So when you see a report of one MPN number or two NPN numbers, does that
really tell you enough to be able to use that information to create the
possible exposures from lots of food that probably are very different
and getting over those assumptions of homogeneous distribution, like Tom
said. We know that organisms aren't homogeneously distributed in
foods, so our models need to start addressing those real life issues.
AUDIENCE PARTICIPANT: I was just going
to say at EPA, we're currently developing further regulations to control Cryptosporidium, and a big part of that analysis is to try to develop
a characterization of the risk to the population at large in the U.S.,
and then what change in that risk will occur as a function of the various
criteria that are considered, so we are engaged very much in using this
information to project national estimates.
I think one of the big uncertainties with this
for us is struggling what to do about subpopulations and what to do about
the interpretation of the infection, in terms of symptomatic response,
duration, severity, because they have some very significant cost implications. So
when we do impact analysis, we try to at least characterize these uncertainties,
recognizing them, but there are huge native gaps that we're dealing with.
AUDIENCE PARTICIPANT: And just to continue
with that point, where I think the data that we find most useful in calculating
these risks, because we're talking about the methods of treatment plants
treating source waters, but not great level of Cryptosporidium.
The finished water levels might be an oocyst in 1000 liters or an oocyst
in 10 million liters, so what the consumer gets is maybe one oocyst or
two. So the data we need most would be dose-response data at that
kind of level. And I'm kind of wondering, what's the feasibility
of the study to address that?
DR. SLIFKO: I'd just like to make a comment,
and it's one of the things that I think we've talked about kind of indirectly.
one of the issues, in developing the dose-responses is does infectivity,
regardless of the model that's used, predict illness. And to better
define what parameters we should be considering when looking at our triangle
between the host pathogen and the matrix that we're testing. And
so I think it's a real weakness, even though we have all of this data
and we have a few models to consider at this time, what is the best definition
to use?
MS. COLEMAN: Tom?
DR OSCAR: I wanted to expand on my comment
that I made earlier about the food matrix. One of the components
of the food matrix that I don't think is being considered, but that would
really enhance the type of data that we have in dose-response modeling,
is the competitive micro flora. In the poultry industry, one of
the techniques that's being used to try to reduce the incidence of pathogens
in birds is to give them competitive exclusion cultures that they feed
the bird early in the life when the bird doesn't have a mature gastrointestinal
tract, and the idea is that the competitive micro flora blocks the Salmonella from being able to take hold in the chicken.
And one of the things that might be very important
in our dose-response modeling studies is to actually deliver the dose
in the food matrix that we're concerned about, because the presence of
the competitive micro flora may act as a protective buffer for the pathogen,
which is usually a minority member of the microflora, may protect it as
it goes through the gastrointestinal tract.
In addition, the composition of that flora, the
types of competing organisms, may also affect the ability of that pathogen
to grow in the intestinal tract and cause an infection. And so
the dose-response may be greatly affected by the competitive microflora
in
the food matrix.
And so the only way we might be able to get at
that question is develop a model system where we could actually test the
food matrix. And one of the potential model systems here would be
the domestic pig. Being an omnivore, we could feet a pig a Big
Mac or a Whopper and at least get relative effect of the food matrix,
on the
infection dose in the animal.
MS. COLEMAN: Great point, Tom. And
if Marianne Miliotis were here, she would probably tell you, guess what,
we're feeding in oysters in the human clinical trial that's starting up
in Maryland.
DR. HAAS: I just want to make one observation.
You know, the question you're raising is interesting and important. I'll
speak now as a non-toxicologist, and this is dangerous with Mary Alice
next to me. After, 30 some odd years of regulatory toxicology and
regulating each individual agent, agent by agent, and neglecting interactions,
the toxicological field is finally getting around to realize that mixture
interactions may be relevant and important to look at.
I don't know how many billions of dollars have
been pumped into doing animal testing on chemical agents over that period.
We're a thousand-fold less on the microbial side in terms of the level
of effort that's been devoted to gathering the underlying data. And
so if we really want that sort of information, which is important, I
think it's also important to start making an effort, in terms of funds
expended, to get the kind of information on a par with the sort of information
that's been obtained on the chemical side.
MS. COLEMAN: Great point. We better
move on to the next question. What lessons can we learn from Cryptosporidium dose-response
modeling that can inform model systems for other pathogens? Remember,
one of the reasons we chose Cryptosporidium to have this
as a kickoff meeting for our first dose-response public meeting on a specific
pathogen was because there was so much more data from different sources,
from animal models, in vitro, human clinical trials. After
having thought about all these data that are available during this time,
are there gems that we can glean as far as developing systems for other
pathogens? Isabel wants to tackle it. Go ahead.
DR. WALLS: I just want to make the comment
that for Cryptosporidium, I don't believe we have any chronic
sequelae, and that it is probably as safe as any pathogen to do human
studies, whereas
with some, you cannot do many other pathogens. As much as we might
like human data, we cannot do it. With Listeria, with Salmonella,
don't think with any, because of the possibility that we will go on to
develop some really nasty chronic sequelae syndrome or arthritis.
So I don't know. I don't know what the
answers are. But I think we do need to retain our focus looking
at animal models. My question to the panel, at some point today,
would be what do they think is going to be the most appropriate animal
models?
Mostly, we've heard about mice, although I know
some of you are working on other animals. And I'd like to know,
where are we in terms of animal modeling and is there going to be a good
animal model. Has anybody looked at rats? Like a chemical
toxicology? Maybe they can address that later.
MS. COLEMAN: Before the panel tackles that,
I just want to mention that there actually was a Campylobacter human
clinical trial that was conducted at Fort Detrick and Dennis is going to tell
us
about it.
DR. LANG: Well, not to say that there can't
be a trial, but I'm at NIH in the enteric diseases program there, so we
have a lot of experience in doing human trials and human exposures. Not
with Listeria, but yes with E. coli and cholera, and
other diarrheagenic agents. And I think it's instructive to me
to see the data that's coming out of this comparison to Cryptosporidium.
Two things are apparent to me. We know
very little about the genetics, regulation, molecular biology, even how
to grow this thing in culture, and yet I think that the pathogenesis models
that were presented today are better than we have with anything else,
including with these other organisms that have been completely sequenced
that we understand hundreds of pathogenesis factors and yet we still don't
have a good correlation with the human exposure data with a good and predictable
animal model. That's the first point.
The second point is I think the only way you're
going to be able to generate that kind of information is exactly the
way you guys have done this study, and that's to take the same culture grown
up under the same conditions and do parallel studies, and that's something
that's not occurred in any other pathogen that I'm aware of.
MS. COLEMAN: Any other comments?
DR. SLIFKO: I'd like to just mirror that.
One of the difficulties with me whenever I was working on handling my
data, was the study design. I didn't design a study to look into
dose-responses of subculture. I did serial dilutions, and I based
these on human count.
So I think maybe a better study design, if that's
what you want, would be something that would be considered for doing cell
culture assays or animals or whatever to determine those low-dose levels. Now
we've defined that this needs to be done, and so we have some other things
to think about.
MS. COLEMAN: I'd broached this question
to Dale Haddis, who's at Clark University, and he's one of the stalwarts
in Society for Risk Analysis who's been active in dose-response modeling,
and he said if you're going to design a study, you're going to have to
not apportion equal numbers, or lower numbers in low-dose groups. You're
going to have to load them up so that you can have greater confidence
from those data, so I agree with you, study design is one improvement
that we could make as we're continuing to research these problems.
AUDIENCE PARTICIPANT: Everyone has commented
on keeping the focus on what the purpose is of the model that you're using,
what the limitations are, and it seems like researchers don't feel like
they're really communicating to risk managers about the limitations of
their models and their data. And I'm wondering if one possible
thing we could learn from Cryptosporidium is that there should
be some sort of forum for people who are actually working on the data
for communicating
the limitations of their model. That's not a popular thing to do
in publications, because it can hurt you with reviewers.
But if any of you are on editorial boards at
various journals, then I know that one way you can change the whole system
is for you to have journals have some sort of requirement that, when you
submit a piece, you make some sort of statement as to what you specifically
think the limitations might be, like this really shouldn't be applied
to this, because we think this.
I don't know whether that's really going out
on a limb or whether that would sort of help, but I feel that a lot of
times, researchers feel like the managers ought to know that. It
ought to be intuitive and it ought to be obvious, but everyone has repeated
over and over here that it's really important to be aware of all of that.
DR. CHAPPELL: I'd like to respond to that.
In a sense, if you wait till you get to the publication, it's kind of
too late. It's much better if I talk to the risk assessment modelers
and they talk to me and we go through these issues at every point, we
do that before the experiments get started and we identify as many issues
as we know. At that point, there will be new issues that come up
as the experiments go on. I think this continuing dialogue is essential
to get the most out of the experiments that we're doing.
We can do a lot of experiments that are extremely
interesting to us, but they may or may not be interesting to risk assessment
folks. Or we may not be expressing those data in a way that you
can use them best. We simply have to talk to each other to know
what our individual issues are to be able to supply the data in the best
way that we can for everyone.
MS. COLEMAN: That seems to speak very highly
of the cooperative agreement type of format that FDA had funded, and we
hope funding continues and maybe from multiple agencies. Other comments?
Let's move on to the last question. Can data from different biological
models shed light on the basic underlying processes that risk assessors
need to develop plausible mathematical models? Who'd like to tackle
that one first?
AUDIENCE PARTICIPANT: A short comment on
this, and I think this one is a definite yes. We'd like to in trial,
but that will never happen. So that's why they're using cholera,
because that's apparently much safer.
So a good situation with Cryptosporidium here is it's got
lots of different models, I mean different animal models. It would
be interesting, if you could do the similar types of cell culture models
and animal models, perhaps use that as a bridge, you could predict human
dose-response with something like Cryptosporidium or some other
relatively safe pathogen in human trial for the anchor in leaping from
animal to human.
MS. COLEMAN: Other comments? We might
be all talked out. Well, that is our last question. Wes,
how should we proceed from here.
DR. LONG: [Off mic.] I think we should
wrap up.
MS. COLEMAN: Well, Wes had asked me to
kind of offer a few conclusions. It does seem to me that having
a forum like this where we can bring in the research community, the regulators,
and really get into dialogue about specific issues and extending the
data
that we know and using scientific method to discover more, that's pretty
exciting.
And I'm glad that the Risk Assessment Consortium
has been able to fund two or three meetings like this now. And we
hope to continue to be part of the analytical deliberative process with
microbial risk assessment and in other areas. And I think I'll
just stop right there.
DR. LONG: I just want to make one comment,
and that is that within two or three months, the entire proceeding should
be up on the web at the risk assessment clearinghouse. I just want to
echo Peg's comments, and I think we had a really good meeting today.
I hope we're able to continue to offer opportunities like this for
groups like this to gather and talk about these issues. Thanks
for coming.
[Applause.]
[Whereupon, at 4:47 p.m., the meeting was adjourned.]
