Getting ready for the bioinformatics shakeup
Pathologists have long suspected, even feared, that bioinformatics
would turn their world upside down.
It will. As test results and clinical patient data merge, and
as the lines blur between anatomic and clinical pathology, no pathologist
should expect her world to remain unshaken.
But if bioinformatics has thrown down the gauntlet, it has also
opened the door to tremendous opportunity for laboratories and pa-thol-o-gists.
The challenge for pathology is both "worse and more interesting
than we originally thought," says Jonathan Braun, MD, professor
and chair, Department of Pathology and Laboratory Medicine, Geffen
School of Medicine, University of California, Los Angeles.
Recent advances in genetics research are yielding profound insights into disease
identification and treatment, as well as suggesting new directions for how pathologists
learn and practice their profession. Making use of these insights will require
pathologists who understand molecular dynamics and can bridge the worlds of
research and clinical care. It will also require pathologists who understand
and can work comfortably with high-throughput technologies as well as the computational
techniques and tools that make these insights accessible.
From the morphological to the molecular
But playing this role will cause deep changes in the profession.
For example, predicts Bruce Friedman, MD, pathology will see a gradual
shift from cell morphology to molecular structure and function of
cells as an adjunct to the diagnostic process. In the future, says
Dr. Friedman, pathologists "must be able to render a tissue diagnosis—most
particularly for but not necessarily confined to tumors—that
takes advantage of not only morphologic observations but also related
genomic and proteomic testing." Ultimately, but "not tomorrow,"
Dr. Friedman says, morphology will dwindle in significance, becoming
more of a diagnostic afterthought, referred to "on the way out the
door, as a kind of final confirmation that you know what you’re
doing and as a final check," as he puts it. Dr. Friedman is professor
of pathology at the University of Michigan Medical School and Health
In many respects, this transition is already under way. He-mato-pa-thol-ogy
and flow cytometry, to take just two instances, already take cell
chemistry as well as morphology into account. Moreover, pathologists
working in these fields tend to play a more active clinical role
in shaping treatment decisions than do many other types of pathologists,
a change Dr. Friedman believes pathology in general will be well
advised to anticipate and seek in the future.
Another indication that this shift is under way is that "many
or most of the new classification schemes for tumors have a genetic,
not a morphologic, basis," Dr. Friedman says. As these new classification
schemes develop, pathology is also learning more about the ways
in which previous systems, largely or completely based on morphological
characteristics, may have blurred functional distinctions with important
clinical implications. "We’re finding that things we lumped together
because they looked morphologically the same sometimes don’t really
behave in the same way biologically," says Dr. Friedman. "We’re
learning that they are genetically programmed to do different things,
most importantly, to either remain in the same place or to metastasize."
In prostate research, for example, researchers armed with new genomics and
proteomics insights are finding that certain smaller tumors, which might well
have been regarded as innocuous in the past, are in fact aggressive and can
metastasize in weeks, while other, larger tumors, which would previously have
been viewed with concern, simply do not have the enzyme system and other cellular
constituents required to implant elsewhere in the host, to metastasize. "Not
all prostate tumors that look bad under the microscope need to be treated or
treated aggressively," Dr. Friedman observes, "because they’re simply not all
programmed to metastasize and become lethal."
Neither CP nor AP
Moving from the morphological to the molecular has blurred the
traditional boundaries between anatomic and clinical pathology.
In the future, Dr. Friedman suggests, pathologists are more likely
to find that "what we will have to learn to do is talk about diagnosis
and about the different ways to analyze the serum and tissue that’s
sent to us, some of which are chemical, some of which are morphologic,
but all of which are essentially part of the diagnostic domain."
UCLA’s Dr. Braun, who also sees deep changes ahead for pathology,
tends to define this particular shift in terms of cell biology versus
biochemistry. "In a way, AP has been focused on classic cell biology
and on using that as a window for formulating questions and framing
the pathologist’s diagnosis, prognosis, and treatment interpretations
and options," Dr. Braun says. Laboratory medicine or clinical pathology,
by contrast, has tended to focus more prominently on the biochemistry
Proteomics and bioinformatics may offer something of a third choice, a bridge
of sorts, one that touches on both of its traditional predecessors, Dr. Braun
says. "So, for example, if you collect some tissue, prepare an expression array,
1 analyze the levels of the proteins and the other specific things that
define a disease state, and look at that over time to understand whether the
disease state is remaining or going away and how your treatment recommendations
are working, you’re starting with cells, but the approach you’re taking is really
a biochemical one," Dr. Braun says. "The process is in some sense right in the
middle between the anatomic pathologist and the lab medicine pathologist, and
connected in significant ways to the work both of them do."
As bioinformatics becomes a more central and significant driving
force in pathology, it is likely to transform even the most basic
day-to-day functions and priorities of the pathologist. Results
reporting, for example, is likely to steadily demand a far greater
proportion of the pathologist’s time and attention in the future.
"In the past, we spent 90 percent of our time actually doing the
test and 10 percent providing the result," says Michael Becich,
MD, PhD. This will change dramatically, he predicts.
"The amount of time spent actually doing the test will decrease,
and the amount of time spent integrating test data and patient data—the
analytical dimension of patho-bioinformatics, if you will—will
consume the major part of the pathologist’s workload," he says.
Dr. Becich is director of the Center for Pathology Informatics and
director for the Benedum Oncology Informatics Center at the University
of Pittsburgh Medical School.
Parenthetically, Dr. Becich sees this change as yet another way
in which the practice of pathology will necessarily blur the traditional
distinction between anatomic and clinical domains. "In the past,
if you had a clinical pathology test, you shot out a page of numbers,"
Dr. Becich says. "If you had an anatomic pathology test, you shot
out a page of text, and never the twain would meet." In the future,
by contrast, neither one of these modes is likely to prove sufficient.
"Integrating and interpreting both numeric and textual data into
one integrated report is going to be the good work of future pathologists."
Closely related to these changes is the need and opportunity for
pathologists in the future to assume a more active role in clinical
decision-making. Pathologists in the past had a tendency to remain
somewhat detached from care delivery, Dr. Becich says. "Until now,
we’ve had a tendency to sit behind a curtain, spew out our reports,
and let the physicians figure them out," he says. "It was a little
like the Wizard of Oz, sitting behind the curtain and uttering oracular
Back in the days when the testing options were relatively simple—"when
we had an SMA 20 and a handful of AP tests," as Dr. Becich puts it—that
posture may have been appropriate, he adds. But it isn’t likely to work in the
future. "If we want the insights that are now becoming available from patho-bioinformatics
to really have an impact on patient care, pathologists will no longer be able
to just toss the diagnosis over the transom, as it were, but will instead find
that we need to get out from behind the curtain and take on the job of integrating
and interpreting reports for our clinical colleagues," he says. "That is the
’sweet spot’ for our profession, actively participating in shaping the process
of delivering care."
What is a "disease"?
At the same time that the world of bioinformatics is promising
to overturn the pathologist’s world, it is also raising difficult
questions about the very object of that work, the very nature of
"The concept behind bioinformatics is to look for significant
patterns in large datasets, not at individual biological entities,"
says UCLA’s Dr. Braun. "This has been profoundly annoying to basic
biologists." In other words, he explains, people who have worked
hard to understand the behavior of individual proteins and how they
are regulated are now being told to let go of that perspective.
"Suddenly a group of rash informatics people is telling them that
that is not the big, the important question," Dr. Braun says. "The
important question, they’re being told, is finding ways to collect
very large datasets and identify patterns or signatures of biologic
Even then, the bioformatics challenge continues. Knowing when
the patterns found in those large datasets relate to a traditional
disease state—or, for that matter, a known biologic condition—is
not always simple. "One of the questions we are still trying hard
to answer is, How do you set up an approach where after a period
of time you can determine that one signature or pattern does relate
to a specific biologic state?" Dr. Braun says. "And even if we can
answer that question, we aren’t sure whether those biologic states
will be ones we’ve formulated before or are entirely new and different."
Answers to these questions are not likely to be quickly or easily forthcoming.
"Talk to anyone who is doing bioinformatics in a basic research lab and they
will tell you that while it probably will take only a few days to collect the
data, it may take months to analyze it, and even more months to understand how
what is found relates to what we already think we know about, say, breast cancer—if
it does at all," Dr. Braun says. "We may look at data from, say, 100 or 1,000
people, and we may find 10 major patterns or signatures, but they won’t necessarily
relate to previously defined diagnostic disease states." We now face this very
challenge, he adds, in current research on lymphoma and breast cancer. "None
of this new work is interesting unless it helps define and guide prognosis and
treatment," he says.
The critical biorepository
Pathology also must understand its stewardship of a resource that
is likely to become more critical to bioinformatics research: the
tissue or biorepository. Says Dr. Becich, "Every tissue that comes
across our surgical benches and every blood sample that comes into
our clinical pathology laboratories has to be highly managed and
refined to allow controlled genomic and therapeutic investigation
Comprehensive and well-organized biorepositories with a sophisticated
"infostructure" are critical to bioinformatics advances, and the
importance of this resource is likely to grow. "There is a lot to
be gained by managing tissue and serum samples and creating refined
products for very well-classified disease states, so we know, for
example, what a true normal is, for comparative analysis," Dr. Becich
says. For example, having a prostate cancer sample that has never
been treated with radiation, never been treated with any kind of
endocrine therapy, and that comes from a patient without a family
history of prostate cancer means it can serve as a standard for
comparative analysis. Or, a repository might be able to provide
equally important longitudinal information, by obtaining samples
ranging from needle biopsy samples from a patient in, say, 1994,
then from that patient’s prostatectomy in 1996, metastatic disease
in 1999, and at his death in 2001. "If you have samples from each
of these stages, you have the entire life cycle of an individual
tumor," Dr. Becich says. "That is the most valuable information
of all, but obtaining it takes a very coordinated effort."
What you gain from such effort is a rich clinical biorepository, a resource
that "will be as strategic to pathology in the future as running a control panel
of electrolytes on cardiac enzymes is today," says Dr. Becich. So strategic,
he notes, that several commercial start-ups, one of which he himself participates
in, offer for-profit services in this space. "If pathology doesn’t rally and
take the lead in this aspect of the bioinformatics initiative, other groups—departments
of medicine, structural biology departments—will."
Getting from here to there
What is the role for pathology in this postgenomics era? "Pa-thol-ogy
is as well poised as any discipline to be able to drive these innovations,"
Dr. Braun says. "On the lab medicine side we have the expertise
in biochemistry and in handling large datasets, and on the AP side
we have the perspective of understanding cell biology, which is
what a lot of this is about."
If there is a danger, it is that the challenge may appear too
risky. "We may find ourselves saying, ’We’ll do it when it’s ready,’"
Dr. Braun says. "But if we take that position, pathology as a whole
may miss the opportunity to be the innovator here." Identifying
and deploying new bioinformatics insights will undoubtedly proceed,
but it could bypass pathology. "We have to have individuals who
can jump into the breach and develop both a deep understanding of
these new dimensions of biology while also working effectively with
the computational systems that are necessary to handle and analyze
the data," he says. "That’s the test ahead."
The solution involves a primary focus on training the next generation
of pathologists to be pathology bioinformaticians. Dr. Friedman
is proposing changes at the University of Michigan that would incorporate
more exposure to genomics and proteomics for medical students and
pathology residents. He is also urging his colleagues to see the
value of "abundant training in advanced pathology bioinformatics
such that the trainees understand the computational tools necessary
to manage sophisticated data and digital images and will feel comfortable
in future careers in pathology where a large portion of their professional
responsibility will involve the manipulation and query of complex
datasets using computers."
Pathologists need to focus on training pathologists as well as
graduate students and other trainees in bioinformatics, Dr. Becich
agrees. "In the same way that we moved pathology informatics forward
by taking an aggressive stance on the importance of training young
docs and graduate students in this area, we need to do the same
thing now with bioinformatics."
1. See Arraying the data: Bringing order to tissue microarray technology," CAP TODAY, March 2001, pp. 60-64.
Eric Skjei is a writer in Stinson Beach, Calif.