By Aaron Krol
August 28, 2014 | Here
are a few facts about Berg, a pharmaceutical and diagnostics company headquartered
in Framingham, MA, that seem calculated to set off all the alarm bells in the
head of a skeptical observer:
The president and CTO, Niven Narain, claims that his company
can cut the time and cost of bringing a new drug to market in half. He plans to
use molecules naturally found in the human body to treat diseases like cancer,
diabetes, and major neurological conditions — among the most intractable
problems in healthcare. Narain believes his company can uncover these elusive
molecules with an artificial intelligence platform, which will choose potential
drugs itself by looking at big data. And he gets almost all the money to pursue
this project from a single investor, Carl Berg, who made his billion-dollar
fortune in Silicon Valley real estate.
After covering the drug industry for a while, you start to
make some mental substitutions for certain phrases. Artificial intelligence:
pattern recognition. Big data: genomes, and heck, whatever else you got.
Naturally occurring: likely to be a hopelessly complicated biomolecule whose
activity we can hardly begin to understand.
Single real estate investor: well, all right, that one
doesn’t come up as much.
Still, let’s lay one fact on the opposite side of the scale.
Berg was founded in 2006 without a lead compound, a chemical area of interest,
or even a cellular target. By 2012, a cancer therapeutic by the name BPM 31510 was
in early clinical trials, and it has since shown enough promise to expand into
multi-center studies for a variety of cancer types. “In the few years that the
company has been developing [our] platform, we’ve taken two drugs into clinical
development,” Narain tells Bio-IT World.
“The platform has also shown diagnostics that have been clinically validated in
prostate cancer and heart failure, and we have others in the pipeline.”
There’s still a yawning gap between the Phase Ib trials Berg
is currently sponsoring, and a successful drug launch. Phase II remains the
industry’s biggest choke point, where in Narain’s own estimate, nearly three out
of every four drug trials end in failure — and BPM 31510 hasn’t even progressed
that far yet. (A topical form of the drug has been approved for a Phase IIb
trial in squamous cell carcinoma, but enrollment hasn’t finished.)
Still, Berg’s “artificial intelligence” platform, dubbed Interrogative
Biology, has demonstrated it can turn out plausible drug candidates from
scratch, faster than most human-led R&D efforts. If even a fraction of
those treatments make a real difference to patients, it would represent a
genuine advance for the industry.
Berg’s first drug candidate also bucks a key trend in cancer
care. Most pharma companies today are looking at narrowly-targeted cancer
drugs, meant to treat small molecular subtypes of the disease. This has led to
some of the biggest recent advances in oncology, with drugs like Herceptin and
Gleevec seeing huge survival gains for their targeted patient populations, but
it has also limited the impact of any one therapy. By contrast, BPM 31510 has a
broad mechanism of action, and Berg is enrolling patients with any type of
solid tumor in its clinical trials. If the therapy does turn out to be among
the 10% or so of drugs that make it all the way from Phase I studies to FDA
approval, the benefit to patients could be especially large.
Even a skeptic has to hold out a little hope for a result
Design without a
BPM 31510 is a slight reformulation of a compound called
coenzyme Q10, or CoQ for short. In healthy cells, CoQ plays an essential role
in cellular respiration, using oxygen to produce the energy source ATP. Because
it can truthfully be claimed that CoQ helps produce “cellular energy,” a term
that is catnip for shady health gurus, the compound has found a niche market in
the unregulated dietary supplement industry. Enthusiasts claim it helps with
everything from controlling blood pressure to treating Parkinson’s disease,
although the evidence for any of this is shaky at best.
Still, Berg didn’t pluck CoQ out of thin air. The story of
how this molecule, with its less-than-inspiring history of home use, became the
subject of a well-financed cancer study is a good illustration of the way
Narain hopes his company can reinvent the drug discovery process.
Instead of starting from a target in the cell that Berg’s
biologists believe may have an impact on disease, Berg enters a new therapeutic
area by creating virtual models of what’s going on in healthy and diseased
cells. “We don’t make any preconceived hypotheses up front,” says Narain.
“Instead of a hypothesis creating data, at Berg we want the data to create
hypotheses. So we start with human tissue samples.”
Niven Narain, co-founder, president and CTO of Berg. Image credit: Berg
Those samples come out of partnerships with large medical
centers, including the MD Anderson Cancer Center in Texas, Beth Israel
Deaconess in Boston, and the Mount Sinai School of Medicine in New York. Berg
also recently inked a deal with the Department of Defense, the first time the DoD
has shared tissue samples with a commercial partner. The important thing is to
find institutions that keep detailed clinical data on the samples in their
biobanks, so each tissue sample can be associated with a disease history.
The samples go to Berg’s facilities in Massachusetts, where the
company maintains a large array of omics instruments. “On the expression layer,
we look at the genome, proteome, lipidome, metabolome,” says Narain.
High-throughput sequencers find genetic mutations and levels of RNA expression
in the cells, while mass spectrometers try to capture as many of the proteins,
lipids and metabolites in the tissue as possible. Berg will also sometimes
perturb its samples to induce disease-like states, then run all its multi-omics
again to see what changes. And in addition to the clinical and omics data,
Narain adds, “I firmly believe that expression does not always translate into
functionality, so the third layer is functional data… We look at mitochondrial
function, oxidative states, ATP production, to look at how the cell is
It’s this mass of data that forms the basis of Interrogative
Biology’s analysis. Each molecule or genetic variant forms a node in
Interrogative Biology’s map of a disease, and those nodes can cluster together
if they tend to co-occur, or cluster with different disease presentations based
on the clinical data. The platform also has some knowledge of chemical pathways
— which genes lead to which proteins, and how those proteins affect other
compounds’ activities — helping it decide which nodes are truly central to a
disease state, and which are downstream effects of other disturbances.
The Cellular Arsenal
Berg isn’t the only company taking an AI approach to the
search for new drugs. Cloud Pharmaceuticals of North Carolina, and Numerate of
California, are both in the business of generating millions of compounds in silico and modeling their activity in
the body. Both also have pharma partners eager to take small molecules created
by their platforms into preclinical studies.
But Berg departs from these companies by not introducing
novel compounds to its models. In an Interrogative Biology disease map, says
Narain, “the node is the drug. We
don’t do any chemical screening. We don’t do any candidate hit-to-lead. We skip
that entire process… All of our drugs are based on endogenous proteins,
endogenous enzymes or peptides.” In other words, they’re looking for the
compounds in human cells that naturally ward off disease.
That approach has some real advantages for drug design. For
example, by flagging molecules that are over- or under-abundant in diseased
cells, Interrogative Biology can often find promising biomarkers to help with
the diagnosis or prognosis of an illness, even when it doesn’t turn up a drug
candidate; Berg has a separate diagnostics division working to commercialize
some of those markers. The same biomarkers can also help Berg to stratify
patients in its clinical trials, providing advance knowledge of which patient
subpopulations are most likely to respond to treatment. Narain even suspects
that Berg will have an easier time with toxicology than other pharma companies,
because its therapeutics are already present in the human body.
Yet there are also challenges to working with endogenous
molecules. Intellectual property will be a continual headache, since by
definition a naturally occurring molecule can’t be claimed as a new and
protected product. “Nobody’s going to get a composition and matter on these,
obviously,” says Narain, referring to the preferred choice for pharma patents,
covering the chemical structure of the drug itself. “That’s why we stayed under
the radar for a good five years [with BPM 31510], developing IP around method
and use, predictive modeling, the stratified approach, the key metabolic
drivers around the mechanism of action, and the formulation.”
There’s also every chance that Interrogative Biology will
turn up some very complex biomolecules. Even CoQ, a small molecule, is on the
large side for what would traditionally be considered “druglike,” and there are
plenty of peptides and RNA molecules in the cell that are much knottier. This
has implications for mass synthesis, and for delivery, both of which demand
some clever biochemical engineering. One of the biggest design challenges
behind BPM 31510 was delivering CoQ specifically to cancer cells. Berg wound up
repurposing phospholipids from the mitochondrial membrane as a delivery system,
a step that a traditional pharma company would rarely have to take.
Back to Biology
BPM 31510 is a result of running tissue samples from a huge
variety of different tumors through the Interrogative Biology process. The fact
that a single molecule could reappear in disease maps as diverse as brain
cancer, lymphoma, and carcinoma is a hopeful sign, but also a potential red
flag: very few successful cancer drugs are that broadly effective.
While Berg prides itself on a hypothesis-free approach to
generating ideas, the company does eventually want to delve into the biology of
its compounds to be sure it can explain the activity its computational
biologists are seeing. “Once we finish the in
silico output of the platform, we go back to a highly coordinated and
functional point of validation to ensure that we’re looking at what happens in
the cell biology,” says Narain. “What is the mechanism of action? Are we seeing
that this target has an effect in the disease we’re studying?”
For that step, the company orders up a batch of its
prospective compounds from its manufacturing facility in Nashville, Tennessee.
The wet lab in Framingham can then start working with live disease models, and
for the first time, form a hypothesis about how it plans to treat a condition.
These preclinical studies can also help to test Interrogative Biology’s
predictions about toxicity and dose tolerance.
Inside Berg's lab facilities in Massachusetts. Image credit: Berg
With BPM 31510, the working theory is that introducing the
drug to a cancer cell helps to restart the process of apoptosis, or natural
cell death, which is partially regulated by the mitochondria where CoQ is
active. Cancer cells may be able to bypass this process by depending less on
the normal mitochondrial cycle for generating energy from oxygen, and more on
sugars, a well-known phenomenon known as the Warburg effect. “Cancer cells want
to survive on lactate,” says Narain. “What our drug does is come into the mitochondria,
[and] switch the fuel out. And then the downstream effect is to re-engage
There’s no guarantee this will work out in practice, of
course — or, if it does, that it will be as broadly effective in patients as it
is in the lab. But Berg is being careful about its trial design, to ensure it
can quickly zero in on any cancer types where BPM 31510 shows real promise. An
advantage of working with Interrogative Biology is that Berg already has in
place a platform for integrating large amounts of patient data, and that
platform continues to collect new information as clinical trials progress.
“We want to keep Interrogative Biology from the lifecycle of
discovery all the way to post-market surveillance,” says Narain. “It’s
continuous, longitudinal mapping of the biological process that’s ongoing in a
patient.” Already, he adds, Interrogative Biology is finding useful trends
among patients enrolled in the Phase Ib trial for solid tumors. “We are seeing
that in highly metabolic tumors, like triple-negative breast cancer,
colorectal, pancreatic, and glioblastoma, we have a quicker and more pronounced
effect” —consistent with the hypothesis that BPM 31510 targets the
energy-producing activities of cancer cells. The Warburg effect is also thought
to be especially pronounced in squamous cell carcinoma, the subject of Berg’s
most advanced clinical trial.
The $650 Million
It’s important not to pin too much hope on BPM 31510
specifically. Most drugs that make it to clinical trials drop out after
patients don’t respond as expected. But what would be more significant than any
single drug’s approval would be if a platform like Interrogative Biology proves
to be a feasible route to new therapies at all. As more hospital systems are
collecting massive amounts of data on their patients throughout the course of
treatment — Berg’s partner Mount Sinai is a sterling example — the ability to
scan that data directly for potential drug leads, rather than just for targets
and biomarkers, could really accelerate the discovery process.
Fortunately, through its billionaire investor, Berg seems to
have deep pockets, and employs a fairly sizable research staff at its Massachusetts facilities. That’s helping to advance a large number of early leads: Narain
says that around 600 drugs are now in preclinical studies aiming for Investigational
New Drug applications, mostly in the areas of cancer and diabetes. If Berg’s
brand of big data mining really is an effective approach to drug design, the
company should have the resources to prove it.
And if that process can actually cut the cost of developing
drugs in half — the biggest if of all
— it would be hard to exaggerate the impact on healthcare. Narain has a bit of
a cock-eyed way of supporting that claim: he says drugs are so expensive to create today, it would
be almost impossible not to do a
better job, if only you can chisel away at steps like chemical screening and
stratifying clinical trials.
“My reasoning is that it takes, on average, $1.3 billion and
twelve to fourteen years to get one drug to market,” he says. “Based on the
metrics that we have internally, and the entire process of skipping hit-to-lead
optimization and screening, that’s helping to reduce the time from a node to a
patient.” And once a drug is in patients, Interrogative Biology ensures that
some level of patient stratification is already in place, narrowing studies to
the most promising populations. Clinical trials may still be a huge expense,
but perhaps Berg will waste less time on trials that are doomed to failure from
“If I spend $650 million, and seven years, from the time I
have an IND to develop a drug,” Narain concludes, “I’m sure you’d agree that’s
still a lot of money.”
Looking at the pharmaceutical landscape today — where
companies pour their resources into novel compounds that strain to make a
sustained impact on one molecular target, only to find their new drug’s activity
is unpredictable in real patients, and the target may not even be the true
cause of disease — well, it’s always a safe bet in this industry to be a
skeptic. But one would like to believe he might have a point.