This is the fourth post in an open science blog series about PERL101 (P101), a small molecule with big hopes.
Demystifying drug discovery
The default setting of biotech startups and pharma giants alike is stealth mode. At Perlara the default is set to open. As the first biotech PBC (public benefit corporation), transparency is encoded in our corporate DNA. That’s why we’re live-blogging the odyssey of a new chemical entity from hit discovery to preclinical data package to validating deal and beyond.
In October 2016, Perlara and Novartis entered into a research collaboration to use whole-organism phenotypic screens to discover and develop definitive therapies for lysosomal diseases, starting with our lead compound PERL101 (P101) for Niemann-Pick Type C. P101 is the protagonist of this story.
We first glimpsed P101 in late spring, early summer 2016. It was one of a handful of validated hits that passed the two-factor authentication test we devised: compounds that both rescue developmental delay of NPC1-deficient nematode larvae and show low micromolar activity in NPC1 I1061T/I1061T patient fibroblasts in the filipin staining assay.
P101 (center panel) doesn’t merely amplify the disease phenotype (left panel) but it clearly doesn’t clear cholesterol (right panel). P101 goes a third way by redistributing cholesterol throughout the cell.
This figure captures the first in a succession of surprise results. P101’s activity in patient cells was not what we expected. Put another way, it was not what the NPC scholarly literature primed us to expect. Conventional wisdom says compounds that result in reduced or no filipin staining, i.e., clearance of cholesterol storage, are therapeutic candidates. But P101 appeared (to the untrained eye) to do the opposite of the control clearance compound SAHA.
Putting aside for the moment considerations about P101’s mechanism of action, we knew the P101 scaffold had legs when it sprinted the gauntlet of drug metabolism and pharmacokinetic studies between September and November 2015. Unconventionally for a biotech startup looking to partner, those data were pitched not at partnering sessions at BIO or BTS or Partnering for Cures but in a blog post in November 2015. Then came a second installment in April 2016, followed by a third chapter in July 2016.
In the Spring 2016 progress report, P101 achieved two critical preclinical milestones in mice. First, P101 passed the blood-brain barrier penetrability (BBB) test with flying colors. Second, P101 performed well in a 90-day maximum-tolerated dose (MTD) study. In the Summer 2016 progress report, P101 checked off two more boxes on the preclinical checklist: a multi-dose PK and tissue accumulation study in wild-type mice, and a P101 mini-tox study in six Day 35 (i.e., declining) NPC1-/- mice.
January to October of 2017 were consumed with fundraising, so I fell behind with P101 dispatches (and scientific writing in general). To catch us up to the present day, I’ll first share the two remaining pieces of the P101 preclinical data package: 1) a pilot efficacy study of P101-treated NPC-/- mice conducted at Vium between July and September 2016, and 2) P101 mechanism-of-action experiments (MoA) performed on NPC patient fibroblasts between November 2015 and September 2016 at Perlara.
All told, the P101 preclinical data package was assembled over 12 months from September 2015 to September 2016 at a cost of approximately $1.5M.
I’ll conclude this report with highlights of advances made in the first year of the Perlara<>Novartis collaboration from October 2016 to October 2017.
Of mice and MoA
We ended last season with two cliffhangers. Let’s start with the P101 pilot efficacy study in NPC1-/- mice. Knowing full well it was a high-risk, high-reward strategy, we planned to dose NPC1-/- mice with P101 in a treatment paradigm with lifespan extension as the primary endpoint without having done PK and MTD studies in NPC1-/- mice. The results of this pilot study, as you might expect, were mixed.
Here are the plots showing body weight over time for two P101 dose groups (20 mpk and 40 mpk) versus age-matched vehicle control mice and wild-type mice:
Body weight data from male (left panel) and female (right panel) NPC1-/- knockout Balb/c mice. Colors indicate different experimental groups.
So does P101 not have any effect or did we simply under-dose? Serum chemistry data from this study suggest that the answer is yes. We did observe statistically significant, or trending toward significant, normalization of several liver biomarkers, which when combined with our previous observation that P101 has a high liver-to-plasma ratio, led us to conclude that we under-dosed. And under-dosing may be aggravated by the late intervention at Day 35, when liver, spleen and lung already show signs of disease but Purkinje neurons aren’t dying yet.
We performed serum chemistry analysis on NPC1 knockout mice after terminal bleeds. Females are red circles; males are blue circles. Black lines are group means.
40 mpk P101 reduced ALT levels to borderline significance, and normalizing effect is stronger in male NPC1-/- mice. Glucose levels are significantly normalized at 40 mpk (p = 0.0276), and there even appears to be dose response. Cholesterol levels increased significantly at 40 mpk (p = 0.0016). Across the dataset, we observed higher variance in female versus male NPC1-/- mice. Bile acid levels (notice the log scale) were reduced by P101 in male NPC1-/- mice but not in females.
We knew we ran a risk of under-dosing, but we were more afraid of over-dosing. There were several reasons why. Before committing to a pilot efficacy study, we completed a replication study with cyclodextrin in early 2016 in order to calibrate baselines for our NPC1-/- colony housed at Vium (formerly Mousera) with published studies of NPC1-/- mice. One aspect of the replication study that was a cause for concern was the lower than expected percentage of NPC1-/- animals per litter (16-17% versus 25%). Also, NPC1-/- animals showed a more aggressive disease progression than what expected based on previous studies.
The upshot was NPC1-/- mice were very hard to come by. Any affected homozygotes we recovered from each litter were therefore precious. In an ideal world, we would have performed at least one single-dose PK study in NPC1-/- mice before progressing to any efficacy study. But in December 2015 we faced a time-sensitive choice. We had 7 knockout animals available. What were we going to do with them? We decided to enroll these animals in a mini-MTD study so see if the maximum tolerated dose we identified in wild-type mice (80 mpk) is also the MTD in NPC1-/- mice. Turned out 80 mpk appeared to be toxic — based on data from only three NPC1-/- mice.
Those mini-MTD NPC1-/- data combined with the 90-day MTD study and 7-day multi-dose PK study in NPC1+/+ mice showing prodigious P101 accumulation in liver convinced us that we had to back away from the 80 mpk MTD — but how much? Human cell data from P101-treated fibroblasts indicated that doses in excess of 10µM were toxic over chronic exposures, further strengthening our conviction that we had to avoid over-dosing.
That mini-MTD study in NPC1-/- mice in December 2015 was flawed in several ways, one of which we appreciated at the time and the others – which proved to be more critical – only became apparent months later when we were breeding up NPC1-/- mice for the pilot efficacy study. We dosed Day 35 NPC1-/- mice but we dosed Day 21 NPC+/+ mice in the 90-day MTD study. Strike one. We didn’t have enough NPC1-/- mice for a contemporaneous control group, so we used the vehicle-treated group from the aforementioned cyclodextrin replication study. Strike two.
Vium changed what turned out to be several cage features between the start of the cyclodextrin replication study (late 2015) and the anticipated start of the pilot efficacy study (mid-2016). One of the variables was cage size, which got smaller. It’s also likely that their staff acquired experience working with the relatively fragile NPC1-/- mice. The result was that we now observed the expected Mendelian ratio of homozygous animals per litter (25%), which meant we generated NPC1-/- mice faster. These NPC1-/- also appeared healthier than the previous homozygous animals.
Given time and cost constraints, we could only afford two dose groups. Siding with the under-dosing doves versus the over-dosing hawks, we chose 40mpk and 20mpk. What if we had chosen 60 mpk and 40 mpk instead? Or 80 mpk and 60 mpk? Had more resources been available, we would have simply included a third higher dose, either 60 mpk or 80mpk. Alas, we did not have that luxury. Most young platform companies don’t.
Are you out of your filipin mind?!
In parallel, we embarked on an almost yearlong exploration of possible mechanisms of action of P101 in NPC patient and control fibroblasts that culminated in the discovery of a robust cellular readout of P101 action that wasn’t based on filipin staining and that was specific to P101. This proved challenging, as I’ll describe. Every experiment seemed to yield an unexpected result that required continuous revising our model of P101 action.
The conventional wisdom in the NPC researcher community is that intense perinuclear punctate filipin staining is the definitive readout of NPC cellular pathology, i.e., accumulating intracellular unesterified cholesterol. So when we observed that P101 increases but redistributes filipin staining rather than reduces or clears filipin staining in both NPC and wildtype fibroblasts, we knew we faced a potentially uphill battle. U18666A, an inhibitor of NPC1, and chloroquine, a non-specific inhibitor of vesicular transport, also increase overall filipin staining relative to untreated cells. We had to find more robust, more specific and more easily interpretable cellular phenotypes caused by P101. We were spinning our wheels focusing on filipin staining.
We thought we found a way out in Q2 2016 when we first saw cellular antibody staining data for LC3-II, the mature form of the autophagosome-forming protein Atg8, and LAMP1, a lysosomal membrane marker. P101 caused an increase in both markers, which appeared to be colocalized. We interpreted that apparent co-localization as P101 restoring autoghagic flux. Critically, U18666A did not show the formation of LC3-II puncta.
A note of caution when interpreting these data: these images were not taken with a confocal microscope. So we can only make suggestive, not definitive, claims about colocalization.
We tested whether P101 affected other lipid stains or fluorescently labeled lipids. After a brief search we tested NBD-cholesterol. We reasoned that P101 might be mobilizing cholesterol in part via increased non-LDLR-mediated cholesterol uptake. Indeed, P101 induces the formation of NBD-cholesterol puncta, which may be lipid droplets. However, chloroquine does not induce the formation of NBD-cholesterol puncta.
NBD-cholesterol is a fluorescently labeled cholesterol analog that is added to the media, passively taken up by cells and equilibrates in the cytosol. NBD-cholesterol-positive puncta form as soon as one hour after P101 exposure.
Critical negative control. Chloroquine, which increases filipin staining, does NOT induce NBD-cholesterol-positive puncta.
We were vindicated when our Novartis collaborators not only replicated our results with NBD-cholesterol, but they also discovered that P101 unexpectedly clears the fluorescent version of a key regulatory lipid after it’s added to cells, thereby normalizing this lipid’s storage phenotype in both NPC patient and control fibroblasts.
This screenotype has enabled genetic target ID studies because now we could assess the effects of knocking down each human genes in a population of cells treated with the fluorescent lipid and P101. We could sort clones that either failed to respond to P101, i.e., very high fluorescence signal from accumulated labeleld lipid, and clones on the other end of the spectrum that showed an enhanced P101 phenotype, i.e,. very low fluorescence from cleared labeled lipid.
A fluorescently labeled lipid in the Fluopack collection accumulates in NPC1 patient fibroblasts to a greater degree than in control fibros. Remarkably, there’s no storage phenotype evident in P101-treated cells.
Based on those genetic target ID experiments, several candidates are in our sights and in the validation stage. In terms of lead optimization, we now have P101 analogs that are 10X more potent and 10X more metabolically stable that are going back into mice for a PK/PD study before the next efficacy study. We have examples of enantiomeric pairs that differ in potency by 10-fold, which is consistent with the overall non-flatness of the SAR, i.e., P101 is binding to a specific site on a protein. We can’t yet make a statement about binding selectivity. We’ll just have to wait for the results of planned chemical proteomics experiments. Bit by bit, we’re triangulating the target.
The typical discovery phase can last up to four years and costs over $5M. One of our main scientific objectives was to see how quickly and cheaply we could advance an unoptimized primary screening hit from a worm screen to patient cell and mouse validation. By nearly all counts, we succeeded in halving the time and cost of the discovery phase. A year from now, we plan on halving it again.
Stay tuned for the fifth chapter of the P101 quest in Q2 2018..
