This post is dedicated to World Rare Disease Day (WRDD) 2017. It was at the WRRD four years earlier when the value proposition for Perlara crystalized in my mind. I went to NIH that frosty last day of February in order to meet with patient advocates and ask them one question: are there yeast and/or worm and/or fly and/or fish models of your disease? It quickly became clear from the answers I received that “animal models” really only meant mouse models, and that human cell-based phenotypic screens were the strongly preferred if not exclusive model system for drug discovery. There was a glaring disease model deficit: many phylogenetic gaps to fill in between human cells and mice.


It wasn’t until we discovered PERL101 for Niemann-Pick C/NPC disease 18 months ago and then put the compound through its paces in worms, human cells and mice in the intervening period that we accumulated evidence — albeit n=1 — for what in early 2013 was just a strengthening hunch. Namely, hits from whole-organism phenotypic screens are selected not only for efficacy (as best as that can be modeled in any given disease model), but also for all the other non-negotiable properties we require for a drug to be a pill: solubility, metabolic stability, oral bioavailability, brain penetrability (for CNS diseases) and tolerability, to name the big ones. None of those pharmaceutical properties can be selected for in target-based in vitro screens and only a few of them can be selected for in cell-based screens, which is why hit compounds traditionally require the immediate time and expense of medicinal chemists.


The first real make-or-break test for just how far an un-optimized primary screening hit could run the preclinical gauntlet came when we assayed PERL101 in a mouse liver microsomal stability assay. As I blogged previously here, here and here, (and will blog again soon), one by one thereafter PERL101 cleared each of the subsequent early-stage hurdles, culminating in a 90-day maximum tolerated dose study and a multi-dose PK study in wild-type mice. Earlier this month, we presented PERL101 data at a talk. During the Q&A, an experienced drug hunter at a large pharma approached the mic first and was baffled that PERL101 could belong to an unbiased compound library. He insisted that its pharmaceutical properties were too good for a random library compound.


Which of course raises the original question again: is PERL101 an n=1 outlier? Is there another example in the literature of an un-optimized primary screening hit discovered in a whole-animal phenotypic screen that was advanced to mouse validation without any lead optimization? Science Twitter to the rescue!


This paper describes a cell-based phenotypic screen to identify compounds that rescue erythropoiesis/erythroid differentiation in Diamond-Blackfan Anemia (DBA) induced pluripotent stem cells followed by hit validation studies in zebrafish and mouse models. DBA is caused by mutations in ribosomal proteins genes (RPS19 and RPL5, for example), and results in a loss of red blood cell production. The single best hit from their repurposing screen is a compound called SMER28, which is near and dear to my heart as I’ll shortly explain. Incredibly, SMER28 rescues anemia in the fish and mouse models, as shown here:


SMER28 fish mouse rescue


So, where the heck did SMER28 come from? I first discovered SMER28 and a bevy of other autophagy modifying compounds in 2003, the second year of my graduate thesis research in Stuart Schreiber’s lab. Since we published the original SMER paper a decade ago in 2007, several groups showed that SMER28, one of the first small-molecule inducers of autophagy that wasn’t rapamycin, could clear autophagic substrates in several cellular models of disease (here and here). But no one had shown that SMER28 was active in any animal more complex than a fly, as shown in these data from our 2007 paper:


SMER28 fly


What the DBA paper failed to appreciate is that SMER28 is an un-optimized primary screening hit from a yeast cell-based phenotypic screen for chemical modifiers of the cytostatic effects of rapamycin. In the paper, they (under)state: “SMER28 was first identified as a mechanistic target of rapamycin (mTOR)-independent inducer of autophagy in models of neurodegeneration.” So SMER28 is the second known example of an un-optimized primary screening hit translating from a simple animal model to a mouse model. And in this case the simple animal model is yeast, the simplest possible animal: the cell is the organism. To my chagrin at the time, the original yeast screening data was abridged and dumped into Supplementary Data. I’ll wrap up this post by resurrecting that data here.


smer28 supp 1


As shown above in 1a above, the SMER (Small-Molecule Enhancer of Rapamycin) hit rate is < 0.1%. Recall that the hit rate for PERL compounds (hits from the primary NPC1 KO nematode screen that also enhance cholesterol uptake and mobilization in NPC1 patient fibroblasts) is also < 0.1%. I would posit that a very low hit rate is a feature of whole-animal phenotypic screens. There are two other notable observations. First, SMER28 is not very potent at all in yeast, with an EC50 greater than 50µM, yet remarkably in the DBA paper, they observed SMER28 activity in zebrafish at 1µM.  Second, 1d shows that SMER28 is extremely selective and only enhances rapamcyin’s fungistatic effects.


smer28 supp 2


I could only get so far profiling SMERs in yeast. So I reached out to David Rubinsztein, an expert in autophagy and neurodegenerative diseases, to see if his lab would be interested in testing SMERs (and SMIRs, or Small-Molecule Inhibitors of Rapamycin) in human cells in an autophagic substrate clearance assay. My hypothesis was that some enhancers of rapamycin as single agents would induce autophagy with a better therapeutic index than rapamycin. As show above in 2c and 2d, only three SMERs induced clearance of mutant alpha-synuclein in human cells, and SMER28 is the best performer.

The whole-animal phenotypic screening mantra goes something like: way fewer hits, much higher quality.

Translate »