But I couldn’t understand how that could be when yeast cells were overdosing on antidepressants like Zoloft, a selective serotonin reuptake inhibitor (SSRI) whose seemingly sole drug target — the serotonin transporter protein — isn’t even present in yeast. Single-celled yeast don’t produce serotonin let alone have a brain, so what’s at the root of Zoloft overdose? Conflating non-specific and uninteresting doesn’t make sense from a mechanistic standpoint, because there are many ways to kill a cell, but it’s a matter opinion as to whether these mechanisms are deemed worthy of study.

So in the Fall of 2007, I let genetics be the arbiter: I and Meredith Rainey McClure, my first technician at Princeton, selected for yeast mutants that are resistant to Zoloft overdose. If the molecular basis of overdose were non-specific, resistance-conferring mutations would occur in genes that are involved in drug removal or detoxification, pathways that would also confer resistance to drug overdose in general, without regard for chemical structure. On the other hand, if the mechanism of overdose were specific, resistance-conferring mutations would occur in genes that did not modify cellular responses to just any drug.

As shown in this table from my Princeton lab’s 2010 Genetics paper, we observed resistance-conferring mutations that fall into two classes:

The first class includes CUP5, TFP1 and VMA9, all components of the V-ATPase complex (below, left). Mutations in the V-ATPase affect drug response by a mechanism that was first elucidated decades ago by Christian de Duve, who discovered the waste management facility of the eukaryotic cell called the lysosome (vacuole in yeast), and after which the drug accumulation known as lysosomotropism is named. I blogged last summer about the second class, SWA2 and CHC1, the latter encoding for a protein called clathrin. Clathrin molecules assemble in to the buckyball structures (below, right) that encapsulate lipid vesicles, including the vesicles that store neurotransmitters like serotonin in the synapse:

As we argued in the Genetics paper, the pattern of Zoloft overdose-resistant mutations was consistent with specific mechanisms that modify the accumulation of drugs called cationic amphipaths, or in slightly friendlier terms, hydrophobic weak bases. So what about overdoses caused by other psychoactive cationic amphipaths? In unpublished work that we’ll be submitting soon, it turns out that the genetic basis of psychoactive drug-overdose resistance is complex in yeast, such that distinct pharmacological classes like antipsychotics and antihistamines exhibit more than one resistance mechanism.

In 2007-2008, the state of the art in mutation detection (“genotyping”) were whole-genome yeast tiling arrays from Affymetrix. (Remember them?) Those arrays cost $150 a pop, and that was with an institutional discount to Princeton researchers, who originally helped design and validate the technology, and as far as I can tell were the only people ordering these arrays. However since 2008, mutation detection has advanced to the point where one can sequence an entire 12 million base pair yeast genome 30-times over for $15 or less, so I myself could in theory do in my second bedroom in a few weeks what it once took a full-time technician six months at more than 10-times the cost. But I’m not sure my wife is too keen on the idea of an experimental Man Cave just yet, so I’m on the hunt for kosher lab space.

If I had gotten a job as an assistant professor, I would have done exactly these experiments right out of the gate. Here’s what I wrote in the research statement that was part of my application package:

“I propose to generalize yeast-based genetic resistance selections that made possible Project 1 to other antidepressants, as well as to three other pharmacological classes bedeviled by molecular promiscuity: antipsychotics (both typical and atypical), antihistamines (both sedating and non-sedating), and anesthetics. Using whole-genome sequencing technologies, robotic automation and miniaturization, it will be possible to perform many single, or even combination, de novo drug-resistance selections in parallel. Follow up investigation of specific conserved cellular pathways revealed by this approach will be guided by the template established by my experience with sertraline, including the use of radiolabeled analogs to validate novel targets or target pathways.”

As I blogged earlier this year, no dice. At least one of the departments I applied had the professional courtesy to actually explain albeit in a terse, unofficial email of rejection why I didn’t make the cut: “one of the issues was the relevance of understanding how neuro drugs like antidepressants work by studying their action in non-mammalian organisms.” I obviously couldn’t disagree more strongly with this short-sighted view, which shows just how far pharmacology has drifted away from the tenets of cell biology and genetics. Eroom’s Law and the wholesale abandonment of CNS drug pipelines by Big Pharma is proof that we don’t have enough basic understanding of how existing drugs work. So how on Earth are we supposed to create new first-in-class drugs for common or rare diseases without the basic understanding that comes from studying simple genetic model organisms like yeast?

So here’s my plan. Before, I was interested in studying individual drug-resistant mutants, but as I’ll explain I now want to take a statistical approach by pooling all resistant colonies from a drug-laden agar plate, like this one:

Each of these colonies sprouts from a single spontaneous mutant. In fact, some of these colonies are clones, so the same mutation may be represented on the plate in multiple copies. And as the above Genetics table shows, en toto there are multiple genes with resistance-conferring mutations. So is there a way in one fell swoop to identify all mutations? I plan to scrape all mutants off a plate like the one above into a gene pool, literally; then extract genomic DNA and outsource sequencing using a site like Science Exchange. These pooled whole-genome datasets would yield the entire distribution of resistance-conferring mutations for each tested drug, with a given mutation present at a frequency in the pool that is proportional to the number and size of colonies harboring it.

I’d start with Zoloft as the baseline, since I have a concrete expectation of the genes that are likely to pop up. For example, I expect that V-ATPase mutants will be common across many resistance selections. The downside of this distributional approach is that I lose information like whether a mutation is recessive or dominant, or even whether resistant colonies harbor one or more than one resistance-conferring mutations. And I wouldn’t be able to study mutants individually. But using yeast strain construction techniques, I could after the fact regenerate specific mutants to test in secondary assays of cellular physiology.

While I figure out where to do these experiments, which in theory could take a few weeks if all goes without a hitch, I’d be interested in finding a computational collaborator who would help me with variant calling from pools of drug-resistant mutants. It would also be helpful to do some simulations before any physical experiments are done to get a sense of sensitivity. In other words, if there are 100 mutant colonies on a plate ranging in size, would this approach be able to detect the rarest mutant? What is the minimum genome coverage required to identify all mutants on a plate? How many independent drug-laden plate replicates would be needed to distinguish in a statistically significant way between close structural analogs, for example between a secondary and tertiary amine tricyclic?

Think of this post as an experiment in open review of a pre-experiment hypothesis. Keen to hear your thoughts!

There are over 7,000 rare diseases, which are almost all caused by having two mutant copies of a single gene. En toto rare diseases affect 10% of the population, so 250-300 million people worldwide. Remarkably, many rare disease-causing genes are conserved in all eukaryotes. Using copiously studied genetic model organisms — yeast, worms, flies and zebrafish — for rare disease drug discovery may have its greatest impact on brain diseases, whose underlying mechanisms are still poorly understood. For example, a tsunami of neurodegeneration is massing over the demographic horizon, and we desperately need genetic illumination. And the cherry on top of the sundae of conservation is that common diseases can be modeled by rare diseases.

Besides scientific curiosity and idealism, there are practical reasons to choose rare disease drug discovery. For one, there’s expedited FDA review under the Orphan Drug Act, and a more recent “breakthrough” track, which decrease the time to translate discoveries from the bench to bottles in the pharmacy. Related to clinical trials, the drug repurposing fad — finding new indications for old drugs — dovetails nicely with rare disease drug discovery because the FDA pharmacopeia has plenty of surprises up its sleeve that can skip safety trials (Phase 1) and go straight to efficacy trials (Phase 2 and 3).

The big questions are how to fund and staff this rare disease drug discovery enterprise, and of course where to house it all. I am, after all, an experimentalist. And being a recent transplant, I also happen to be looking for a place to live, so I’ll focus on the search for lab space here, and save funding and staffing for future posts. One of the many professional reasons to relocate the Bay Area is the opportunity to access unconventional lab space outside Academia and Pharma. It looks like I have two ready-to-go options: 1) lease a bench from a life science startup incubator, or 2) join a community lab, sometimes called a biohacker space. There are also more creative options, like carving out a niche in a local academic lab, but the regulatory and administrative hurdles would be non-trivial even if I were to find a magnanimous host.

As of last count, my new home boasts seven sites where lab space can be leased on a monthly basis. Five of these sites are life science startup incubators in the QB3 network, which promotes and advances translational research from the Bay Area UC schools. However, as I mentioned above, there is also a bottom-up alternative sprouting up around the Bay Area, as well as in other cities. Biohacker spaces like BioCurious in Sunnyvale and GenSpace in Brooklyn level the playing field with tech startups and hackathons without compromising safety or reproducibility, and even allow independent scientists to commercialize their results. I’ve been introduced to members of the BioCurious team and I plan to drop in soon.

So, it’s between biotech incubator vs biohacker space. Or is it? A question I keep asking myself for the near-term is: do I need one central lab space, or can I manage projects that are distributed across research sites with unique communal resources? Evidence that the distributed model works is the Crowd4Discovery project, which is currently underway in the Sulzer Lab in New York City while I manage the project remotely. Of course one could also leverage websites like Science Exchange or Assay Depot to source experiments from third-party providers, the way Atul Butte of Stanford Med and others have been doing to great effect.

In these situations, I do what I’ve always done: strategic cold emailing. I scoured the QB3 site for contact@XYZstartup email addresses and blasted out 20 messages; almost half replied within 72 hours. From the subsequent Skype/phone chats and in-person meetings, I gleaned the distribution of startups at QB3: a good mix of companies founded by serial entrepreneurs as well as n00bs. Many of the representatives I spoke to had actually graduated from QB3, which as the name incubator implies only nurtures companies with 5 people or less, after which point they’re ready to leave the nest (or plummet perilously to the ground).

I also got a tour of the East Bay Innovation Center, which is the newest of the QB3 incubator sites. Turns out this space is sublet from an independent commercial entity, as is another QB3 incubator site. Some equipment is communal, like the fridges and freezers and the occasional incubator or shaker. Several of the QB3 folks I chatted with affirmed the positive community ethos. The catch is QB3 costs $900 per bench per month. For some sense of scale, my old lab at Princeton was comprised of 9 benches as defined by QB3, so it would cost $81,000/year to recapitulate a comparable footprint. For the uninitiated, here’s what a typical bench in a basic biomedical research lab looks like:

Lots to think about. If anyone has any other suggestions about where and how to reconstitute Perlstein Lab, I’m all ears. Per usual, the comment thread is open.

Then we hit some walls. First, the most recent batches of VM cultures have not been healthy, and so it was impossible for us to interpret amphetamine vs no drug treatment of VM cultures used by Danny in his initial experiments. But as of last week, Danny has switched to primary neuronal culture cells derived from a different mouse brain region called the striatum, and these new cultures appear to be healthier. (For most people who don’t have first-hand experience with mammalian cell culture techniques, these kinds of technical difficulties happen).

Regulatory delays created additional headwinds. By mid-March, we had worked out protocols to store and experiment with radioactive amphetamines (labeled with tritium or carbon-14), but ordering these compounds has been slowed by issues on the provider’s end, in this case American Radiolabeled Chemicals (ARC). Since amphetamines are controlled substances, we’re now waiting on the DEA to sort things out with ARC.

Unrelated to research, we’ve also had technical issues with our MakerBot Replicator2 printer. We’ve experienced software and hardware hiccups with 3D-printed methamphetamine production. In response, Danny has streamlined the 3D meth model so that it consumes less polylactic acid (PLA), and as a result the models are now smaller but much smoother. We’ve delivered around half of the total orders. That means another 100 models left to deliver. Thankfully, our 3D printer is up and running again, but keep in mind that each model takes roughly two hours to print. We apologize to the delay. As an excuse to hear feedback directly from C4D supporters, I’ve been hand delivering 3D meth to as many donors as possible. There were enough C4D supporters in the Northeast to keep me busy all winter. I’ve gotten great individual feedback, and comment threads around here are always open.

We are also beta testing G+ On Air hangouts as our version of an online lab meeting. The first C4D hangout will include two supporters who know a thing or two about pharmacology and wrote about us during the $25k fall campaign. I invite any C4D supporter who wants to participate in a future C4D hangout to email me at info@perlsteinlab.com.

I don’t need to see Hunger Games — I have a non-tenure track position in academia.

— Ethan Perlstein (@eperlste) March 29, 2012

It was the second-most popular tweet in my 2+ years on Twitter. Then two months ago, once it became clear that my second go-around on the academic job market would be fruitless, I penned a farewell letter to Academia entitled Postdocalypse Now. To my surprise the post went viral, garnering ~9,000 pageviews and over 60 comments. By far it was the best showing of any content I’ve posted on my lab website since it launched last summer.

Now that the #postdocalypse surge has receded, I offer this postscript at 36,000 feet as I hurtle through the atmosphere on a one-way voyage from New York City to San Francisco. I’m embarking on the next phase of my professional evolution as an independent scientist, leaving the Academia-Pharma Complex behind. I still experience occasional pangs for the professorial fantasy I’d clung to since I was 17, but the withdrawal is gradually giving way to an emboldened optimism.

That’s not to say that there aren’t days that sting, like when I discovered that I have my own trolls who seem to think that I’m a malcontent or a failure. The fact that this invective against my character and productivity was made by peers hiding behind a pseudonym was especially disheartening, and it really hurt the first time I read it. But we all know that the second phase of a troll infestation is sheer amusement at how ridiculous haters sound. Thankfully, like any rash, the final phase is subsiding irritation.

Just last week, an unexpected source of inspiration flitted across my Twitter feed, reaffirming my declaration of scientific independence. Professor Henry Bourne of UCSF, whose lab famously studied G protein coupled receptors, wrote a courageous, no bullshit commentary about The Tenure Games. The one part of Bourne’s jeremiad that really bothered me was when he stated that many biology PhDs abandon Science altogether when they can’t break into an academic job. We still seem to be stuck in a pre-Internet mindset where basic biomedical research only happens in Academia, and applied biomedical research only happens in Industry – and never the ‘twain shall meet.

Moving forward I will continue to argue the case for a hybrid evolutionary approach to drug discovery that combines basic and applied research, and more to the point, I will try to lead by example. One of the reasons why I blog about my scientific journey is to leave a digital trail for the next generation of scientists who are scared shitless or simply turned off by all the hyper-competition, granstmanship and academic balkanization.

Declare your scientific independence with me, and starting small, we will forge a third way.

My plan is to be an independent drug discoverer with an evolutionary pharmacology twist. Some might simply call this a flavor of life science entrepreneur. Apparently, I still struggle with the academic tendency to rechristen an old concept with a new name. Whatever you want to call me, how do I get from here to there?

As part of my self-directed crash course in bioentrepreneurship, I’ve been scanning the Burrill Report podcast library for interesting case studies of hybrid business models that are also capital-efficient, or lean in the startup argot. Hybrid refers to combining fee-for-service work with an internal R&D pipeline. The episode entitled “A Proof-of-Concept For Fast Exit Strategy” piqued my interest. Podcast host Daniel Levine interviewed Jim Posada, founder of Resolve Therapeutics LLC.

Resolve Therapeutics was founded by Posada in 2010 to develop a new kind of treatment for lupus, an autoimmune disease associated with chronic inflammation. Its single-asset value proposition is a genetically engineered fusion molecule made of two different proteins: the business end, a nuclease that shreds strands of DNA (or RNA) into nucleotide bits, is attached to a Fc domain that targets immune complexes (indicated below by the orange arrows). According to lupus research, inflammatory responses are triggered by these immune complexes, tangles that form in the kidney and other tissues when rogue antibodies glom onto nuclear proteins, DNA, RNA, and related cellular debris that is normally disposed of without incident.

From: Demonstration of Immune Complex Deposits Using Fluorescence Microscopy of Hematoxylin and Eosin–Stained Sections of Hollande’s Fixed Renal Biopsies. James T McMahon, Jonathan L Myles and Raymond R Tubbs

The idea for a Fc-nuclease fusion protein originally came from academic co-inventors Jeffrey Ledbetter and Kevin Elkon. Jeffrey Ledbetter is a veteran of pharmaceutical R&D departments of yore; the glory days of the 1970s and 1980s, when groups headed by pharmacologists like David Wong at Eli Lilly spearheaded the creation of new first-in-class drugs, e.g., the SSRI antidepressant Prozac. Jim Posada, himself a one-time academic researcher turned life science entrepreneur, licensed the Fc-nuclease technology from the University of Washington with one objective: to partner with an established Pharma company to take the Fc-nuclease technology all the way to the clinic.

Keep in mind that Resolve isn’t the only one pursuing a protein-based experimental therapy for lupus. For example, Genentech has developed a monoclonal antibody that neutralizes interferon alpha, a small protein that is released by immune cells during inflammatory responses. The rationale for taking out interferons, which are elevated in lupus, is straightforward.

Resolve Therapeutics will be successful if they can score positively in a biomarker assay of interferon activity, the first benchmark of efficacy in lupus trials. As of last month, they’re well on their way. Resolve only needed to spend $3M to reach their partnering event with Takeda, a top-15 Pharma company based in Japan. Takeda gave Resolve $8M upfront, and then promised to shell out up to $247M in milestone payments at the end of Phase IIa.

Before I did any experiment as an academic, I had my due diligence ritual. Basically, it entailed scouring PubMed for hours a day, weeks on end, assimilating every last datum related to my inquiry. Now as a budding independent scientist, my challenge is not only scientific but economic, because I don’t have an employer who will guarantee me salary and benefits and a plum startup package. I need to develop a viable business model to support my evolutionary approach to drug discovery. What I’ll be doing over the weeks ahead is more of these case studies, finding examples of people and companies who have done something similar, or are simply sources of inspiration.

I was fortunate to meet and interact with several rare disease parents, mostly Moms. Lori Sames was introduced to me by my friend Sean Ekins, a proponent and practitioner of Open Science Drug Discovery. Lori and her husband live in Albany NY. Their daughter Hannah has Giant Axonal Neuropathy, or GAN. Lori is executive director of Hannah’s Hope Fund, which would raise as much as $1,000,000 this Spring if an all-or-nothing challenge grant were matched. The three of us converged on a greasy spoon in downtown Bethesda the night before the meeting. In a whirlwind tutorial over mozzarella sticks and 7-Up, Lori breathlessly explained everything known to humankind about gigaxonin, the damaged protein underlying GAN, which is usually but not always associated with curly hair.

Lori told me she had studied economics and it showed. She rattled off obscure sounding gene names just as quickly as she remarked on VCs, CROs and INDs. It was almost like watching an entrepreneur crushing it on Shark Tank, except instead of pitching some random startup idea, Lori was effortlessly discussing how to bring new GAN treatments and cures to market.

Later at the meeting itself, I sought out the parents of kids with one of the ~50 lysosomal storage diseases (LSDs), of which Tay-Sachs disease may be the most well known. My affinity for LSDs stems from my lab’s work on vacuoles, the yeast version of lysosomes. I met parents just like Lori. Patty Burkholder Taormino’s son has Sanfilippo Syndrome and is part of group called Team Sanfilippo. Melissa Hogan’s son has Hunter Syndrome, another mucopolysaccharidosis, and her foundation is called Saving Case. And then I met the Hempels of Nevada, whose twin daughters have Niemann-Pick Disease. The meeting organizers screened a poignant documentary about the Hempels called Here. Us. Now.

So how we deploy this amazing human capital to cure diseases faster?

I want to build rare disease research into my hybrid drug discovery business model. I went to #RDD13 specifically to talk to parents, researchers and administrators about the use of simple genetic models in rare disease research. Three of the most well-studied genetic model organisms are brewer’s yeast (Saccharomyces cerevisiae), nematode worms (Caenorhabditis elegans) and fruit flies (Drosophila melanogaster):

I was very surprised to find out that gene therapy, stem cells, and in some case enzyme replacement, were the only therapeutic pipelines people were talking about. Seems to me that taking an evolutionary approach to drug discovery is long overdue. And based on work from Susan Lindquist’s lab or Aaron Gitler’s lab or Stephen Sturley’s lab, I’m not the only one with this idea. Turns out yeast cells can teach us about human diseases like Parkinson’s or ALS even though they lack brains.

In terms of the science, February was about the appetizers leading up to the experimental main course: autoradiography. Danny has prepared the first draft of an autoradiography protocol, which we will continue to flesh out. Protocols, raw data and ultimately processed data generated by the C4D project can be viewed and downloaded on Figshare, our open data repository of choice; all project-related content will be tagged “Crowd4Discovery.” And in the spirit of Open Science, we will be soliciting feedback from experts on comment threads and in G+ hangouts. (Skype dates for C4D donors are also available upon request!)

In this update, I will discuss experimental appetizers. Since the beginning of last month, Danny has been working on experiments that leverage published experimental observations of where hydrophobic weak base drugs – a.k.a. cationic amphipaths – accumulate inside cells. Using vital stains that localize to specific compartments inside drug-treated versus untreated primary neuronal cell cultures derived from mouse ventral midbrain, we hope to gain preliminary insights into where amphetamines accumulate.

Recent work from Professor Jeff Krise’s lab at the University of Kansas shows that human fibroblast cell lines treated with psychoactive cationic amphipaths for  extended periods have greater capacity to take up a vital dye that collects in lysosomes. In other words, drug-treated cells adapt to drug exposure by increasing their total lysosomal volume. For the cognoscenti, please see Krise’s paper here.

In mid-February we “figshared” some raw data from vital dye labeling experiments with untreated versus amphetamine-treated cells. Danny stained for an autophagy protein called LC3 and a pH-sensitive dye called pHrodo, which labels acidic compartments called endosomes (the red puncta in the featured image above). There are many reasons why we care about autophagy and its evolutionarily conserved role as a cellular drug response. Based on the yeast work my lab did at Princeton, we know that cationic amphipaths related to amphetamines, namely antidepressants, trigger autophagy.

And that’s where the multilamellar bodies (MLBs) I discussed in the late January C4D update come in. MLBs are a mystery. Are they pit stops on the autophagy pathway? Are MLBs structural variants of other compartments, like endosomes? We suspect that MLBs form independently of autophagy, though they may ultimately be substrates for autophagic clearance. Under this scenario, MLBs are altered compartments formed by asymmetric drug accumulation in organellar membranes — basically the bilayer couple model extended to acidic intracellular compartments. To test the role of autophagy in MLB formation in response to drug treatment, we will eventually take advantage of an ATG7 mutant mouse created in the Sulzer Lab that specifically lacks the ability to perform autophagy but only in dopamine-positive cells that express DAT, which is the transporter that actively concentrates amphetamines.

Based on preliminary observations thus far, there doesn’t appear to be co-localization between LC3 and pHrodo, but Danny is currently doing this experiment more carefully with a confocal rig, and with quantification of co-localization. One interesting possibility is MLBs may contain amphetamines and LC3, but don’t stain with pHrodo because amphetamine is a weak base that neutralizes acidic compartments in which it accumulates. To explore this possibility, we will examine co-localization between LC3 and another vital dye called monodansylcadaverine, or MDC, which is known to accumulate in “autophagic vacuoles.” These experiments are now underway, so more on the autophagy story in the weeks ahead.

Finally, we will be ordering radioactive amphetamines soon, at which point we’ll have much more to say about autoradiography. And that will also be a good time to post our first open budget report, because so far our only expenditure has been on the MakerBot Replicator2 3D printer we purchased in mid-December.

The people who inherit two broken copies of a single gene called CFTR get cystic fibrosis. The CFTR gene was discovered (“cloned”) in 1989 at a cost of almost $200M. But it took 23 years for the first genuine breakthrough drug called Kalydeco/ivacaftor to be developed for only a subset of people with cystic fibrosis (CF) harboring a particular mutant form of the CFTR gene called the Celtic allele. For some perspective, the discovery of CFTR predates my former students by several years, many of whom weren’t even aware of the disease.

I opened the slide deck with a karyotype. Most students quickly realized that it yielded several clues about the person who supplied these chromosomes, namely this person is a male with Down Syndrome:

But cystic fibrosis isn’t caused by having an extra chromosome, which is easy to spot in karyotype. The mutations that cause CF are tucked away in a single gene called CFTR, which encodes a chloride ion transporter protein that is expressed in the epithelial tissue of the lungs. Some day we may be able to visualize point mutations microscopically with some future version of the iPhone, but we can’t today and geneticists weren’t able to 25 years ago.

The recombinant DNA revolution that swept across genetics in the 1970s presented a solution to the problem of finding a needle (a mutation) in a haystack (a chromosome) of haystacks (a genome, or 46 chromosomes). David Botstein and colleagues published a wonderful paper that’s been cited almost 6,000 times on a set of molecular techniques that now seem quaint in comparison to next-gen sequencers. Restriction fragment length polymorphisms (RFLPs) are like fingerprint patterns that, in the limit, distinguish any two people in the world:

With RFLPs, geneticists could link specific banding patterns with a disease in sick people. But back in the day, it wasn’t easy to shuffle around chromosomes, which had to be propagated and amplified by living cells and then chemically extracted and manipulated in the lab. Somatic cell hybrids presented a crude albeit effective way to isolate chromosomes of human DNA against a backdrop of non-human genome, e.g., mouse, in search of the one chromosome that contained polymorphisms linked to cystic fibrosis.

And so the hunt began. By 1985, the year Michael J. Fox was lighting up the silver screen in Back to the Future, the CF-causing gene was tracked down to a chromosome. But there isn’t just one singular CF-causing mutation. In fact, an individual with CF may not even have inherited the same exact broken version of CFTR from both parents. Turns out the majority of people with CF have a mutant allele of CFTR in which 3 DNA bases are deleted, ∆F508. Here’s the original table from the paper proving that a polymorphism called DOCRI-917, which is present in people with CF, resides on chromosome 7:

How exactly do CFTR mutations break the CFTR protein? CFTR like many channels and transporters, resides in the plasma membrane, or at the cell surface. But it isn’t created there, rather it is synthesized inside the cell and must then wend its way to the cell surface via vesicular transport pathways, which shuttle cellular components from one location to another inside tiny compartments called vesicles. Mutant CFTR can’t make its way to the surface, where they normally accumulate:

This gave researchers the idea that if they could find “pharmacological chaperones” that would suppress the misfolding effects of the mutation on the CFTR protein structure, then CFTR would survive the quality control system and make its way to the cell surface. This idea bore fruit with Kalydeco:

Right now clinical trials are underway with Kalydeco and another compound that may restore the function of ∆F508, which in the US is half of the people with CF.

I previously blogged about two of a favorite precept puzzles. One features a classic paper describing one my beloved natural products called rapamycin. The solution is a model that explains how rapamycin inhibits yeast cell growth. The other precept puzzle features a paper on an often-misunderstood concept from genetics called epistasis. The solution is a model that explains how a group of genes regulate each other’s activity in the mustard weed, experimental biology’s favorite plant.

I will gradually post content from my Princeton teaching archive over the next few months. In this first tranche, I’ve uploaded to Slideshare two decks from the population genetics module of ISC236, which is the Spring course. The first deck is about cystic fibrosis, and the second slide deck is about the group of rare diseases called lysosomal storage disorders, for example Tay-Sachs disease.

This weekend, I will annotate each of these decks for posterity in two separate posts on my lab website, so please stay tuned..

A conserved ethylamine side chain (in red) pervades the psychopharmacopoeia

Discovered in the early 1950s, imipramine is the original tricyclic antidepressant, and it begat second- and third-generation antidepressants with fewer side effects but not necessarily greater efficacy, including the selective-serotonin reuptake inhibitors (SSRIs): Zoloft, Prozac, and Paxil, among others.

So how do antidepressants work at a molecular level? Which cellular bits do they bind to? You’d think we’d know the answer to that question after more than 50 years of antidepressant use. As suggested by the name SSRI, antidepressants inhibit with different degrees of selectivity the reuptake of monoamine neurotransmitter transporter proteins at synapses in the brain. Neurotransmitter transporter proteins include SERT, which is specific for serotonin, or 5-hydroxytryptamine (5-HT). Evolution has conserved the SERT gene in mammals, flies and even nematode worms. The human version of SERT is thought to be the primary drug target of SSRI antidepressants. In other words, SSRIs bind to SERT expressed on the surface of serotonin-producing neurons and then inhibit serotonin reuptake into cells.

But there’s a catch. The serotonin-centric model has an exposed, soft underbelly. In the mid-90s, psychopharmacologists Eric Nestler and Ronald Duman and others showed that antidepressants induce neurogenesis in laboratory rodents after chronic but not acute treatment. The chronic vs acute part is key. Inhibition of SERT by antidepressants occurs within minutes to hours, yet people who respond to antidepressants don’t feel better until they’ve been taking them for weeks or months. So the rodent observation meshes with long-standing clinical observations in people taking antidepressants who experience a “therapeutic lag.”

For the record, acute is defined as 30 minutes after injection of antidepressant. One of the behavioral assays employed in the antidepressant R&D pipeline is the tail suspension test. Briefly, mice are suspended upside down by their tails. As you or I would if subjected to a comparable stress position, inverted mice struggle to right themselves but eventually submit to gravity. Turns out mice treated with antidepressants fidget longer than their placebo-treated compatriots. This difference between treated and untreated mice has traditionally been the green light to move a candidate antidepressant onto human clinical trials.

In 2011, Randy Blakely’s lab at Vanderbilt published an insightful paper in PNAS that informs any serious discussion of antidepressant pharmacology, which is clearly complex. The Blakely Lab has a unique place in the history of antidepressant research. Blakely led the team that first cloned SERT from rats in 1991, validating the crude extract-based neurotransmitter reuptake assays that were originally employed by the pharmaceutical industry in the 1960s and 1970s in R&D and that ultimately yielded the SSRIs. With the 2011 PNAS paper, Blakely’s efforts came full circle. He and his group found that acute behavioral effects of antidepressants did not occur in a special SERT mutant mouse they genetically engineered.

This mutant mouse’s SERT gene was mutated at a single amino acid position in the SERT protein. As a result, this one mutation rendered the mouse SERT protein unable to be bound by many antidepressants but still able to bind and transport serotonin, its natural substrate. As such, this mSERT mutant mouse is a clean genetic test of the role of serotonin reuptake in the complex pharmacology of antidepressants.

As shown in Figure 1A of their paper, mutant SERT (M172) interacts with and transports serotonin, or 5-HT, just as well as wildtype (I172). However, several SSRI antidepressants (but interestingly not paroxetine/Paxil), as well as cocaine, must be used at higher concentrations to block serotonin transport by mutant SERT. In one case (E),  up to 1000-times more citalopram/Celexa is required:

The single amino acid change, which swapped an isoleucine for a methionine at position 172, was not made by serendipity but on the basis of data in a 2007 Nature paper by Eric Gouaux, whose lab is at Oregon Health & Science University. The Gouaux Lab studies neurotransmitter transporter proteins. However, the Gouaux Lab couldn’t crystallize a SERT protein of mammalian origin, but they were able to crystallize an old relative of mammalian SERT – from bacteria! Bacteria don’t produce serotonin, but they express a distantly related yet structurally similar transporter that recognizes the amino acid leucine (see featured image).

Blakely’s group has demonstrated that SERT is required for the acute behavioral effects of antidepressants in laboratory rodents. So here’s the $64,000 question: is SERT required for neurogenesis in laboratory rodents after chronic antidepressant treatment?

Two years later this experiment still hasn’t been performed to my knowledge — what are we waiting for?

I could place my fantasy on life support as a second postdoc or claw my way into a tenure-track position through an adjunct appointment, but in my case these moves would be a step back. I just spent the last five years managing a $1M budget and a two-person lab while also teaching, and published two papers on novel insights into how antidepressants actually work. At age 33, with plans to start a family, and a desire for – gasp! – life/work balance, entering the equivalent of a professional holding pattern offers little appeal or dignity.

So how the hell did I not foresee this outcome? Like so many of my contemporaries, I’ve been a contestant in the Tenure Games since I was a teenage summer research intern, plenty of time to see the writing on the wall. Most PhDs don’t become professors. But I was in denial. At each transition in my academic career I watched people drop off, and the refrain in my head was “I’ll be different.”

For a time that was true until one day it wasn’t.

To be fair to myself and to the scores of other rejected postdocs, the Tenure Games were not of our making. The National Institutes of Health (NIH) budget doubled from $15B to $30B in the late 90s and early 00s:

People and projects followed in droves, but the capacity to absorb trainees into stable academic careers didn’t increase proportionally, resulting a predictable glut of postdocs that was exacerbated by the Great Recession.

I was loath to admit all of this publicly till now, partly out of insecurity and partly out of academic self-preservation. The latter is no longer a concern. Engaging the online scientific community on Twitter has helped me overcome my insecurity, because it provided concrete evidence for the first time in my training that I wasn’t alone.

If you’re like me, there’s no reason to be ashamed. Almost every single assistant professor I know has admitted that it was dumb luck, idiosyncratic departmental tastes or plain old academic tribalism that landed them their job, because they all had impressive CVs, stellar recommendations and solid proposals.

So what’s next? As the shell shock begins to wear off and more and more thwarted postdocs emerge from their bunkers, I hope we can take comfort and inspiration from each other by sharing our journeys. Younger trainees can benefit from our peer-to-peer mentorship. And practically speaking, we can start to mobilize and brainstorm new ways to do the science we love outside of traditional academic (or even industry) settings.

The C4D project will proceed in two phases. In Phase One, we will optimize autoradiography and electron microscopy for mouse brain cells that will be grown in the lab and exposed to radiolabeled amphetamines. In Phase Two, we will extend these validated protocols to examine postmortem brain samples from mice treated with radiolabeled amphetamines.

As of last week, lead project experimentalist Danny Korostyshevsky was officially on board and settling in the Sulzer Lab of Columbia University Medical School, where experiments will begin in earnest soon. Consider this post a prelude to what we, the C4D team, hope will be an open, interactive and sustained scientific colloquy in the weeks and months ahead.

Let me back up for a minute and explain why we’re so interested in determining where amphetamines accumulate at a molecular level. Conventional wisdom in psychopharmacology insists that amphetamines trigger the release of dopamine from brain cells by blocking the dopamine reuptake transporter protein, DAT, and that all the ensuing cellular effects of amphetamines, including toxicity, can be traced back to direct drug action on DAT.

However, I blogged last Fall in the run-up to our crowdfunding campaign that C4D team member David Sulzer first showed over 20 years ago that both dopamine and amphetamines accumulate inside synaptic vesicles because of a chemical attraction of positively-charged (“basic”) molecules for negatively-charged (“acidic”) environments, turning the DAT-centric model on its head. And the cellular effects of amphetamines don’t stop there.

Amphetamines are more hydrophobic than their chemical cousin dopamine, and this affinity for lipids drives drug accumulation in cell membranes. Since the 1970s, many labs, including the Sulzer Lab, showed that cultured mammalian cells or tissue slices removed from whole animals that were treated with hydrophobic weak bases like amphetamines contained abnormal structures called multilamellar bodies (MLBs), while untreated samples were free of MLBs.

My former lab at Princeton even observed MLBs in yeast cells treated with the antidepressant Zoloft, which is more hydrophobic than amphetamines, proving that these structures form in the absence of classical protein transporter drug targets, and therefore are a conserved cellular drug response exhibited by all eukaryotes:

There are two possibilities: either amphetamines are directly associated with MLBs or precursor structures that are ultimately incorporated into MLBs, or amphetamines accumulate somewhere else entirely and MLBs are a sequela of altered cellular lipid homeostasis. While we make preparations for  autoradiography, we will attempt to distinguish between those two possibilities. MLBs appear to be intermediates of the autophagy pathway, an ancient cellular recycling program that my former Princeton lab showed is activated by Zoloft accumulation in yeast cells.

The Sulzer Lab has recently described a transgenic mouse in which the autophagy pathway has been genetically disrupted only in brain cells that express dopamine and DAT, which actively concentrate amphetamines. We will see if amphetamines induce the formation of MLBs in cultured neurons derived from this autophagy mutant mouse. If the answer is yes, then autoradiography will confirm the association. But if MLBs don’t form without functioning autophagy, the drug molecules still have to be somewhere and autoradiography will reveal their hiding place.

Please stay tuned for three upcoming developments: (1) release of our first open budget report; (2) posting of autoradiography protocols online for feedback; (3) our first online lab meeting, which will be open to C4D supporters and the general public.

As always, your comments are welcome!

In a follow up post, I will speak to that 40% of Crowd4Discovery (C4D) supporters who are “out of network.” In this post, I’ll finish the preliminary analysis of the in-network 60%. During the C4D campaign, I blogged about the distribution of donors in my Facebook network. Two results jumped out at me. First, my “science” friends turned out strongly, and by science friends I mean current or former scientists whom I met over a decade of academic training. Second, donation size was more a measure of individual cash flow or personal resonance with the project, and less a reflection of the strength of our friendship per se, or where donors are located in the network.

Thanks to Tony Hirst, who kindly generated my Twitter network data files on December 3, 2012, I was able to generate the above graphical snapshot of my dynamic follower network, at the time numbering 1,315. Keep in mind that unlike my Facebook network, the above is a directed graph, which means the connections, or edges, are one-way — A follows B, but B doesn’t necessarily follow A. I’m the biggest node (in blue). Like all the network graphs I’ve visualized using Gephi, the size of each node is directly proportional its degree, or connectivity.

98% of my followers — the other 2% are orphans — comprise 6 modularity classes, or clusters, as shown below:

These clusters represents distinct online scientific communities to which I belong, with the exception of an orphans cluster (blue), which contains mostly non-scientists. I first ascertained whether C4D donors are spread out randomly across my follower network or concentrated in specific clusters:

The total turnout of my Twitter network was 10%, which is lower than the 17% turnout of my Facebook network. The distribution of donors in my follower network (right; donors in yellow and non-donors in blue) appears to be random with respect to cluster membership (left), with some clusters performing slightly above or below the network average. Interestingly, there is no correlation between the amount donated and donor degree, just as we observed no correlation in the Facebook network data:

One regard in which donors in my follower network behaved differently from donors in my Facebook network is the timing of their donations. It’s fair to say that our closing surge was stoked by a fusillade of retweets and mentions. 38 of my “tweeps” turned out on the last day of the campaign, and almost half of the 124 donors in the final 26 hours came from Twitter:

Our closing surge was more robust on Twitter than on Facebook. It appears that I had “tapped out” my Facebook network over the course of the 52-day campaign. I can think of several reasons why Twitter proved to be so pivotal in the home stretch. First, the ephemeral nature of tweets plays well with a looming deadline, and it didn’t hurt that I tweeted hourly funding updates in the final 8 hours of the campaign, adding to the sense of urgency. Second, the amplification potential on Twitter due to retweeting is higher than that of Facebook. Even though one can technically share content on Facebook, we never got much traction with Facebook sharing.

What do you think? Did we use the Twitter megaphone effectively? Is 10% Twitter turnout an underperformance, and overperformance, or right on target?

As of this writing, 116 out of 385 C4D supporters filled out the survey (but the data below is based on the 113 responses as of two weeks ago). Keep in mind that only 249 of the 385 actually opened the email containing the survey link, so our sample is not entirely representative of all C4D supporters. Let’s start with the age breakdown:

80% of respondents are age 25-44. Science crowdfunding appears to be a young person’s game. I don’t know how the C4D age distribution compares to crowdfunding projects in general. If anyone has any insights, please share them in the comments thread below. So where do these respondents live? Ours is an international coalition, though 2 out of 3 respondents are American:

We had 1 donor each from: Indonesia (an economist friend living in Jakarta); Mauritius (a science blogger I met at SpotOn London); Mexico (a scientist who follows me on Twitter); Singapore (a grad school friend); Sweden (someone I don’t know); and Switzerland (a science publisher I also met at SpotOn London). So what do these respondents do for a living? Here’s a breakdown of their professions:

During the course of our campaign, I was keeping tabs on who’s a scientist, a determination I could only make for people I knew. The informal tally suggested that the fraction of scientists was at least 25%. The working hypothesis was that scientists are doing most of the heavy lifting. The above data are consistent, and demonstrate that the just over half of respondents are scientists, which includes university professors, academic trainees (grad students and postdocs), and researchers in Pharma. Interestingly, the next largest group is comprised of people working in management positions in for-profit and non-profit organizations, typically in the science sphere. After that are the people in technology writ large, which includes IT analysts, computer programmers, and web designers. My favorite profession (n=1) is the self-described “full-time mama, part-time web geek,” a person I don’t know who heard about our campaign on Twitter.

And that leads me to the next question, which is how did our supporters hear about us? For this question respondents could select one or more of the following choices: 1) Twitter; 2) friend/family; 3) Facebook; 4) word of mouth; 5) other social media; 6) in person at an event. Here’s what we found:

In what is in my opinion the most interesting result of the survey, almost half of respondents said they heard about us on Twitter. This jibes with my own social network analyses, which show that 24% of all C4D supporters follow me on Twitter. Interestingly, 14% of responses are other social media. My strong suspicion is that includes Reddit and Hacker News. The least effective marketing strategy was in person at an event, though in fairness the only formal campaign gathering was our launch party.

Finally, we asked why supporters chose to open their wallets. Again, we allowed respondents to pick one or more answers: 1) supporting a friend; 2) supporting research in general; 3) supporting this specific project; 4) supporting new avenues of funding. Here are the results:

I was a bit surprised to find that most respondents said they were supporting research in general and new avenues of funding, rather than our specific project or a friend. My interpretation of these data is that science crowdfunding has a lot of room to grow..

Reform begins with a diagnosis of what ails us. Many roads lead to the Bayh-Dole Act of 1980, long ago in the PreInternet Age. Bayh-Dole grants patent rights to non-government entities for inventions resulting from publicly funded research. These non-government entities include universities.

As described in a trenchant analysis in The Economist from 2005, the primary legislative intent behind Bayh-Dole was to spur (and simplify) the commercialization of publicly funded research, which prior to Bayh-Dole was stagnating inside numerous disparate federal agencies engaged in R&D efforts. Once in the private sector, discoveries would be forged into products, in this case new FDA-approved drugs.

In a nutshell, here’s how it works. Professors in biology departments spend NIH-disbursed grant money on project proposals that have been positively evaluated by academic review committees. The study of biological processes invariably yields patentable results. In those instances, Professor John or Jane Q. Smith makes a beeline for the university technology transfer office, which has the Herculean task of shepherding patents into the promised land via licensing agreements that generate revenue streams to the university.

However, in practice a deluge of public and private funds and research scientists flow into the drug discovery pipeline but fewer and fewer drops trickle out the other end. Although I found examples of Bayh-Dole boosters, a balanced scholarly review of the law published in 2006 spelled out the flaws of a closed approach:

“By vesting such comprehensive discretion and flexibility in patenting and licensing with individual institutions, the Bayh-Dole Act provided the nation and the world with a large-scale experiment in how public institutions manage public assets as private goods. The outcomes have been positive on nearly all counts, but the Act inadvertently created a misalignment between the private interests of university technology transfer offices and public interests that benefit the innovation system at large or that enable access to IP for humanitarian purposes.”

I dug around and found more data-driven support for opening up biomedical research. First, take a look at this graph of Pharma productivity, which shows the number of drugs per billion US$ R&D spending over time, and please note the log scale on the spending axis:

What you’re seeing is the opposite of the famous efficiency gains in computing power dubbed Moore’s Law, hence the flipped moniker Eroom’s Law. Sure, we just had a bumper crop of new drug approvals in 2012. Leading the way was the groundbreaker Kalydeco, a new drug approved for the treatment of some cases of cystic fibrosis, the poster child of rare, single-gene diseases. But new first-in-class drugs for many devastating diseases, e.g., psychiatric and neurological diseases, are nowhere in sight. Not coincidentally, the success of Kalydeco depended on open collaboration between academics, industry, disease advocacy groups, not-for-profit foundations and patients.

Second, consider the age of scientific independence, i.e., age at first R01 award. The R01 is the bread and butter grant for tenure-track and tenured professors in Academia that provides $100,000s in public funding over several years:

In 1980, the average age of professors receiving their first R01 money was 36. By 2011, the last year for which we have data, the average age had climbed to 42, where it plateaued at the end of the booming late 1990s, when NIH’s budget was doubled to around $30 billion. It’s hovered there ever since.

This prolonged and unnecessary apprenticeship exacerbates the distorting effects that Bayh-Dole has on research choices, and encourages academics to engorge grant proposals with preliminary results and skimp on truly daring, basic research aims. It’s also a serious mis-allocation of social capital whose enormous potential would be unleashed in an open system. According to recent stats, less than 20% of people who enter the NIH-funded graduate training pipeline emerge on the other with a tenured professorship. The downward trend was apparent even in the early 90s, when the ratio was closer to 50/50.

Think about this for a second. The very capable and creative people who run the gauntlet of graduate school, one or more postdocs, AND an assistant professor search committee are a highly selected bunch. Even in so-called good years, R01 rejection rates were around 70%, and we are currently at historic highs of 83%. We’re squandering so much talent when we ask people who’ve endured a decade of intense, specialized training and established excellence as researchers to hang out for an extra decade at precisely the same time when many of them are starting families, usually after delaying parenthood.

Of course, I anticipate challenges from establishment thinking as well as organizational resistance. But the first step to recovery is calling a spade a spade. I expect secrecy from Pharma, but not from Academia, which I think abuses the freedom to pursue knowledge for its own sake on the public’s dime. At the same time, I expect more innovation to originate within Pharma and not simply be imported from Academia with taxpayer subsidies.

Eternally an optimist, I’m encouraged by the diversity of experimentation. Among the experiments I’m watching closely is the maturation of academic drug discovery, and efforts by the pro-entrepreneurship Kauffman Foundation, e.g., the free agent model, to empower scientists.

The open science Cambrian Explosion continues apace in 2013…

As we transition from campaign mode (Crowdsourcing Discovery) to research mode (Crowd4Discovery) in the weeks ahead, I want to take stock of exactly how far we’ve come in the form of a retrospective study. As I recently tweeted, the first scholarly publication to result from Crowd4Discovery will be an analysis of crowdfunding metrics. Basically, all of my weekly campaign dispatches will be condensed into a single citable document.

This exercise achieves two aims. First, I want to create an official “how-to” guide that demystifies the crowdfunding process for aspiring crowdfunders in the sciences and other creative endeavors.  Second, I want to preview the open process by which we will assemble future scholarly publications borne out of the actual research our team will conduct in the next few months.

Let’s start by taking a look at the distribution of donors across the eight donation levels spanning $10 to $1,000:

In percentage terms, over half of our donors contributed at the $25 level. Why was this our most popular donation? Well, we offered a pretty nifty tangible reward: a 3D printed methamphetamine model. But it also turns out that the median donation for many successful campaigns is $25, according to stats disseminated by crowdfunding site blogs.

However, small-dollar donors are only one part of a winning albeit lopsided coalition:

For example, $5-$49 donors (in blue) constitute 63% of all donors but they fund only 21% of our $25,000 goal. On the other hand, $200-$1,000 (in yellow) donors make up only 7% of all donors but contribute 35% of all donations in dollar terms.

So what did a typical day look like? As shown in this histogram, on most days we had fewer than 5 donors:

Along the bottom axis is the number of donors in a day. So according to this histogram we had five 1-donor days; ten 2-donor days; eight 3-donor days; and so on. There were only a dozen days when we had anywhere near double-digit donor “turnout.” And the outlier all the way to the right is the last day of our campaign.

To get a better appreciation for the really active days, take a look at these plots, where each orange circle is a day in the life of our campaign:

Again, the outlier is the last day of our campaign, which was unlike any day preceding it. Excluding the final 48 hours, the average daily haul was $367. Including every day boosts that average to $482, just above the magic number $481, which is what you get when you divide our campaign goal by the length of our campaign.

Digging a little deeper into the demographics of our donors, it turns out that 70% of our supporters are men, at least 25% are scientists, and roughly half are strangers or 2+ degrees removed. But can we say anything more specific about the composition of our donor pool?

In the aforementioned weekly campaign dispatches, I’ve been visualizing my Facebook network as a graph, with friends represented as nodes (circles) and friendships represented as edges (lines). The size of each circle is proportional to its connectedness, or degree, in the network. Let’s look at the distribution of donors (in yellow) across my Facebook network:

As I noted before, the distribution of donors doesn’t appear to follow any obvious spatial pattern, at least from this bird’s eye perspective. For example, it doesn’t appear to be the case that my most highly connected friends donated larger sums to our campaign. However, something interesting happens when you break up my Facebook network into friendship clusters, which correspond to different life phases or shared interests:

Donors are in yellow, and the fractions indicate the number of donors in each cluster. Several patterns emerged, at least to my eye. First, while the total donation turnout from my entire Facebook network is 17%, individual clusters donated either more or less or the average. Second, there appear to be at least two qualitatively distinct cluster topologies, or configurations in plain speak. On the one hand, the center-left cluster is a tight-knit core group of friends with more loosely connected individuals radiating outward. An extreme version of this topology is the bottom-left cluster. On the other hand, the center-right cluster is anchored by a central hub to which several smaller, non-mutual spokes of friends attach. In graph theory speak, what’s different about these two kinds of clusters is the average degree.

Does the average degree of donors differ from the average degree of their cluster as a whole? In other words, do yellow nodes tend to be larger or smaller in some clusters versus others? In some instances, the answer seems to be yes. Take a look at the two clusters below:

The average degree and total size of these two clusters is comparable. Yet the donors belonging to the right cluster have an average degree that is nearly twice the average degree of the entire cluster, meaning our message appealed more to core friends rather than to peripheral friends. On the other hand, the donors belonging to the left cluster have an average degree that is approximately equal to the average degree of the entire cluster, meaning our message appealed to core friends and peripheral friends the same. So content and connection both matter, depending on where friends reside in the network.

Now there are two caveats I’d be remiss not to mention. First, each friend in my Facebook network doesn’t see my posts equally given Newsfeed preferences and Facebook’s secret algorithm for presenting and prioritizing content. Second, some fraction of people in general are Facebook lurkers, which means they miss a lot of shared content.

I’ll keep plugging away at this Facebook dataset, and soon I’ll begin analyzing the donor distribution in my Twitter network. I’d love to get feedback from network aficionados out there!

For a summary of the final 24 hours, please go here.

For a summary of week 6, please go here.

For a summary of week 5, please go here.

For a summary of week 4, please go here.

For a summary of week 3, please go here.

For a summary of week 2, please go here.

For a summary of week 1, please go here.

And for a summary of the first 96 hours, please go here.

Spoiler alert: we got our outlier, and it was awesome:

Here’s how it went down.

On the previous day, Black Friday, we received a jolt of fresh media, which when combined with my exhortations on Facebook and Twitter netted us 25 donors, tying the previous one-day donor record set on Day 1 of the campaign. These 25 donors contributed around $1,000.

So we needed to do 5-times better the very next day to close the gap. But we hadn’t experienced a day anywhere close to 100 donors, nor raised more than $2,000 in a 24-hour stretch. What’s the difference between a campaign that stalls at 80% versus a campaign that reaches 100%? Did we have the secret sauce?

Then I remembered what I had calmly written a week ago in my final campaign update. According to stats published on crowdfunding site blogs, a successful campaign raises 50% of the goal in the first and last 10% of the crowdfund drive, and slightly more in the last 10% than in the first 10%:

We were on pace out of the gate, but then we grew more slowly than average, so we had to make up 37% of our goal in the last 5 days of the campaign. Thanks to a great turnout of Facebook and Twitter supporters, we raised most of the shortfall in the final 24 hours:

Throughout that last day, we average 4 donors per hour, and at our peak just before midnight on Sunday November 25^th, we had 14 donors in a single hour. Sure, that’s nothing compared to the tens of thousands of dollars raised per hour by Mathew Ingram (aka @Oatmeal) during his crowdfunding campaign to build a Tesla museum. But for us, we were raising in an hour what we used to raise in an entire day!

By the time the dust had settled, we’d raised $6,000 from 124 new donors, almost exactly 5-times higher than the preceding day. Who were these donors? As I just mentioned, my Facebook network turned out in a big way, but my tweeps really take most of the credit, as I’ll explain in the next post-game analysis, which will be out this weekend. In the interim, let’s focus on my Facebook network, because I started visualizing it a few weeks ago.

These are graphs of my Facebook network, which is comprised of 698 friends (nodes) and over 3000 connections (edges). The size of each node is proportional to its degree, which is the number of edges to other nodes. In other words, the large nodes are the hubs, the small nodes are orphans. The graph on the left is color-coded by friendship clusters. For example, the large red cluster is Grad School. The graph on the right has the connections wiped away for ease of viewing, and now there are only two colors: donors are in yellow, and non-donors are in blue.

17% of my Facebook network contributed to our campaign, and with the exception of a cluster of science friends I’ve made in the last 2 years, being a donor doesn’t depend on how I know you or how well you are connected to my other friends. Content over connection, perhaps?

I’ve only begun to think about these questions more carefully now that the campaign in behind us. Which brings me to my last point. The experiment part of our project will start in the weeks ahead, but the meta-experiment has already yielded a fundraising dataset that will be of use to the greater community.

I will be analyzing the Crowdsourcing Discovery campaign dataset and then writing it up as a paper in the open, a process that will serves as a model for how we intend to analyze our actual dataset when experimental results start to trickle in.

Thanks again to our 385 supporters — vive le Open Science!

For a summary of week 6, please go here.

For a summary of week 5, please go here.

For a summary of week 4, please go here.

For a summary of week 3, please go here.

For a summary of week 2, please go here.

For a summary of week 1, please go here.

And for a summary of the first 96 hours, please go here.

With 48 hours left the $6,754 question is – can we go all the way?

Here’s a plot of money raised over time, which I introduced several updates ago:

The black dashed diagonal line represents an idealized constant rate of donations over time, such that a quarter of the goal is raised in a quarter of the time, and so on. The ideal rate is $481/day. The orange line charts our actual progress. The observed average daily rate now stands at $372/day.

For the first two weeks, we over-performed relative to the diagonal, averaging close to $1,000/day post-launch. We were outstripped by utopia in the third week, and thereafter we steadily grew, with a few spurts along the way (see day 40) to even out the really slow patches, but a rate too slow to reach our goal unless it were to accelerate. Basically, we need a hockey stick finish.

How often do campaigns asking for $25,000 fund all the way when they need to raise 25% of their goal in the final 2 days?

One unfortunate fact I’ve learned during our campaign is that fundraising stats on specific crowdfunding projects are in short supply, making comparisons let alone projections difficult. I’m aware of at least one thorough post-game analysis. However, analyses based on averaging many fully funded projects are available on the blogs of crowdfunding portals like Kickstarter and IndieGoGo. This past summer, the folks at IndieGoGo blogged about a well-known feature of successful crowdfunding campaigns that seems to hold across project topic areas: 50% of contributions are made in the first and last 10% of a campaign.

Here’s IndieGoGo’s aggregate data showing funds raised as a % of goal (bottom axis) by % of campaign length, broken up into 10% chunks:

This is what is meant when people say crowdfunding campaigns are “U-shaped.” In the above rendering the U has fallen over on its right side, forming a C: an initial wave of contributions is followed by a long trough that gathers into a closing upswell that exceeds the opening burst.

Here’s that same plot with our data as of the start of this week. Keep in mind that our campaign is 52 days long. We’ll need a strong surge (yellow) in the homestretch, and with 2 days left we actually only need to raise 25% of our goal:

If I can indulge, Nate Silver-style, in numerically informed prognostication, our challenge boils down to finding 100 new donors, half of whom need to be strangers. I came up with these projections based on the following two pieces of data. First, the average contribution to our campaign has hovered around $70 since the outset: the remaining $7k divided by $70 equals 100.  Second, approximately ½ of CSD’s 258 donors are people I know IRL and ½ are people I don’t know, including online-only connections.

So can I realistically hustle up 50 friends? Using the open source graph program Gephi, I created a graph representation of my Facebook network. The size of each node is proportional to its degree, or the number of connections to other nodes. In other words, the largest nodes are hubs and the smallest nodes are orphans. The network on the left has been color-coded to reflect modularity classes, essentially friendship clusters. For example, my graduate school friends comprise the red cluster. The network on the right has been color-coded to reflect donors (yellow) vs non-donors (blue).

Two things jump out at me. First, the distribution of donors appears to be random with respect to degree, which means both hubs and orphans and everyone in between are donating. Second, the distribution of donors also appears to be random with respect to modularity class, which means there are donors from different phases of my life. I might have hypothesized before we started that all my science friends would donate but not my non-science friends. Interestingly, that doesn’t seem to be true, suggesting that the content of our project matters more than membership in a specific clique. In the remaining hours of our campaign, I will be curious to see how late-donating friends distribute across my network, and whether there will be evidence of a contagion effect.

And for my fellow data nerds, here are the daily box scores through the end of week 6 (note that I will be posting these data on figshare as soon as our campaign ends):

As a parting thought, everyone I’ve talked to in these last few days has wished me good luck, which is incredibly uplifting after a 50+ day campaign, I can promise you!

The thing about getting lucky is that it actually happens from time to time. Take tweeting, for example. My two best tweets of all time, judged by the number of times they were retweeted, are two random remarks that at the time seemed just like everything else I ever tweeted, but they get incredibly loud on the human mic:

The number of retweets (x-axis) and the number of impressions (y-axis) are plotted. Impressions are what the Twitter analytics tool called Crowdbooster calculates, in analogy to a Klout score, I suppose.

In yellow, my all-time best tweet is:

The title of second-best tweet is currently held by:

Getting a tweet with a link to our donation page to be retweeted over 100 times in the homestretch would constitute getting lucky.

Chance favors the prepared mind, I’m told.

For a summary of week 5, please go here.

For a summary of week 4, please go here.

For the summary of week 3, please go here.

For the summary of week 2, please go here.

For the summary of week 1, please go here.

And for the summary of the first 96 hours, please go here.

Week 5 of Crowdsourcing Discovery spanned the tail end of my northern CA trip and the 5-day breather back home in NYC before my trans-Atlantic campaign swing. Why the UK?

Above is geographic data from Google Analytics, a heat map showing the number of unique visitors from across Europe to my lab website between October 4^th and November 6^th. In other words, from the day we launched our campaign to a few days before I flew to London to attend the annual SpotOn conference series, at which I was scheduled to be a panelist in three awesome policy sessions: a session on crowdfunding, a session on the future of scholarly publishing, and a session entitled “What do you need to start a revolution?” Thanks to the miracle of technology, you can even watch archived video footage of each session on YouTube!

We started out week 5 with the 50% mark in our sights, and we ended the week just north of the halfway point. In concrete terms, in just over a month of online and IRL (“in real life”) campaigning, we received the support of 191 contributors from North America, Europe and Australia/New Zealand, demonstrating that our message has international appeal.

As I’ve done since my initial comparisons between previously successfully crowdfunded science projects on RocketHub and our campaign, here’s our donor distribution through week 5:

The ratio of $25 donors to $100 donors appears to be settling out at 2.5:1. I’m still a bit surprised at the dearth of $10 donations. Perhaps the 3D meth is really raising the floor to $25?

The average contribution = $67 and the median contribution = $25. Per usual, here’s the daily breakdown:

As I stated above, our campaign had close to 200 supporters by the end of week 5. Nearly 1/3 of our supporters are friends in my Facebook network, a number large enough that allows us to look for patterns of how these 60+ people are distributed across my FB network. One hypothesis is that my most connected friends are more likely to be contributors than friends who are less connected.

Sam Arbesman blogged about us again on Monday of this week, and he included a preliminary version of the figure I’m about to show you here, which is an up-to-date version of my Facebook network that shows who among my friends has donated:

In yellow are donors, and everyone else is labeled blue. The size of each node reflects its degree, which is the number of other people he/she is connected to in my network. For example, someone with degree 1 is connected to a single mutual friend. Someone with degree 55, which qualifies as a hub, is connected to 55 mutual friends.

So what’s the pattern? Well, I think the pattern is that there is no pattern (yet). The donors appear to be randomly distributed in two different ways. First, there doesn’t seem to be a relationship between donating and membership in a specific cluster. To see what I mean by cluster, take a look at the same exact network but with colors corresponding to different clusters (modularity classes, in graph theory parlance):

There seem to be donors in each of the major clusters, some of which I’ve named. I might have thought, for example, that all of my Princeton/postdoc cluster (green) would be chalk full of donors as we share not only friendship ties but also professional ties and a mutual love of science. But that doesn’t seem to be the case, suggesting that content may trump connection.

The second interesting feature of the apparent random distribution of donors across my FB network is that there doesn’t seem to be a relationship between donation size and degree:

With the exception of my hubs (degree > 40), there is no significant difference between the means of the the donor-level distributions, which is interesting because I might have thought that donors belonging to the champagne and Lexus demographics would be heavily invested in me and therefore in my creative endeavors as well. Instead, I think the simpler explanation may be that cash flow drives the variance in how much donors contribute, rather than something more complicated like network position.

The week 6, or penultimate, digest will be all about SpotOn and my London tour. Also, I’ll look more carefully at conversion rate, which I think will tell me which aspects of marketing worked and which didn’t..

For a summary of week 4, please go here.

For the summary of week 3, please go here.

For the summary of week 2, please go here.

For the summary of week 1, please go here.

And for the summary of the first 96 hours, please go here.

This simple 10-person network is comprised of 9 contributors to Crowdsourcing Discovery, and yours truly, A. What I find most interesting are the three shapes that emerge from our connections.

The first and simplest is the line, linking me to an “orphan” friend, F, who’s not connected to anyone else. The second is the triangle, which is formed between me and two friends who are also friends. Two examples are the triads A-B-C and A-D-E. The third and more complicated structure is the modified diamond formed by A-G-H-I-J, which is centered around H.

What happens as the network grows to include supporters K, L, M, N, and I tweak the color scheme to specify how I know these friends?

Blue indicates friends from college, red indicates science friends from grad school, yellow indicates non-science friends, and gray are other scientists I met through H.

The triads A-B-C and A-D-E are unaffected, while the modified diamond grows into a more complicated structure anchored by two mini-hubs, L and H. L is cool because he/she forms a triad with N and M, but a diamond with K, H and G.

What does this mean for the final two weeks of our crowdfunding campaign in terms of marketing strategies? Short answer: I’m still figuring it out! In the meantime, I decided to take a look at a graph of my entire Facebook network, which may provide clues how to convert more friends from likers to contributors in the home stretch.

Because the Internet is awesome, a tutorial on how to generate such a graph has already been crowdsourced! Start by downloading and running the suggested FB app, which exports a .gdf file that can be opened in Gephi as an undirected graph. (Undirected means mutual in graph theory jargon. An example of a directed graph is your Twitter network, where A may follow B but B may not follow A).

After you load the data file, use the algorithm called Force Atlas, as explained in the tutorial, to expand the initial hairball of connections into something with meaningful architecture. Here again is my network, which is comprised of 669 friends (nodes) and 3,227 friendships (edges):

The size of each node is proportional to its connectedness, and are most highly connected nodes are blue. The orphans and lowly connected nodes are more red, and green/yellow/orange nodes are somewhere in between. I’ve taken the liberty of labeling majoring clusters for ease of interpretation.

What stands out the most is that my college friends are the least cliquish of all my friendship clusters. In other words, I was friends with a lot of people who were each part of separate social groups. This more dispersed organization is unique to college. By contrast, from middle school to grad school and beyond to my postdoctoral years, distinct clusters of mutual friends emerge. I’m not sure what this means in the absence of comparisons. It could be a function of my attending Columbia College, or my early adult angst, or both. (Oh, and Burners are really tight).

If you’re interested in seeing what your FB network looks like but you’re a Luddite when it comes to network graphs, send me your .gdf file and I’ll render your network!

Let’s start with the headline numbers. As of this writing, we’ve raised $12,305 from 182 contributors. The average donation = $68; the median donation = $25. The average daily haul is $405, and there are on average 6 contributors per day.

Now let’s dig a little deeper to examine how our project is being financed as a function of contribution size:

In (A), I show the percentage of the total amount raised in three donation ranges: the beer demographic ($5-$49), the champagne demographic ($50-$199) and the Lexus demographic ($200-$1,000). In (B), I show the percentage of the total number of contributors from each of those three demographics.

How does our data compare to other science crowdfunding campaigns? Kristina Killgrove’s RocketHub project from the first round of the SciFund Challenge is the best apples-to-apples comparison. I don’t have access to her exact numbers, but if I apply the aforementioned demographic segmentation to her total amount raised, the “beer: : champagne :: Lexus” ratio comes out to 25% :: 30% :: 45%.

For the aficionados, here’s the daily breakdown of the four weeks spanning October 4^th to October 31^st:

As the title of this post suggests, the trickle occasionally turned into a flood, for example on Friday October 26^th and Tuesday October 30^th. What happened on those days? Well, on the morning of 10/26 we were just shy of the $10,000 mark, and I made a concerted push on Facebook and Twitter to push us over the hump. Also, on the preceding day we received positive press on the social media site Mashable, and I suspect there were lingering effects.

The spike on 10/30 was triggered by our being featured on the front page of the Reddit-like site called Hacker News. One of our evangelists, Bilal Mahmood, posted a link to our project page that morning. I was alerted to this fact by two different science tweeps; I confess that I’d never heard of Hacker News before. Of all the online buzz generated by press, blogs or trackbacks, the Hacker News post yielded the highest conversion rate, suggesting that Bay Area and NYC metro techies are a huge untapped market for our project, and possibly science crowdfunding in general.

The proof is in the project video plays, which are courtesy of Vimeo. For some perspective, the previous daily high was set on the first day of our campaign, and the Hacker News spike is 3-times larger:

On Friday November 9^th, I fly to London to attend the annual SpotOn Science conference (#solo12). I’ll be speaking on a panel about science crowdfunding, and a second panel on the future of scholarly publishing. Obviously, my attendance is part and parcel of our strategy to close strong in the final week of our campaign. Please email or tweet at me if you’ll be in London for #solo12, too!

For the summary of Week Three, please go here.

For the summary of Week Two, please go here.

For the summary of Week One, please go here.

And for the summary of the first 96 hours, please go here.

According to Google Analytics, the second largest source of traffic to my lab website is California, with 1,226 visits since the end of June:

So while I spend Week Four in the Bay Area spreading our message our transparency and engagement, here’s how we managed to raise $10,220 (as of last count) from 153 contributors with an average contribution = $66, and a median contribution = $25. I like presenting the data as a box score so you can digest the numbers at your own pace:

Lance Stewart, (@LJStewartTweet), one of the earliest pledges and then supporters of our effort, remarked in an email to me that we’re really doing two experiments – one scientific, one metascientific – for the price of one. So now that we have accumulated a decent amount of data, let’s plot daily contributor and daily contribution totals for the first 21 days, essentially just visualizing in graphical form the data presented above in the table:

The dotted line indicates the median, because the first day of our campaign was an outlier, both in terms of contributors and contributions. The second outlier was courtesy of my very generous 93-year-old Grandpa Walter,  though in absolute terms only 4% of contributors are members of my family. Another notable stat is the gender ratio of contributors. Turns out it’s 65/35 in favor of men.

By the third week of our campaign, all of the amplification generated by the initial volley of press coverage had faded, and fewer and fewer strangers were contributing. In order to keep the momentum going, I had to boost the conversion rate by focusing and personalizing our marketing strategy until the next media-drive spike in traffic came along.

Here’s a prime example. When we were mentioned in The Economist, I shared on Facebook a picture that I snapped of the actual article from my print copy. This picture received over 85 likes, including many by friends who’d already contributed. This got me thinking that a Facebook like might be a predictor of willingness to contribute, so I followed up with many of these likers in 1:1 correspondence. Lo and behold, it did the trick: 10/16 contributors in the 48 hours after I reached out on Facebook were my friends who had liked the picture share.

So what does this all mean going forward? If you were to imagine a perfectly linear relationship between the amount of money raised and the amount of time elapsed, we would be receiving contributions at a constant rate. Stop all the presses – we don’t live in an ideal world! Instead the world of crowdfunding is U-shaped, in that a disproportionate amount of contributions come in the beginning and at the end of a campaign:

We have a lot more work to do, and we intend to close strong. All I can say is that I’m looking forward to the next three weeks, and I hope you are, too!

If this was your first time reading about the behind-the-scenes of our campaign, I’ve been releasing weekly status updates since we launched Crowdsourcing Discovery on October 4th.

• For the summary of Week Two, please go here.
• For the summary of Week One, please go here.
• And for the summary of the first 96 hours, please go here.

As of this writing, the headline numbers are: $8,565 raised from a total of 125 contributors. The average contribution = $68; the median contribution = $25.

The big news from Week Two was that our project was mentioned in an article about science crowdfunding that appeared in the print and online editions of The Economist! This has to be the most awesomely alliterative title in the history of science journalism: “Many a mickle makes a muckle:”

(the tag line under Aahnold reads: Redistribute this, punk)

How did this come to pass? Lou Woodley (follow her @LouWoodley) of Nature, whom I met at the Spot On NYC (formerly SONYC) monthly salons, graciously put me in touch with Akshat Rathi, a UK-based science writer. Akshat interviewed me the day after our campaign launched, and we had a great discussion about what motivated the CSD project and science crowdfunding in general.

Switching gears, let’s look at some data and charts and networks. Let me begin with what should by now be the familiar donor distribution:

After two weeks, there’s a rougly 2:1 ratio of $25 to $100 donors. So how are the average and median contributions holding up? Here’s an updated table of daily stats:

To be expected, the average number of contributors fell by around half from 9 in Week One to 5 in Week Two. Similarly, the average contribution dropped three-fold from $766 in Week One to $255 in Week Two. And weekends are always a drag. Another notable milestone from Week Two: we crossed the 100^th contributor mark!

The question that I get all the time is how many contributors are already in my social network? Let’s start by crunching some numbers. Plotted here is a breakdown of all $25 and all $100 contributors according to whether they are connected (or not connected) to me in a social network:

More strangers gave me $25 vs. $100. Well duh. How often do you give $100 to a stranger?

What I find interesting to think about is the connectivity between contributors who are also my friends. I used a freely available program called Gephi to create the first of many social-network representations of contributors and their connections to me and to each other. In other words, people are nodes (the filled circles), and friendships are lines.

Consider this real-life 10-person social network comprised of me (the green node) and 9 contributors, 8 of whom are my friends. To protect the innocent, let’s refer to everyone as A thru J. Notice the triads connecting me to different pairs of mutual friends, e.g., A-D-E. Also notice the two different kinds of singletons, F and I:

There’s a lot more I can say, but I’m interested in hearing your feedback first. The comment pool is open. Jump in!

In the next update on Week Three, I’ll take a more bird’s eye perspective on these social networks of science benefactors. In other words, more fun with Gephi!

I thanked PZ Myers on Twitter but he’s a busy, famous guy who probably gets tweeted at all the live long day. So in the off chance he ever sees this post: thank you, PZ! I thought his depiction of our project and our intent was fair. But I strongly disagree with his characterization of crowdfunding basic research as having to “sing and dance.” Jarrett Byrnes, co-founder of the SciFund Challenge and a professor of ecology at UMass Boston, made a stirring, well-articulated defense of crowdfunding in the comment thread (#14 to be precise), so I won’t belabor the point. Rafael Najmanovich, a professor of biochemistry in Canada, also contributed thoughtfully to the debate (he’s comment #37).

Alas, with a few other exceptions, the rest of the comment thread left something to be desired. For better or worse, the comments do represent the breadth of the ideological spectrum. Some comments were downright hilarious, and I mean double over in uncontrollable convulsions of laughter.

I took the liberty of reproducing here on my lab website nine of representative selections. Enjoy!

The Perfectionist

One problem I can see is that people may not know what they’re supporting. I can imagine gullible woo-believers giving money to hacks attempting to discover the miraculous power of water that’s somehow not just water, and exposing the horrors of how the government created HIV to control the population.

The Kvetcher

Let me guess: You haven’t been to a university committee meeting yet, right? Scientists, especially when they are trying to run anything as a “collective”, are no better at cooperating than any other group of people with similar interests but competing individual goals.

In my experience, that means a lot of bickering and power games, then some half-assed solution is quickly agreed upon when participants realise that they need a result to show, and then it’s back to power games and trying to use/change/re-interpret/sabotage the agreed plan to one’s own advantage.

Oh, and as a lot of others pointed out, for basic working, a business needs an actual business concept (i.e. what to sell to whom), appropriate marketing (i.e. actually selling the stuff) and strategic planning (i.e. knowing what to sell tomorrow), as well as starting capital as well as ongoing resources to generate whatever it’s trying to sell. And then we haven’t even talked about how many expensive-to-run branches of research have no commercial applications for decades, and obviously it’s exactly these that are in need of funding.

The Canadian

Man, if only there were some way to have a peer-reviewed crowd-sourced funding mechanism for supporting robust scientific research, where everyone giving just a little money would add up to a lot… if only…

Oh wait, we HAD THAT before our governments decided that science wasn’t a worthwhile investment. It’s called TAX REVENUE and the NATIONAL SCIENCE FOUNDATION, or here in Canada, NSERC. *headdesk*

The Self-hating Libertarian

Sigh… Back in my benighted days as a libertarian, I read L. Neil Smith’s series of propaganda science fiction novels that took place in an anarcho-capitalist utopia. In the first novel, the series’ main character discovers that his wife has a fatal disease that is incurable even with the magical technology that a world with an absolute free market and no regulations is supposed to produce. Since the woman was on borrowed time and there were none of those evil, freedom-destroying governments to steal the productive capacity of the achievers, they put her in suspended animation and started up a charity fund to find a cure.

This was considered by the author (and, admittedly, myself at the time) a GOOD idea.

The PETA person

Great news. Now we don’t have to have our tax dollars taken from us, forced to support the torture of sentient animals. I appreciate that.

The PETA hater

The vast majority of animal torture research is privately funded. Except for the research into killing bipedal animals that wear clothing and organize themselves into groups — one of which we call “the enemy”. The government’s 100% behind that.

Seriously, have you any idea how much HUMAN suffering animal research prevents. Please go fuck yourself sideways with a rusty porcupine for suggesting that researchers conduct experiments with animals merely to get their jollies.

The Seemingly Reasonable Person with a Well-Concealed Bloodlust

I’m well aware that a lot of research isn’t profitable, especially in the early phases. It may not be ideal for the pace of research some may want, but at least it a) employs researchers and engineers and gives them something to do and b) gives them some control on what they want to research or engineer.

With problems of peer review, simply don’t have peer reviewers with a conflict of interest. This should be obvious.

As for factory workers, technicians, and who ever else are needed for the operation, hire them. It creates jobs; which even politicians can’t figure out how to do, apart from pleading to capitalists. Buy the supplies. After all, it is the sale of goods that bring in money. At least then scientists wont have to beg the government or private business, which incidentally is what I mean by “self-sufficient”.

@Rutee Katreya
>Randroid

Don’t even compare me to such lowly creatures. As a socialist, I hate capitalism. To compare me to an ultra-capitalist is a grave insult to my person. I spit on the grave of Ayn.

But as to who is going to build it: builders of course.

The Daily Kossack

Keith is a Randroid or Looneytarian obviously. With nonsensical answers to simple questions. Try these.

1. Hey Keith!!! Why don’t you find a Libertarian paradise somewhere and join it and stop wasting electrons and photons on the internet? There are 220 countries in the world, some you can buy cheap, and they are always looking for an edge. The current leader is Somalia. I’m sure they can use a new body. The old ones keep getting shot full of holes.

2. What is stopping you and John Galt from setting up a Gibbertarian paradise in the USA? There are huge areas that are emptying out as people move away, east of the Cascades and Sierras, the Midwest. Could it be that you and John Galt are just babbling idiots? My guess is that you would miss your mommy, get lonely, and get real hungry in a hurry.

The Ayn Rand Fetishist

That’s cute and all, Kevin, but I ask you the same thing I ask Randroids who think they can make a Galt’s Gulch:

Who’s going to fucking build it?

IMHO, there are two really important concepts in genetics that need to be impressed upon undergrads. The first is dominant vs recessive and the second is, forgive the dreadfully neoclassical diction, epistatic vs hypostatic. In lieu of ancient Greek I prefer plain English: expressed vs suppressed. I previously blogged about dominance and recessiveness using a drug resistance selection as a case study.

Part of the confusion over epistasis can be corrected by being clear about definitions. Dominant and recessive describes the outward effects of mutations in a single gene that exists in two potentially different copies, while expressed vs suppressed describes the influence of mutations in one or more genes on the outward effects of a mutation in another gene, and the fact that there are can be two different copies of every gene only makes matters worse.

Naturally, I thought one could convey the essence of epistasis by way of a good paper. After some Pubmedding, I came across this article from PNAS:

Ladies and Gentlemen, representing all plant life on Earth, I present the welterweight champion of the angiosperm universe: Arabidopsis thaliana, aka the mustard weed!

(despite my moniker, I am awesome)

The Zhou et al study features a series of easy-to-understand experiments, starting with a genetic screen and then transitioning to what’s known as epistasis analysis, which involves creating double mutant strains whose traits are compared to the traits of the single mutant strains from whence they came.

Turns out that Arabidopsis seedlings don’t grow leaves when exposed to high concentrations of glucose, the thinking being that photosynthesis isn’t necessary if there’s free sugar lying around. A clever geneticist can turn this leaf-growth suppression by glucose on its head and look for mutants that are resistant to glucose’s effect. That’s precisely what the authors of this study did, and in the process they identified a mutant they christened gin1-1, which is short for glucose insensitivity:

Compared to wildtype (“Ws-0″), the gin1-1 mutant grows just fine in the presence of excess glucose, and that’s apparent when comparing individual specimens side-by-side (as in A); when comparing many dozens of individuals (as in B); and when comparing individual specimens over many days of observation (as in C). However, the gin1-1 mutant grown under normal greenhouse conditions is smaller than wildtype, as shown in D.

Next, the authors dutifully crossed the gin1-1 mutant back to wildtype to determine whether the mutation is dominant or recessive, in this case recessive:

As stated above, the reason why sugars like glucose suppress leaf development is metabolic efficiency. Leaves are where photosynthesis occurs, so why should a plant spend all the energy required to make leaves if the product of photosynthesis is freely available? To prove this point, the authors measured the activity of genes required for photosynthesis in the presence or absence of glucose:

Then the authors set out to perform some important control experiments. What would cause glucose insensitivity? One trivial explanation is that the gin1-1 mutant is unable to absorb glucose from its environment, say because the glucose transporter protein is defective. Or, maybe the glucose isn’t correctly metabolized it once it’s inside cells, which might happen if an enzyme in the pathway that metabolizes glucose is broken. To rule in or rule out these explanations, the following experiments were done, including the first double mutant experiment:

The gin1-1 mutant take up glucose from the environment at a rate identical to wildtype (as shown in A). The enzyme HXK1 is hexokinase, which phosphorylates glucose after it enters cells, thereby preventing glucose from diffusing right back out into the environment. As shown in C, the gin1-1 mutant crossed to a strain that contains multiple copies of HXK1 resulted in a double mutant (“35S-AtHXK1 gin1-1“) that looks exactly like the gin1-1 single mutant exposed to excess glucose. In other words, GIN1 is not HXK1. Moreover, the protein encoded by the GIN1 gene does whatever it does after the HXK1 enzyme phosphorylates glucose.

The authors noticed that the gin1-1 mutant displays many of the same phenotypes as wildtype seedlings (“Ws-0″ and “Col-0″) exposed to the plant hormone ethylene, or as mutants that overproduce or overreact to ethylene, such as eto1-1, ctr1-1 and etr1-1 (note that ACC is a chemical precursor of ethylene):

Panel B revealed an interesting observation. The ethylene-response mutant etr1-1 exhibited the opposite phenotype of gin1-1 when exposed to different amounts of glucose (0%, 4% and 6%), namely the leaf development of the etr1-1 mutant is more suppressed than wildtype. Struck by that observation, the authors created a etr1-1 gin1-1 double mutant and exposed it to excess glucose. Would this double mutant look like gin1-1 or etr1-1??

This is where the epistasis rubber meets the road:

As demonstrated in a variety of conditions, the etr1-1 gin1-1 double mutant clearly resembles the gin1-1 single mutant, and looks nothing like the etr1-1 single mutant. The conclusion is that the GIN1 protein normally performs its function after the ETR1 protein performs its function, as shown in the final figure of the paper, which I had my student teams attempt to recapitulate at the chalkboard to varying degrees of success:

Out of the starting gate we benefited from tailwind generated by a post on Sam Arbesman’s wired.com blog, Social Dimension. That amplification, plus my own volley of tweets and Facebook sharing, netted us in 24 hours 35% of the total contributions donated in the first week of the campaign. Then in the 24-48 hour interval, more coincident blogging by the science blogosphere, a post on Derek Lowe’s In the Pipeline, followed by a post on Ash Jogalekar’s Scientific American blog Curious Wavefunction, sustained our momentum going into the first weekend.

If you’re never heard of or seen the “Lowe Bump,” behold it in full bloom (orange arrow) in this plot of site traffic to perlsteinlab.com (courtesy of Google Analytics) from October 1 to October 11. (Chemists, please note that Ash Catalyst was added to the reaction):

Here’s a closer look at the first 72 hours post-launch, spanning October 4-6:

As I blogged two months ago in a post called Quantified Self Publishing, those blips that peak at 30 visits correspond to retweet (RT) events. Several RT blips occurred around Noon EDT, which is when I usually send out my first substantive tweet of the day. Each RT blip is short-lived, lasting approximately one hour. The Ash-catalyzed Lowe Bump was an entirely different beast. First, it was an order of magnitude above baseline traffic levels. Second, it faded gradually over the course of 12 hours.

Okay, traffic data to my lab website is nice, but what about traffic to the Crowdsourcing Discovery project page on RocketHub? As a proxy for pageviews, which are not directly available from the RocketHub website, I’ll use data from the Crowdsourcing Discovery project video page, which I scraped from Vimeo. A similar picture emerges:

So how much money did we raise? How many people contributed? Who’s contributing?

As of one week + one day, we’ve raised $5,990 (24%) from 75 contributors, so an average contribution of around $80. The median contribution = $50. The daily totals are shown here:

It’s pretty clear that getting more than 10 donations/per day requires popular press amplification. Coordinated tweeting by me and a few project supporters, including the setting of small, realistic calls to action, resulted in the 3-fold bump on October 11^th. Going forward, the only way we’ll top 25 contributors in one day will be if we’d be fortunate to get a RT or mention from a massively followed tastemaker or fresh press coverage garnering 5,000 – 10,000 pageviews. What happens at the 100,000-pageview level I can only imagine…

Here’s the distribution of donations at the 8-day mark:

Who are these brave souls? 60% are friends and family. Of the 30 people not directly connected to me, at least 5 are directly connected to one of the other team members. Less than 25% are scientists, defined as a grad student, postdoc, tenure-track professor or private-sector researcher.

For the next crowdfund drive update in a week, I’ll start to visualize the data as actual networks in order to see if any interesting connectivity patterns emerge. And in case you missed it, I blogged about the first 96 hours after launch here.

In the meantime, your feedback is welcome in the comment thread below. Thanks!

In the 96 hours post-launch – so from Thursday October 4^th to Sunday October 7^th – 47 people contributed to our campaign. (Thank you again, you all rock!) The average contribution = $79, which is consistent with other projects on RocketHub and with projects on other crowdfunding portals. The median contribution = $50. 57% of the 96-hour total was contributed in the first 24 hours by 25 individuals, so roughly one person per hour.

The daily breakdown is shown in this table:

In my first post outlining the genesis of CSD, I generated plots of the distribution of donations for several successful science crowdfunding projects that were part of the #SciFund Challenge consortium led by Jai Ranganathan and Jarrett Byrnes. (Incidentally, #SciFund just finished recruiting for their 3^rd round of projects; learn more here). Based on the above data, here’s the distribution of contributions for CSD:

I’ve reserved the comment thread below for interpretations.

Some of the buzz generated by the project can be found here, here, and here.

Next time, I’ll present updated data for the first week, and preliminary data on the social networks of contributors.

Well, not exactly. I also described in detail an alternative mechanism called the weak base model, whereby amphetamines increase neurotransmitter levels by chemically interfering with the sequestration of neurotransmitters into synaptic vesicles. In the years following the debut of the weak base model in 1990, the lab of my Crowdsourcing Discovery collaborator and original weak base proponent, David Sulzer, has observed that the accumulation of amphetamines has other effects on neurons, as well as non-neuronal brain cells, e.g., glia. For example, the toxic effect of unsequestered dopamine, which can be damaging to cellular components in ways that are similar to free radicals.

But Sulzer’s lab observed that amphetamines also have dramatic effects on cell membranes. In 1994, Sulzer’s lab published a paper in which they measured the effects of 24-hr, high-dose methamphetamine treatment on dopamine-expressing neurons grown in the laboratory. Here’s an example of what they saw:

(reproduced from Figure 4 of Cubells et al)

The neuron in A1 was the control, while the neuron in B1 was treated with meth. The meth-treated cells showed evidence of “vacuolation,” or an increase in the number of intracellular compartments (those tiny bubbles). On the right, you can see the same neurons shot through with a laser that illuminated fluorescent microspheres (those white specks) that can only enter the neuron through an ancient process called endocytosis, which involves the transport of cellular cargo in vesicles. What you should take away from these data is the fact that the microspheres reside in the same intracellular compartments induced by meth, suggesting that meth accumulation affects the flow of membranes.

And it wasn’t just Sulzer’s lab that observed membrane-related drug effects. I previously blogged about the seemingly overlooked body of literature from the 1960s-1980s documenting drug distribution across tissues and throughout cells of whole animals treated with psychoactive drugs. In fact, scientists have observed different drug accumulation rates in different tissues, and the cellular adaptations to drug accumulation also vary between tissue types.

Examples of psychoactive drug distribution include the work of Italian researcher Francesco Fornai and colleagues, who published a paper in 2002 on the effects of MDMA (“ecstacy”) on brain cells of mice. They observed curious looking membrane structures inside neurons of the striatum, like the one in the right panel below:

(reproduced from Figure 2E of Fornai et al)

Very similar looking membrane “whorls,” so described because they resemble the patterns on your fingertips, can be seen inside the cells of other organisms treated with chemically related psychoactive drugs, in this case a humble yeast cell exposed to the SSRI antidepressant Zoloft (left panel), unpublished data from the Perlstein lab.

Unfortunately, several lost decades of “high affinity drug target” focus have left pharmacology in a state not too different from swiss cheese, with holes in our basic understanding of how cells respond to drug accumulation, in particular drug accumulation in cell membrane, which is caused by a primordial, in fact prebiotic, chemical attraction between certain psychoactive drugs and phospholipids, the stuff of membranes.

But do these drugs actually get enmeshed in cell membrane, or are these drugs accumulating in some other part of the cell, and we just see ripple effects in the form of aberrant membrane structures?

Autoradiography is the perfect experimental tool to resolve that question. By far the biggest advantage of autoradiography is that you don’t have to alter the structure of the drug being studied. Recall Heisenberg’s uncertainty principle: the act of measurement changes the object you are trying to measure!

Knock knock.

A radioactive (“radiolabeled”) version of a drug is infinitesimally heavier than the non-radioactive parent compound. For Crowdsourcing Discovery, we will use amphetamines that contain carbon-14. Perhaps the best way to understand the power of autoradiography is to consider a real world example from the annals of psychopharmacology: lysergic acid diethylamide, or among friends, LSD.

(reproduced from Figure 1 of Yagaloff & Hartig)

What you’re looking at a section of the adult rat brain. The white areas show the distribution of radioactive LSD against a black background. The brightest areas (indicated by the white arrows) are where most of the radioactive LSD wound up. Based on this “map,” one can isolate the regions of high drug accumulation and do further experiments to implicate specific drug targets, also using the radioactive version of the drug.

Now, there have been autoradiography studies using radiolabeled amphetamine before anyone says that we’re claiming to be the first to try this experiment. But those studies were done a long time ago (in the early 1970s), haven’t been reproduced using modern technological advances, and were performed at poor resolution:

(reproduced from Figure 1 of Placidi et al)

You can see swaths of the mouse brain darkening, but each region of radioactive amphetamine accumulation contains too many individual cells to count. If we really want to be able to describe how amphetamines work, we first need to determine precisely where it’s going deep inside cells, down to their membranes.

Believe it or not, that’s not hard to do using a powerful electron microscope that magnifies tiny structures like cell membranes, which are almost 100-times smaller than what can be seen at the limit of light microscopy. Daniel Korostyshevsky (@badomens) has been the Perlstein Lab resident electron microscopist since 2009, and has honed the craft of electron microscopy both in yeast cells and a rat neuronal cell line. The Sulzer Lab has decades of experience working with mice and amphetamines.

Together, we have a team in place that can execute a set of focused autoradiography experiments with amphetamines in lab mice. Could this be funded by traditional sources? In theory, yes. Will that happen any time soon? In practice, probably not. Just consider the age when independent basic research life-scientists get their first stable funding stream (the R01) from the National Institutes of Health: 42. I’m 32. And when I contemplate the 80% rejection rate for R01 grant submissions, crowdfunding starts to look really attractive.

We launch Crowdsourcing Discovery in less than 24 hours! Thanks to everyone who volunteered their time and feedback over the last few weeks. You know who you are, and I’m working on a formal acknowledgement post once we clear the launch pad.

As a reminder to those tuning in for the first time, the following posts cover all the major areas of the project:

1) overview

2) engagement and perks

3) what’s already known about amphetamines

Science instructors, it’s simple. The paper-as-puzzle approach involves only three steps, and 1-2 hours of lead time. First, select a concise, seminal and accessible paper from a domain of expertise, in my case genetics, and understand it backwards and forwards. Second, copy and paste all of the figures and tables from said paper into PowerPoint and present to your class, datum by datum (approximately 30 minutes). Third, have the students form breakout groups in which they deliberate( for another 30 minutes) and formulate a three-minute summation that one volunteer from each team presents at the blackboard in front the entire class.

Think journal club, but more didactic, more interactive, and the participants don’t get the benefit of reading the paper beforehand! Now I admit that there’s nothing revolutionary about the idea of incorporating papers into science pedagogy. My first-year graduate seminar at Harvard – MCB100 – was just reading assigned papers on our own and discussing them with a professor in class. There must be countless courses just like it in every biology department.

The novelty comes from the timing. Based on my own early exposure to the primary scientific literature when I was still in high school, I believe that it’s never too early for budding scientists to start interpreting experiments. But later, when I was an undergrad at Columbia, my sophomore biology recitations consisted of a review of that week’s lecture material by the teaching assistant, and an anemic group problem-solving session.

Once upon a time those problems were derived from honest to God papers. However, years of editing and drift have concealed their origins and diluted their complexity. It’s to be expected: grading problem sets is tasked to overworked and underpaid graduate students who want to simplify and standardize evaluation as much as possible. And it takes time and commitment to revitalize a problem set archive. Has this arrangement changed much in 10 years? I doubt it…

So, the paper I selected as my puzzle was originally published in 1991 in Science by Prof Michael Hall’s lab, and entitled “Targets for cell cycle arrest by the immunosuppressant rapamycin in yeast.” This paper describes the isolation of mutants that are resistant to the natural product and therapeutic drug rapamycin. Based on the set of genetic observations in the paper, one could reasonably propose a physical model of how rapamycin works, i.e., what are its targets and how does it interact with these targets?

I chose the rapamycin case study because I worked extensively with rapamycin in graduate school, and so I understand the mechanism well. (They say you should teach what you know). Also, I did a trial run with this paper last year, and I received some of the best feedback from students afterwards. Although none of the students this year or last year solved the puzzle and explained how rapamycin works, some teams got really close.

I’ll go step by step through my presentation of the data, adding commentary and annotation along the way. I opened with this figure, featuring yeast cells treated with rapamycin (A) vs. untreated yeast cells (B):

The students should discern, or be alerted to, two observations. First, the rapamycin-treated cells aren’t budding. Second, the enlarged rapamycin-treated cells appear to be arrested at a pre-mitotic step in the cell cycle, either G1 or G2. It’s not evident from the data so I told the students that the effects of rapamycin are reversible.

Next I showed them this table:

These data depict the growth responses of rapamycin-resistant mutants. The drug resistance selection yielded three complementation groups: fpr1, tor1 and tor2. I explained that the fpr1 mutants are recessive, while tor1 and tor2 mutants are dominant. Most students picked up on the fact that amino acid residue 65 was targeted by multiple independent mutations. Some students also noticed that the nonsense allele (fpr1-12) exhibited stronger resistance than the missense mutants.

Next up was this figure:

In this experiment, the wildtype FPR1 strain was compared to the fpr1 whole-gene deletion mutant. The students know that the fpr1 mutants are recessive for rapamycin resistance, yet here they learn that the FPR1 gene is completely dispensable for yeast cell growth. They can only reconcile these observations by concluding that rapamycin requires the FPR1 protein for its growth inhibitory effects, and that loss of FPR1 renders rapamycin harmless. Consistent with that interpretation is the fact that loss of FPR1 protects yeast cells across a 1000-fold concentration range of rapamycin.

Here comes an essential clue to the puzzle, which most of the students failed to appreciate fully:

FK506 is a structural sibling of rapamycin. In fact, one half of each of the two molecules is completely identical. From the plates above, the students should glean that FK506 and rapamycin mutually suppress each other. At this point many of the students realize that FK506 and rapamycin both bind to the same target, namely FPR1. TRP1 is a bit of obfuscation, to borrow from the Car Talk guys. Loss of TRP1, which encodes a gene required for tryptophan biosynthesis, sensitizes yeast cells to FK506, but has nothing to do with rapamycin’s mechanism of action.

For good measure, I showed them this sequence alignment data for FPR1:

What’s clear is that residue 65, which the students already knew was the frequent target of resistance-conferring mutations, is evolutionarily conserved and therefore functionally important. Most students appreciated that this meant that residue 65 comprises the binding site for rapamycin on FPR1.

However, in the end students seemed to forget about the other rapamycin-resistant mutants, tor1 and tor2, which display dominant resistance. In fairness, the data in the paper revolved mostly around FPR1, but the puzzle cannot be solved without the TOR proteins and an explanation of the dominant resistance phenotype, which I’ll leave to you, astute reader.

The solution to the puzzle is that rapamycin (lavender) binds to FPR1 (blue), and this drug-protein co-complex then binds to TOR (red), which is a kinase required for passage through the G1 cell cycle checkpoint:

As a Lewis-Sigler Fellow at Princeton, I’ve been fortunate to take part of grand pedagogical experiment called the Integrated Sciences Curriculum (ISC), though some would call it a return to olden times, before the de facto premed track engulfed undergraduate biology whole. This slide from applied quantitative reasoning toward cartoon models and rote memorization is coming home to roost; curricula like ISC aim to reverse the trend. In a nutshell, the goal of ISC is to reconstitute a separate biology track with calculus and probability theory at the fore. Obviously, this means a much smaller cohort of undergraduates  — n=20 by the sophomore year — who actually want to be card-carrying scientists when they grow up. Layer on top of that quantitative foundation a generous helping of computer science know-how, mostly in the form of MatLab proficiency, and voilà: Integrated Science!

Since 2010, I’ve been lead preceptor – that’s Princeton-speak for teaching assistant – for the yearlong sophomore bio course ISC235/236. From the outset, I knew I wanted to experiment with the precept form. (Apologies again: precept is Princetonese for “section”). I envisioned a Socratic Journal Club, in which the students would read beforehand a paper corresponding to the week’s lecture material and come to class prepared to turn it inside and out, with yours truly serving as moderator-in-chief.

Last Thursday was the first precept of the Fall semester. In the two previous years I had used the famous Meselson & Stahl paper, which demonstrated semi-conservative replication of DNA, as the icebreaker:

However, this year I wanted to lead off with the paper I usually saved for the second week, namely Hopfield et al, better known as the paper that elegantly demonstrated kinetic proofreading in protein translation. This go-around I wanted to mix things up a bit, so I didn’t give the students any warning. They glimpsed the paper for the first time in precept, and I gave them 45 minutes to read all five densely packed, crisply written pages in small breakout groups. Then for the remaining 45 minutes we walked through the paper’s methods, results and conclusions, figure by figure:

I asked my students to tell me the purpose of each experiment, and what could be gleaned from the data. I also asked them to prove that they understood how the experiments had been done, even though few of the techniques are widely used today. At the end of precept, I asked them to rate the paper on a difficulty scale of 1 to 5. Most found the paper to be a 2 or 3, which is where I had hoped it would score.

Now if reading a paper in class doesn’t seem particularly radical, you’re right it’s not. However, what I do think is refreshing is exposing undergraduates to papers as early as possible, instead of waiting till their junior or senior years, or until the first year of graduate school. The Journal Club is really just Lab Meeting with someone else’s data, and a lab meeting is the crucible in which data are forged into knowledge. So why not expose budding scientists to the scientific method in action at the very same time they’re acquiring independent computational and analytical capabilities?

The floor is open to discussion..

Transmission electron microscopy is a powerful technique that allows us to peer deep inside cells. Dynamic cellular processes, like the fission and fusion of internal compartments, are frozen in time by an aldehyde-containing fixative, allowing us to get a glimpse of a population of near 100% genetic clones.

The cell in this gallery is a wildtype strain (BY4716) of Saccharomyces cerevisiae, or budding yeast. As is evidenced in the high magnification zooms, the formation of multilamellar compartments is a normal event in the lives of vacuoles and related trafficking organelles.

If you are interested collaborating on quantitative image analysis of this kind of data, please contact me on Twitter!

[The above image was lifted from Mike Mitchell, and I added the chemistry bits]

The idea for our $25k meth crowdfunding project came to me earlier this summer after a meeting with David Sulzer and members of his Columbia Med School lab. Sulzer’s lab has studied how amphetamines work for over 20 years. I had emailed Sulzer a few weeks prior to our meeting with an invitation to discuss the broader implications of my lab’s April PLOS ONE paper, which revealed new insights into the effects of the SSRI antidepressant sertraline, aka Zoloft, on the structure and function of cell membranes.

The foundation of our mutual interest is the 1990 Neuron paper by Sulzer and his mentor Stephen Rayport, in which they debuted the “weak base” hypothesis of amphetamine action, to which I’ll return after a brief historical interlude.

It’s been known for decades that amphetamine and amphetamine-like compounds stimulate the release of catecholamine neurotransmitters, e.g., dopamine, from the synapses of neurons, yet we (pharmacologists) still haven’t enumerated all of the cellular targets of amphetamines. By cellular targets I don’t just mean protein targets, but also the vital chemical equilibria that are perturbed by amphetamine accumulation in cells, e.g., ion gradients.

Conventional wisdom holds that amphetamine directly competes with dopamine for access to the dopamine transporter protein (DAT), which is expressed in dopaminergic neurons, and selectively slurps up dopamine after each round of neurotransmission so that it can be repackaged into fresh synaptic vesicles for future rounds of excitation. This model is based on structural similarity between amphetamine and dopamine, as shown here:

However, the affinity of amphetamine for DAT is reportedly weak in some studies but strong(er) in others. The trouble with measurements of drug binding affinity is that different pharmacologists do the experiment differently. In other words, people compare the ability of a panel of drugs A-Z to displace a primary compound that binds to DAT with high affinity and selectivity, but the choice of primary compounds, which can bind to different locations on the DAT protein, often varies between studies.

A comparison is often made between cocaine and amphetamine, since their addictive properties are both thought to involve increasing dopamine levels. Remember those self-administration drug studies with lab rats? This excerpt from a 1987 Science paper of that ilk by Ritz et al sums up the situation well:

“CNS stimulants can be divided into two classes on the basis of their biochemical effects on catecholamine-containing neurons in brain: the amphetamines and the nonamphetamines (cocaine and methylphenidate). Drugs in the former class inhibit reuptake and are potent releasers, while drugs in the latter class inhibit reuptake but are more restricted in their releasing properties.”

Alas, the rabbit hole goes deeper still. Amphetamine and methamphetamine have been shown by some labs to cause DAT to be removed from the synaptic membrane, resulting in DAT sequestration in intracellular compartments, e.g., endosomes, lurking below the membrane surface. But as is often the case in experimental pharmacology, the reproducibility of these results depends on the duration of drug treatment and whether cells were exposed to drug in vitro vs in vivo. Annette Fleckenstein’s lab does solid work here, and her lab’s publications are required reading for budding psychopharmacologists.

So, getting back to the weak base hypothesis. It’s really a simple idea. Even a fleeting glance by the trained eye reveals the chemical propinquity between amphetamine and dopamine: both molecules possess an amine that is ionized, in this case protonated, at physiological pH. Nature exploits this amine in an ingenious way. Synaptic vesicles are loaded with amine-containing neurotransmitters like dopamine on the basis of a pH gradient between the inside of the synaptic vesicle, which the cell spends energy to make acidic, and the surrounding cytoplasm, which is buffered at a comfortable near-neutral pH. The rest basically flows from acid-base equilibrium theory, e.g., Henderson-Hasselbach equation.

So the reason why amphetamine causes dopamine release from synapses is that amphetamine is a stronger base than dopamine, and competes with dopamine for the finite storage capacity of tiny synaptic vesicles. Now there’s more dopamine in the cytoplasm, which seeps out of synapses through DAT.

A hypothesis is only as good as its testable predictions. To wit, 22 years after the original weak base paper was published by Sulzer & Rayport, a group led by Teja Groemer and collaborators showed – coincidentally also in the pages of Neuron – that antipsychotic drugs, which are also have a pesky amine, are subject to weak base forces; are accumulated in synaptic vesicles; are released into the synaptic space in response to neuronal activity. (If anyone’s interested, I started a running discussion of Groemer’s paper on my website, which includes comments by Groemer himself, here).

To be totally fair though, the notion of weak base accumulation may have been new to the field of psychopharmacology in 1990, but Christian de Duve, the discoverer of the organelle called the lysosome, appreciated back in the 1960s that anti-malarial compounds, e.g., chloroquine, which accumulate in acidic organelles, are also weak bases.

Like a lot of powerfully simple ideas in biology, they seem to be forgotten and re-remembered over and over again.

So why did I email Sulzer in the first place? On top of all the weak base considerations, there’s another facet of amphetamines that actually distinguish them from their natural counterparts. They are more hydrophobic. Go back to the above comparison figure. You’ll notice that the catechol group in dopamine is plain vanilla benzene ring in amphetamine. By way of a culinary analogy, that means that given a choice, amphetamine is much more likely to choose oil over water if dissolved in vinaigrette.

My lab has shown that the antidepressant sertraline/Zoloft accumulates in yeast cell membrane with both rapid and slow-acting effects of yeast cell physiology. In 1994, Sulzer’s lab showed that methamphetamine, which has an extra methyl group vs amphetamine, causes gross changes in the endocytic pathway of cultured neurons, including the induction of autophagy, which my lab showed is induced in yeast cells treated with Zoloft, and so may be a general property of what hydrophobic weak base drugs. I’ll review the literature on the direct effects of amphetamines on membranes in my next post.

So where exactly are amphetamines accumulating inside brain cells, and what is the physiological significance of this accumulation? Seems like a question ripe for Crowdsourcing Discovery…

Here is the previous post on perks and price points.

In this post, I’m asking for feedback on project perks and price points. A few weeks ago, I asked my Twitter followers what would motivate them to fund a basic science crowdfunding project, and they dutifully replied.

Responses fell into two camps. On the one hand, scientists approached the question as though they were on a grant review panel: is the science justified; is the experimental design technically sound; is the team experienced and competent; is this project likely to receive support from traditional funding sources?

On the other hand, non-scientists were more concerned with the big picture: will there be a tangible outcome that laypeople could understand; is the science engaging and interesting; do the scientists present a compelling narrative?

The trend toward openness and increased collaboration is picking up steam in academia. To reflect the burgeoning new world order, here are the proposed Open Science rewards for our meth project:

Free: All project data will be deposited on the data-sharing site figshare. Frequent blog and video progress reports will be posted on the online home of the project, perlsteinlab.com. The “crowd” will be formally acknowledged in any resulting scientific publications in a manner that reflects their contributions.

$10 or higher – A follow and shout out from the project Twitter account, and a public acknowledgment of scientific citizenship on perlsteinlab.com.

$25 or higher –A 3D laser printed molecule of methamphetamine, and an open-invitation to be a collaborator on figshare.

$50 or higher – A periodic, signed spending report will be emailed to you, and you’ll have the opportunity to ask a question about the project on Skype or G+.

$75 or higher – You’ll be invited to weekly “Open Lab Hours” on G+ to discuss project updates.

$100 or higher – You’ll get a hearty thanks in person, and the opportunity to talk science over a round of beer or glass of wine at a NYC watering hole one night after work, or when you visit NYC within the next 6 months.

$200 or higher – Invitation to a brainstorming session at a NYC communal workspace, e.g., Wix Lounge. We will discuss the project, and how to scale up an open approach to basic research crowdfunding.

$500 or higher – Get access to a Dropbox account where you can deposit input. We have final cut but we broadly invite thoughtful, constructive feedback.

$1,000 or higher – Attend up to 2 lab meetings during the project and 1 publication brainstorming session at the end of the project. You will also receive access to a Google Doc during the manuscript writing stage. Supporters who contribute substantially to the final manuscript may receive co-authorship.

Please let me know what you think about the updated perks and price points.

The next post in this series addresses what we already know about how amphetamines work.

And here’s the previous post, which is a project overview of Crowdsourcing Discovery.

For this project, my lab at Princeton, which has spent the last five years developing a new evolutionary approach to studying the effects of psychoactive drugs, will team up with the Sulzer Lab at Columbia Med School, a leader in the study of amphetamines for over twenty years. We will use the decades-tested technique called autoradiography, combined with high-magnification electron microscopy, to resolve once and for all where amphetamines accumulate inside brain cells.

In the next post, I will describe the nitty gritty of the science behind the project, as well as our team, in greater detail. For the remainder of this post, I want to put our proposal into context. To date, no one has achieved our stated goal in the science crowdfunding space. In fact, breaking through the $10,000 barrier is no small feat, as I was recently gently reminded by Jai Ranganathan, co-founder of the awesomeness that is The #SciFund Challenge. And even then science crowdfunding successes have come in the least cost-intensive corners of the natural sciences, like ecology, archaeology and anthropology.

So how did we arrive at $25,000, given that it’s an order of magnitude higher than the average successful science crowdfunding bid? Besides the unpersuasive argument that someday, someone will succeed in raising that kind of dough and why not us, the simple answer is the cost of labor.

In contrast to previous science crowdfunding campaigns in the $500 – $5,000 range, to demonstrate scalability in the life sciences we must not only cover the inflated costs of consumables and equipment, but we also have to pay someone to carry out the research. As I described in a series of Open Budget posts last month, a full-time, Masters-level research technician with 3+ years of experience will run you between $5,000 – $6,000 per month.

So is our goal realistic? In order to make this process data-driven from the get-go I’ve collected numbers on a few fully funded (and then some), biology-related SciFund projects from the last year, which provide a decent apples-to-apples comparison:

There were multiple paths to victory. For example, it made it much easier if you did well in the champagne demographic ($50-$200), as was the case for Siouxsie Wiles’ Evolution in Action project (second from left), which met its goal within the first two weeks and then kept chugging along, ultimately over-raising by 170%.

The more common outcome was to crush it, at least in terms of raw totals, in the beer demographic ($20-$50). Take a look at Kristina Kilgrove’s Ancient Roman DNA project (rightmost), which raked in $10,171, or 170% of the stated goal of $6,000. Out of 170 crowdfunders, 94 donated in the $20-$50 range, but this sum only amounted to around 1/3 of the total. The Lexus demographic ($250+) was key to her success, and constituted close to half of her total.

How did she do it? A great summary of her strategy explains that she targeted several different receptive audiences with personalized messages, and then invested several hours of online promotion each day of the campaign. But she was also fortunate to receive favorable mainstream media exposure (here and here), which, if I had to guess, enabled her to snag some of those critical angel funders.

If I extrapolate from Kilgrove’s donor distribution, we’ll need 400-600 funders to reach our stated goal. Friends and family are the first stop on the crowdfunding whistle stop tour! I have over 600 friends on Facebook but I doubt 90% of them would fund a basic science project, so FB alone won’t cut it. I’m just south of 1,000 Twitter followers, and my following is enriched for Open Science devotees, academic and industry scientists, and science writers to whom our crowdfunding pitch would be appealing. Based on one analysis by Jarrett Byrnes, the other co-founder of The #SciFund Challenge, a healthy Twitter following certainly helps. And of course I and the other team members will tap our professional email and LinkedIn networks, but will that be enough?

The wild card is whether we can level up, and leveling up can take many shapes in the Internet Age. Kilgrove’s Ancient Roman DNA project received the equivalent of a massive retweet when her story was written up on a CNN science blog, which is a testament to the reach of professional science bloggers like Ed Yong (@edyong209), who penned that piece. Based on my own experiences poring over Google Analytics traffic data for my lab website, if you want orders-of-magnitude vs. several fold changes, it really helps if someone with > 100,000 followers mentions you. The best example was when a massively followed science writer retweeted a link to one of my blog posts last month:

Even though I reached a theoretical audience size in the 10,000s as a result of my own tweeting and my followers’ retweets (first spike), a single mention by a massively followed tastemaker brought more unique visitors to my site in a 48-hour span than several weeks worth of garden-variety referrals. So the lesson for me is if I want this crowdfunding project to succeed (and possibly even break through the stated goal), I will need to get noticed by at least a few tastemakers.

Please stay tuned for the next post in this series, which deals with project perks and donation price points.

And here’s the previous post on the genesis of Crowdsourcing Discovery.

Meanwhile, back at the Los Alamos National Laboratory, a band of theoretical high-energy physicists traded the printed page for a web page, specifically a preprint server called arXiv. Paul Ginsparg, the creator of arXiv, chronicled the Internet-driven transition in a fascinating interview in Nature last year. Reflecting on the big picture, Ginsparg recognized that scholarly publishing is an evolutionary process:

“It is heartening, 20 years later, to see a stable and successful arXiv, running some of the original software and providing services to a community nearly a thousand times larger than expected. But at some point a thorough overhaul will be needed to keep pace with new online trends and opportunities.”

Unfortunately, not all scientists got the memo. In particular, the non-quantitative biologists who wouldn’t be caught dead posting to arXiv, either for fear of being scooped by a competitor, or because posting a preprint can disqualify the manuscript for consideration in some high Impact Factor journals, e.g, Cell, or just because preprints “don’t count for anything” in The Tenure Games.

But thanks to this cool new science broadcast and alert system I use called Twitter, I can see “new online trends and opportunities” in real-time, as they speciate. Two examples come to mind: the new membership-based business model of publishing startup PeerJ; and F1000 Research flipping peer review on its head.

I’m firmly in the “publish, then filter” camp. Actually, it’s safe to say I’m in a splinter group called “self publish, then filter.” That’s why I refer to my lab website as a self publishing platform. I know for a lot of people the term “self publishing” is a sticking point. Let me be clear: a Tumblr doesn’t cut it!

My site is really just a customized WordPress theme implementing smooth API integrations and responsive design principles. In turn, WordPress is really just a content manager for my research and blogging outputs, which I can make citable and URL-decay proof using tools like Direct Object Identifiers, DOIs, on the data-sharing site figshare. To stimulate speciation, I decided to make the base code freely downloadable so that others can adapt my template.

Two serious critiques to science self publishing that I’ve come across on Twitter are 1) long-term storage, and 2) URL decay prevention. However, there are existing solutions to these problems. For example, CLOCKSS, and possibly other dark archives, will help in the crowdsourcing of content preservation over centuries – millennia, I hope. And by implementing the aforementioned DOIs, URLs may come and go, but the associated metadata never die.

And if you’re still that concerned about preserving your content, then maybe you should create a scholarly “will and testament,” spelling out the technologies that will ensure stable, intergenerational transmission to all your scores of future devotees. For now, I’m content to have my content scraped for the real gems that are worth passing down.

There has always been an energetic cost to content preservation. Today, it’s mostly a hidden cost from the perspective of individual academic scientists, though don’t tell that to university librarians, even at Harvard’s library. My argument is it’s only a matter of time before someone, or more likely some collaborative team, successfully innovates a better and cheaper solution, as has been the case repeatedly throughout the history of invention.

George Church is onto something with that DNA repository, maybe.

Last week several key opinion leaders in the field of genetics, including a collaborator at Princeton, Leonid Kruglyak (@leonidkruglyak), made a splash after posting papers – well, manuscripts technically – on arXiv, the preprint server of choice for physicists and mathematicians for several decades running.

Before that, an announcement last month of the partnership between Faculty of 1000 and the data-sharing site figshare heralded the beginning of an era of disruption in scholarly publishing (despite Kent Anderson’s implacable dyspepsia). Before that still, and in the realm of actual policy making, the Finch Report commissioned by the UK government gave its blessing to open access to all publications whose research is supported by public funding.

And I would be remiss if I forgot to mention the late Spring social-media-roots campaign spearheaded by John Wilbanks (@wilbanks), which netted in less than two weeks over 25,000 signatures for a White House e-petition calling for the policy just embraced by the UK. (Though months later it’s not clear the that White House has done anything about it).

But in the trenches, I have my concerns about the sustainability of Open Science outside of the Twitter echo chamber. As I learned the hard way from the Occupy movement, adrenaline only takes you so far.

Of course I wouldn’t be writing this post if I thought the future will be all gloom and doom. It is my belief that Open Science will jump to a higher orbital once more scientists are convinced that communicating their science with other scientists online, ideally in blog form, is not only worth their time but also an outreach mitzvah. In a tip of the hat to the “quantified self” movement in medicine, I present you with quantified self publishing, courtesy of Google Analytics: ‘cause nothing appeals to the academic’s self interest more than numerical approbation in the form of lab website traffic data.

It obviously helps if you have a shiny new self publishing platform to play with, as I do with Perlsteinlab.com. My site formally launched on June 28^th, and I took a snapshot of site traffic data after one week in the wild. Now that more than a month has elapsed, I wanted to see if I’d actually amassed an audience, however small; and whether my promoting the site almost exclusively via tweeting is effective. What follows are Google Analytics data that attest to both.

First, here’s a nice summary of the basic suite of Google Analytics metrics from the time interval spanning 6/28 – 8/11:

Doesn’t seem too shabby, though the low pages/visits suggests that on average most of my readers are landing on the home page and then clicking through to only one piece of content (or vice versa). Remember, I’m luring visitors to my site principally by tweeting (referral) and emailing (direct), as evidenced by this breakdown of total site traffic:

So how did all those visits shake out day by day? Here’s a plot of unique visits per hour over the ~50-day observation period:

As expected, there was a flurry of immediate post-launch activity, followed by a long, essentially flat stretch punctuated by blips here and there, and then an upswell of traffic over several days toward the end of the observation window. To make a long story short, the blips were caused by retweet events. And the massive spike, which dwarfed even my post-launch buzz, was triggered by a tastemaker retweet, i.e., a retweet from someone with a massive number of followers.

First the garden variety retweets. On July 12^th, I tweeted out my statement of teaching philosophy, which is part of my application package for junior faculty searches. Two people, one a follower of mine and another a follower of this follower, each retweeted my link, as shown here:

This plot is representative of most days on my site, in that I received approximately 100 unique visits per day, and roughly 10 visitors per hour. The tweet-driven spikes are transient – never lasting more than a few hours and usually just one hour.

However, things look a lot different when a retweet comes from a massively followed tastemaker. Case in point is Ben Goldacre, a British science writer who has over 200,000 followers. He tweeted the following:

This one innocuous tweet brought tons of eyeballs to my site, as evidenced here:

Instead of the doubling of traffic that usually resulted from a garden variety retweet, Goldacre’s mention caused a logarithmic jump. The U-shape corresponds to the overnight hours on the East Coast. (Remember, Goldacre is on Greenwich Mean Time). What’s more, the second peak has a “long tail,” as traffic data go.

There’s a lot more where this came from, and I’ll be posting future updates as new and interesting patterns emerge. In the meantime, I would love to get feedback from others who’ve analyzed traffic data for their science blogs, or from academics who are contemplating starting their own self publishing journey.

Also, this exercise in disclosure goes hand in hand with recent Open Science gestures by colleagues in the field of Ecology, @ethanwhite and @ctitusbrown, among others, many of whom have uploaded both successful and unsuccessful grant proposals to the data-sharing portal Figshare.

I left off last time with an enumeration, in broad strokes, of monthly burn rates over my 5-year appointment. The most expensive line item in my budget is, not surprisingly, the cost of labor. Out of the total $1,000,000 “under management,” I spent $529,249.82 (53%) on personnel, specifically Masters-level research associates (aka technicians). However, labor costs fluctuated from month to month, as shown here:

Fiscal years are color-coded; there is missing data in FY 2008 (orange) and FY2011 (yellow).

And what’s the second largest line item? At ~40% of total expenditures, you guessed it: materials and supplies. For a more refined view of non-personnel spending, I’ll use the numbers from FY 2010, which represents the halfway point of my fellowship, as a case study:

Salary and benefits for two full-time technicians came out to 42% of my total FY 2010 budget, lower than other years because I made a one-time purchase of a $40,000 instrument in September 2009. Core facilities and overhead totaled just over 6%, and this included use of a departmental electron microscope. The cost of academic conferences was less than 1%. And I wound up with a non-trivial surplus (9%) that rolled over into the following year’s budget.

Here’s a plot of every Materials & Supplies purchase made over the course of FY 2010, with the exception of the aforementioned $40k instrument (note log scale):

Most purchases fell into the $100-$1000 range, and there weren’t any obvious seasonal variations. The ~$10k spike in early July 2010 was for a shaking incubator. Close to 60% of material and supplies were purchased from just 3 mega-vendors: Fisher (18%), Sigma-Aldrich (14%) and Invitrogen (26%).

Purchasing was a bit streaky in that the lab would sometimes go a week or two without any, so to smooth out the data I graphed the same numbers on a per month basis, resulting in an average of $4,165/month:

I hope sharing is as contagious as everyone claims it to be. If so, the scientific community will be one step closer to crowdfunding on a scale that rivals the sums currently provided by the NIH and NSF.

The Abstract opens: “We propose to demonstrate in this paper that the chromocytes [red blood cells] of different animals are covered by a layer of lipoids [lipids] just two molecules thick.” You don’t encounter an understated turn of phrase like “propose to demonstrate” in contemporary biology papers, because so much published today is a bold proof-of-concept, which, ironically, often doesn’t survive the test of time or replication.

The difference between modern and classical papers glares in the Methods section. Obviously, one has to control for the fact that experiments of yore were technically less challenging. It’s the loss of narrative flow and any sense of first-person action in explanations of today’s experimental protocols that is so distressing, and I would argue it contributes to the lack of reproducibility lamented by many biologists. By contrast, consider this passage from Gorter and Grendel’s paper: “After several extractions, the acetone was filtered into a glass beaker and the liquid evaporated on a water bath. This procedure was the most difficult part of the operation because loss was very liable to occur at this time.” Such a line in a paper today would be tantamount to an admission of guilt rather than an honest account of manual experimentation.

Finally, what you see is what you get in terms of data presentation. No supplementary files. No byzantine 12-panel figures. Just an easy-to-interpret table:

Granted, there are few moving parts to the experiments performed by Gorter and Grendel and so the paper is a brisk 5 pages. But my larger point is that older papers are entirely self-contained. Perhaps we should resurrect the tradition of writing shorter, intellectually crisper papers – more communications, fewer tomes.

Which gets me thinking: on average, how much does it actually cost to run a small academic lab? There’s not much out there in the blogosphere in terms of budget datasets – if there are, please provide links below in the comment thread – so I decided to create my own. I’ve been a Lewis-Sigler Fellow at Princeton since 2007, where I’ve managed a $200,000/year budget and overseen a two-employee lab for the last 5 years.

By releasing these figures into the wild, I hope to stimulate several conversations, first and foremost, regarding efficiency. To clarify, I don’t exclusively mean preventing waste arising from ill-advised equipment purchases. Rather, I mean tackling the asymmetric marketplace in research transactions, along the lines of what science startups like Science Exchange and Assay Depot are doing. Second, if crowdfunding is ever going to compete with the government monopoly on funding, the global research community needs to share their spending habits in order to expose non-obvious inefficiencies, and to illuminate unmarked paths toward greater efficiency that become apparent when comparisons are made across many labs, regions and disciplines.

Okay, with that preamble out of the way, let’s talk numbers.

Each month, I receive an electronic statement detailing expenditures – in the parlance of commerce, my “burn rate.” Here I graphed my lab’s monthly burn rate over the length of my soon-to-be-completed appointment at Princeton:

I’ve color-coded fiscal years (July 1 – June 30) to make comparisons easier. Aside from a spike in September 2009, when I made a one-time > $40,000 instrument purchase, my burn rate has been fairly consistent. The single most expensive line item is – you guessed it – personnel, which consumed well over half of my budget. I should note that my lab is atypical in that I’ve employed two full-time Masters-level research associates (technicians). Experienced technicians are more expensive than trainees, e.g., graduate students and postdocs. Based on these numbers, one can begin to understand why the pitched cries over protracted apprenticeships as a form of indentured servitude have a basis in reality.

Interestingly, there appear to be lulls when my lab spent less money than average, and, conversely, volatile periods when my lab spent more money than average. To see if there are seasonal patterns, I graphed the burn rate data by month:

The Fall is the most volatile season, followed by a stretch of below-average consumption in the Winter and Spring, and concluding with a modest ramp up in the Summer. The Fall volatility is mostly driven by an artifact of my appointment start date. I attribute the under-consumption of Winter and Spring to thrift and my desire to maintain reserves, even though my funds roll over. However, by May it was usually clear how much money remained, and I became bolder with purchases in the Summer, not to mention that Summer is the academic conference season, and these meetings are expensive.

In Part 2, I examine my lab’s spending patterns at higher temporal resolution, and take a closer look at the costs of personnel vs. consumables/reagents/services.

The Perlstein lab generated these unpublished images of red blood cells treated with different concentrations of the the SSRI antidepressant sertraline/Zoloft®, which is a cationic amphipath. Sertraline accumulated in the inner monolayer of the red blood cell plasma membrane, causing the plasma membrane to curve inward, culminating in cup formation. The degree of cup formation is proportional to sertraline concentration, beginning at single-digit micromolar concentrations up to a critical concentration at which membrane lysis occurs.

Don’t worry. The extremely low serum concentration of Zoloft in people taking the drug means that your red blood cells are just fine.

As an experimentalist by training and by disposition, I tend to wade into complex problems. Before rushing headlong into a new experiment, I remind myself of that kernel of wisdom imparted to me by my mentors: do a literature search first!

So that’s exactly what I did. And like a lot of great ideas whose time has come, convergence is commonplace. After poking around the Internet I discovered that someone’s already tried the experiment, but it didn’t work and I needed to figure out why.

For the completely uninitiated, I penned a few general thoughts on crowdfunding here; for meatier fare, I recommend perusing the blog of the SciFund Challenge, a science crowdfunding consortium of mostly ecology-centered projects organized by Jai Ranganathan (@jranganthan) and Jarrett Byrnes (@jebyrnes).

I think it’s safe to assume that most people are familiar with Kickstarter, the Abraham of crowdfunding sites that begat numerous descendants. What might be less known is that the science crowdfunding space is currently convulsing with competition, resulting in several different portals: RocketHub, which hosted the SciFund Challenge; Petridish, which was founded by a former venture capitalist, Matt Salzberg (@mattsalz); and Microryza, a Seattle outfit whose motto is essentially “by scientists, for scientists.” The major difference between these portals is whether they abide by the all-or-nothing rule: Petridish and Microryza do, but RocketHub lets you keep what you raise.

The case study I want to discuss here is an underfunded (and, alas, unsuccessful) project on Petridish called “Listening for cancer: Early detection using laser ultrasound.” This project was proposed by John A. Viator, an Associate Professor of Biological Engineering and Dermatology at the University of Missouri. Viator asked for $7,500 to purchase equipment and provide additional financial support to trainees in his lab, but he raised $2,415 from 40 backers, and according to all-or-nothing house rules, he received zero funds.

I spoke to Viator a few days ago and we played Monday morning quarterback. What follows is a recap of our conversation, relying on notes I made immediately after the call. Viator told me that he found out about Petridish from a colleague. Viator noticed right off the bat, as would any biomedical researcher who runs even the smallest of groups, that the typical requested funds, ranging from $500 – $5,000, would not make a dent in the budget of most academic labs. For a benchmark, my two-person lab at Princeton costs on average $15,000 to run – each month!

I then asked Viator about his interaction with the university grants administration office, because I suspected that they would be a bottleneck. Viator wanted to be able to say to potential backers on Petridish that their contributions would be tax-deductible, but that required accounting maneuvers by the grants office. Not surprisingly, inertia and/or inexperience with crowdfunding in the grants office meant Viator’s requests languished for several weeks. Undaunted, Viator plowed ahead.

As an aside, I broached the possibility of using a lab’s gift account to store the funds, and I would love to get comments/feedback below from other university scientists who are familiar with the policies of their home institutions.

Finally, I wanted to know if he could do it all again, what would Viator do differently the second time around. Above all else, he said that his proposal didn’t receive enough press coverage, and that next time he would conduct a more proactive media campaign. I remarked that perhaps his video call to action could have been more targeted and shorter. Most people recommend 2-3 minutes top, his was ~7 minutes long because it was a talk he had given that was on hand before he launched on Petridish. When you watch the video you get the sense that this is a scientist who gives a good talk and knows his science backwards and forwards, but I’m not sure that alone convinces the greatest number of prospective funders to open their wallets.

I’ll stop there and open the floor to discussion.

The first bump around day 65 was caused by my lab website going live. A review of Chen et al by neuroscience blogger The Cellular Scale (@cellularscale) stimulated the second, slightly larger bump around day 70. For ease of viewing, here’s the same data graphed as daily pageview counts (note log scale; blank slots are missing data):

I’ll note several things. First, thousand-pageview days are creatures of early buzz; the most my paper could muster since was 224 pageviews in response to that blog review. Second, my lab website going live seems to have stabilized the gradual slide in pageviews that set in after the first month. I see evidence of this new readership from Google Analytics traffic data for my lab website.

Now I’d like to shift gears and examine closely one article-level metric in particular, namely the ratio of HTML views to PDF downloads, hereafter “HTML/PDF.” Recently, Martin Fenner (@mfenner), the PLOS article-level metrics guru, and I exchanged a few thoughts on what HTML/PDF might mean in terms of scholarly impact.

Let’s start with my paper’s HTML/PDF, which is 15 (6748/446). So for every 15 HTML views, only one reader downloaded the PDF. In an ideal world, you might imagine that HTML/PDF would be close to unity. But that only makes sense if the sequence of events is: 1) land on article page, 2) download article PDF. Turns out that sequence describes academic readers pretty well: using traffic data from PubMed only, the HTML/PDF for my paper is 1.1 (63/57).

But most readers of my paper don’t appear to be academics. Both Fenner and I (and I’m sure others) interpret HTML views as calling cards of casual readers who might exit the article page after perusing the abstract or after scrolling through the text and realizing that the material is way over their head. On the other hand, a reader who clicks through a second time in order to download the PDF is probably an engaged reader, and I would argue more likely to be an academic or “expert.”

Now let’s try to put my HTML/PDF of 15 into context by looking at a larger sample of PLOS papers, controlling for topic area, in this case cancer genetics.

(Image provided by M. Fenner)

The correlation between HTML views and PDF downloads is stable over a 10-fold range. But the HTML/PDF is much closer to one than my paper, indicating many fewer casual readers of cancer genetics papers. Okay, that’s nice. But what about an apples-to-apples comparison? Here’s a plot comprised of PLOS papers in my specific topic area:

(Image provided by M. Fenner)

Interestingly, the HTML/PDF is higher for this group of papers than the cancer genetics papers. Therefore, pharmacology papers may be more likely to pique the interest of non-experts. But is there any way to test to that hypothesis? Take a look at the off-diagonal outliers. The one furthest to the right is my paper. The size of the circles is an indication of how much interest was generated on Facebook, the Mecca of casual readers.

The next stage in the post publication life cycle of my paper is the (long?) wait for the first citation…

Briefly, the neurogenesis hypothesis states that structural defects in the hippocampus are at the heart of major depressive disorder (MDD). The origin of this hypothesis can be traced back to rodent experiments by Ronald Duman and Eric Nestler in the late 90s, which sometimes elicits a chorus of skeptics decrying “mouse psychiatry!” However, other groups have independently replicated their core observation in subsequent years: modest albeit statistically significant hippocampal neurogenesis in rodent brains chronically exposed to antidepressants or electroconvulsive therapy.

Pivoting to humans from rodents, the first prediction of the neurogenesis hypothesis is that people suffering from MDD would have smaller hippocampi. Examination of postmortem samples has borne out this prediction. However, there’s variance in published studies on human hippocampal volumes. One experimentally validated source of variance, at least in rodents, is the animal’s age; older individuals have lower baseline levels of neurogenesis. Presumably other factors are at play, too.

The second prediction of the neurogenesis hypothesis is that people suffering from MDD but taking antidepressants (ADs) would have normal hippocampi again. Since there’s no way to measure neurogenesis in live humans yet, neurogenesis researchers are again limited to postmortem samples. All the usual caveats apply when examining postmortem samples, and Boldrini et al do a reasonable job of controlling for age, prior substance abuse history, etc. Postmortem samples are clearly not ideal experimental substrates but they shouldn’t be dismissed out of hand given the mountain of rodent-based data.

Okay, with the introduction out of the way, let’s dive into the paper proper. The authors augment and tweak their previous study design by including matched triplets comprised of a non-psychiatric control sample, an untreated MDD sample, and an antidepressant-treated MDD sample. The data set is summarized in Table 1 of the paper:

Not being a neuroanatomist myself, I had to defer to the authors’ prudence in the selection of protein markers of neurogenesis. They chose a protein nestin, which labels neural progenitors cells (NPCs) and the microvasculature, a double whammy of sorts because you also get spatial information on angiogenesis. (Cell types can be distinguished to a first approximation by morphological features, e.g., axons) The authors gained more confidence about which cells are newly born by staining for a nuclear protein called Ki-67, which is an accepted marker of cell division.

The authors spend a fair amount of time explaining the pros and cons of this marker versus that marker because their entire case rests on specificity. For example, glial cells re-express nestin after stress, and so they used a glial specific marker to discriminate between neuronal nestin and non-neuronal nestin. That said, others in the neurogenesis field don’t think nestin is the best marker of neurogenesis, and instead swear by doublecortin.

Moving on, the authors compared the untreated MDD hippocampi to the antidepressant-treated MDD hippocampi, and found evidence for neurogenesis and angiogenesis in the same locations. For example, on average 85% of NPCs (brown/purple blobs) appear to colocalize with capillaries (brown tubes). Quantification aside, the putative neurogenic and angiogenic effects of antidepressant treatment are discernible to the naked eye, as shown here in Figure 4 of the paper:

The coincidence of neurogenesis and angiogenesis makes sense physiologically, and it’s seen in other developing tissues, as well as in tumorigenesis. But as was stated above, cause vs correlation cannot be disentangled using a portmorten approach. There is a contrarian view that neurogenesis may be part of a general stress response, and may not actually constitute the mechanism of the antidepressant response.

Apropos, this paper has a wrinkle that you’d miss if you didn’t read the Discussion section carefully. Close examination of the dentate gyrus, a region of the hippocampus containing “neurogenic niches,” in non-MDD samples vs untreated MDD samples revealed no differences in volume, though I stated above that there is considerable variance in this measurement across independent studies. Here’s what the authors had to say:

“The relationship between MDD, neurogenesis, and angiogenesis is not known. Deficient adult hippocampal neurogenesis is hypothesized as contributing to MDD pathogenesis (10), but we are not finding an effect of MDD on NPC number or vascularization, as we previously reported…it is possible that later stages of NPC maturation or cell survival are compromised in MDD. According to this hypothesis, NPCs would not be fewer inMDDversus control subjects, but neuroblasts or mature granule cells would be. Moreover, a deficit of neuropil could contribute to smaller DG in MDD. We are currently testing these hypotheses.”

It is advisable not to bow at the altar of Serotonin or Neurogenesis. The answer is not something in between, just more complex.

To recap, here are the take-home messages of, and response rate data for,  my sociological experiment:

1. 8 out of 166 (~5%) university professors agreed to move post publication review from the email world to my article’s comment thread
2. Professors with whom I had prior contact replied at nearly thrice the rate of strangers
3. The 8 solicited professor comments placed my article in the top 0.1% of PLoS ONE articles
4. Commenting is not a favorable process, to borrow a phrase from thermodynamics; one must supply energy to make the reaction go forward

Now let’s dig into the numbers. There were 122 non-responders. 12% (15/122) of “no replies” were actually autoreplies, or my email bounced outright because the recipient’s email address is defunct. In the food processing industry, they call unavoidable losses “shrink.”

What about the 14 professors who engaged in multiple rounds of email colloquy but never commented? Why didn’t these professors surmount the seemingly tiny hurdle separating emailing from commenting? It wasn’t for lack of interest, because two of the 14 non-commenters volunteered to chat over the phone, which is certainly worth more than a comment. Some professors may prefer private over public correspondence, period.

In addition to the eight successful cases of email-to-comment conversion, there are six other comments on my PLoS ONE paper for a total of 14 comments. Two of those comments don’t really count: one is a press coverage digest generated by PLoS ONE, and the other one is a digest of relevant tweets about the paper that I posted. So that leaves 12 bona fide comments.

Only one of the 12 was unsoliticited and anonymous, and, not surprisingly, a bit odd: “I followed this discussion from In the Pipeline. I’m not sure this is related, as the information seems self-reported, and therefore anecdotal. The sample size is quite small, as well. The reference is from a study published in the Canadian Journal of Psychology about the rare side effect of yawning as a cause of orgasm in female study patients taking clomipramine, which can be found at the link below.”

“In the Pipeline” is a blog maintained by noted science blogger Derek Lowe, which brings me to the most interesting comment of the 12. Lowe reviewed my paper, and his post garnered a whopping 25 comments! And it isn’t just quantity. The “In the Pipeline” comment thread is more complex than my paper’s comment thread, with later comments responding to or amplifying earlier comments, and multiple posts by the same commenter, etc. Interestingly, most of the commenters posted pseudonymously, a custom in many online comment forums, and yet the content didn’t suffer one iota.

However, the “In the Pipeline” comments are invisible in the official PLoS count, and, regrettably, isolated from the discussion already in progress on the PLoS site, even though after the fact Lowe was kind enough to copy and paste them into a comment for me. Alas, I don’t have a good answer to comment-thread balkanization. I wish it were as easy as highlighting text in an email, and then clicking a “send to comment thread” button that would instantly cross-publish at the PLoS ONE comment thread, or any science comment thread of one’s choosing.

With more online comment forums adopting a common infrastructure, e.g., DISQUS, the fragmentation should lessen over time (I hope). Yet even with a stable commenting identity that is portable across a wide swath of blogs, there’s no way to recreate overnight the informed, loyal readership of “In the Pipeline,” which took years to cultivate I’m sure. But I’m not asking for miracles, just more efficient coordination, and where that fails more technology that facilitates cross-blog pollination.

So how many comments are enough? 25? 100? 200? Put another way, what is the natural comment ceiling for a given research article? Obviously the answer is complex, and depends on the popularity of the research area, the reputation of the authors, the reputation of the journal, among others. However, as I argued in my everyONE post, the 90-9-1 engagement rule is a good rule of thumb. 90-9-1 also jibes with my paper’s article level metrics.

Let’s assume that the 440 PDFs downloads of my paper are a proxy for university- and industry-based readers, as I argued in a previous post, and as others have recently argued. For that many PDF downloads, I would expect several dozen comments. If you add the 12 bona fide comments residing at PLoS ONE to the 25 comments residing at “In the Pipeline,” that comes out to 8.4% (37/440), which is almost exactly equal to the aforementioned email-to-comment conversion rate.

Here, amiodarone-treated yeast cells accumulated the drug in vacuolar membrane, causing cup formation. Bilayer expansion is evident at the concavities as white stretches in the membrane.

In order to eliminate this blind spot, psychopharmacologists have to stop ignoring the direct effects of ADs on membranes, even if some of these effects are observed in cellular models in the lab or at doses that are ostensibly “too high” to be clinically relevant. (For more on the specious “too high” concentration argument, see my recent post about worms on Prozac). That’s obviously easier said than done.

So let’s start where we have consensus. First, everyone agrees that ADs (tricyclics + SSRIs) are slow acting. Second, everyone agrees that the effects of ADs on serotonin reuptake are fast. The uncertainty arises when atttempts are made to reconcile the rapid inhibition of serotonin reuptake by ADs in all experimental model systems examined with the slow-to-emerge neurogenesis induced by ADs in specific brain regions of chronically dosed rodents, changes which presumably also occur in the brains of people taking ADs.

Is that time lag the result of a slow cascade of physiological adaptations triggered directly by increased synaptic serotonin concentrations? Or are there non-serotonin-based, slow-acting mechanisms involved as well?

One important clue comes from a simple electron microscopy study described in an elegant paper by Renate Lüllman-Rauch’s group from 1979, and published in Biochimica et Biophysica Acta (my favorite journal name to say aloud). The authors used transmission electron microscopy and the freeze-fracture technique to examine various tissues, including retinas and adrenal glands but alas not brains, of rats that were chronically treated with the phenethylamine drug chlorphentermine. Chlorphentermine was once used as an appetite suppressant, not surprising given its structural similarity to amphetamine. (As an aside, sertraline/Zoloft, a major focus of the Perlstein Lab, is actually a chemical descendent of amphetamine).

It was known back in 1979 that lipophilic weak bases like chlorphentermine induce the formation of “myeloid bodies,” or densely packed, often concentric, arrays of phospholipids. The mechanistic interpretation is that acidic phospholipids are bound to amphipathic weak base drugs, and the resulting drug-lipid complexes are no longer recognized by the digestive enzymes (e.g., phospholipases) that normally break down phospholipids in the maintenance of cellular membrane homeostasis.

I’ve taken the liberty of reproducing several stunning electron micrographs from the Lüllman-Rauch paper, which tell better than a thousand words what chlorphentermine accumulation looks like in animal tissues.

Two myeloid bodies are indicated by orange arrows. They are located inside lysosomes, degrading organelles where all cellular components, including phospholipids, are recycled. High-resolution magnifications revealed the fine structure of those inclusions, which are indicative of lamellar phase phospholipids:

The authors also observed non-lamellar (i.e., non-bilayer) architecture, namely hexagonal phase lattices:

My lab has produced very similar images of hexagonal lattices, but from yeast cells treated with the SSRI Zoloft. These images are obviously 2-D slices. So what is the 3-D structure of those chlorphentermine-induced inclusions? Here was an artist’s rendition of the freeze-fracture data, revealing concentric bundles of hexagonal tubules of phospholipids:

Besides providing visually arresting proof of phospholipidosis in response to chronic chlorphentermine accumulation, the authors noted that phospholipid inclusions were different in different cell types, in this case retina vs. adrenal tissue:

“The above-mentioned biochemical and physicochemical knowledge supports the interpretation of the present findings, namely that retinal pigment epithelium prefers to accumulate the phospholipids in hexagonal rather than in lamellar phase. In other types of cells, which develop both types of inclusion bodies, the composition of stored lipids as well as the local conditions may be different such that both lipid phases, and additionally transitions between both, may occur.”

The question you’re probably asking yourself right now is, do antidepressants induce cell membrane changes in the human brain over long time scales, and is it required for their therapeutic efficacy? The short answer is no one seems to have looked carefully. In the next post on psychopharmacology’s blind spot, I’ll dig a little deeper..

Very similar multilamellar bodies were observed in the granules of human natural killer cells.

With that preamble out of the way, here are some snapshots of my site’s traffic data after a week and a half in wild. First, take a look at daily unique visitors. The site officially launched two Wednesdays ago, but I didn’t commence the social media roll out until the following morning. As a rule of thumb, Fridays and the weekend are traffic killers (the first trough after the initial surge). The 4^th of July presented headwinds after traffic flow resumed on Monday July 2.

The bounce rate seems high to me, but then again if I reflect on my browsing habits I can’t say I’m that surprised. Oh, and the returning visitor is probably me. I used two strategies to stimulate traffic. First, I emailed a direct link to dozens of academic contacts, as well as friends and family. Second, I shared a direct link on Facebook, and I tweeted a link to the homepage, and links to specific single page posts. I also shared and tweeted press and science blog coverage about my new site.

So does all that tweeting matter, as was recently claimed? Yes, in fact, I think tweeting makes a huge difference! Here’s a nice example.

You’re looking at visits to my site on Monday July 2. See the spike around 11AM? That was generated by a tweet sent by Dan MacArthur (@dgmacarthur), a geneticist in Boston with over 6,000 followers me about using the Mendeley API to host journal club discussions about papers of interest on my website (Apologies for the initial incorrect attribution). That tweet was retweeted by 13 people to 1000s of their cumulative followers. These Twitter-generated spikes were apparent each day last week. In fact, to show how important Twitter was overall, here’s a breakdown of all referral traffic.

Half of referrals are tweets. MedCity News, Figshare and BMJ Blogs wrote reviews of my site here, here and here, respectively. Obviously, favorable press coverage brings in a decent chunk of traffic.

That’s it for now! Thanks for your interest, and stay tuned for future updates as the site matures…

sciseekclaimtoken-4ff7b6827b61c

To warm up, I’ll present a beautiful little paper by Hiruma & Kawakami (Folia Histochemica et Cytobiologica 2011; Vol. 49, No. 2, pp. 272–279), which resuscitates dormant observations about the effects of weak bases on living cells. The authors measured the response of several essential cellular components to high concentrations of the water-soluble weak base 4-aminopyridine in primary cell cultures derived from dissected mouse dorsal root ganglia.

As shown by light microscopy (Figure 1, above), 4 mM 4-aminopyridine caused many internal compartments (vacuoles) to appear, e.g., crater-like structures in panel B, especially at the 90-minute mark. Washing away the 4-aminopyridine after a brief (minutes to hours) treatment caused all those induced vacuoles to disappear gradually. However, long-term (hours to day) treatment with 4-aminopyridine induced irreversible “vacuolization,” which was toxic to cultured neurons and non-neurons alike.

Christian de Duve, proposed a theory called lysosomotropism to describe the cellular accumulation of weak bases. It invokes the Henderson-Hasselbach equation to explain the skewed subcellular distribution of weak bases, which accumulate in acidic compartments, even synaptic vesicles.

So what, you say? 4 mM of anything you can buy from Sigma would eventually kill cells “by a non-specific mechanism.” What’s happening to other dynamic, essential cellular components besides acidic compartments during a 4-aminopyridine overdose?

Well, the authors did a few controls (Figure 3, below):

First, they verified that the 4-aminopyridine-induced vacuoles observed under differential interference contrast (DIC) are indeed lysosomes by using a fluorescent lysosomal protein marker. Next they showed that the mitochondrial and actin filament networks are basically unaffected by an acute 1-hr treatment with 4-aminopyridine (4-AP).

Then they performed a bunch of before-and-after 4-AP treatment experiments with LysoTracker Green, which “stains acidic compartments,” according to the manufacturer. I’m not exactly sure what to make of these data, other than that vacuolization is probably being conflated with the induction of autophagy.

The specificity pièce de résistance is Figure 5 (below). The V-ATPase complex, responsible for acidifying organelles, is essential for the initial vacuole induction and the ensuing cascade of physiological effects:

To be honest, parts of the discussion suffer a tad from lost in translation. Throughout the paper, I got tripped up when the authors referred to organelles within vacuoles. I assume they were talking about autophagosomes or autophagolysosomes. The fact that these words don’t roll off my tongue and I’m a native English speaker attests to the potential for confusion.

The authors concluded the paper with the following model:

“In addition, the contents of vacuoles are serous without proteins or amino acids and also without weak base. Thus, it is possible that vacuoles are formed by extrusion of H+ from acidic organelles along with water. Further studies are needed to prove this.”

I never quite understood what they meant by “serous vacuoles,” a phrase they repeated over and over. My interpretation is that they are seeing evidence of phospholipidosis. How weak base accumulation actually triggers phospholipidosis is a whole nother story…

I encourage whoever is interested to read the entire paper here. I’m hosting a discussion of it below. Who wants to break the ice?

Two weeks ago I was fortunate to attend the Kauffman Foundation’s first Life Science Ventures summit in San Francisco – #lsvs2012 for the Twitterati – which was hosted by UCSF and its entrepreneurial support arm, QB3. Like a lot of startup-related news these days, I first caught wind of the summit in March on my Twitter feed. The two-day, breakout session-filled program featured panel discussions led by key opinion leaders from the Bay Area startup cluster and elsewhere. For example, business model canvas guru Steve Blank (@sblank), founder of Strategyzer Alex Osterwalder, and the CEO of Genomic Health Kim Popovits all spoke about the grand themes and practical challenges of life science entrepreneurship in the Internet Age.

I learned way too much to recount it all here in a single blog post, so for now I’ll focus on one specific topic area that was of keen interest to me as an early stage life science entrepreneur, namely the Small Business Innovation Research (SBIR) and the Small Business Technology Transfer (STTR) programs. SBIR and STTR are basically taxpayer-funded technology transfer grants; in entrepreneur speak, an attractive source of non-dilutive funding.

Most academic trainees, and non-tenure track researchers like myself, are familiar with the alphabet soup of taxpayer-funded basic research grants, especially the R01, the workhouse of the National Institutes of Health (NIH) extramural research program. However, not as many denizens of academia are familiar with SBIR and STTR grants, which can disburse monies commensurate with RO1 levels, i.e., greater than or equal to $100,000 per annum.

I attended a breakout session on innovations in funding therapeutics that was moderated by Doug Crawford, assistant director of QB3, and included Jesus Soriano, program director of SBIR/STTR at the National Science Foundation (NSF). Here are some of the highlights.

Out of $7 billion in NSF funding outlays, $150 million goes toward SBIRs. Sums up to $150,000 over 6 months are available in Phase 1, which is all about proof of concept. Phase 1 peer review assesses intellectual merit, originality, and societal and commercial impact. To use Clay Christensen’s ubiquitous turn of phrase, it must be disruptive. Phase 1 has a 10% funding rate, so be prepared for cycles of feedback and revision.

Phase 1 awardees are eligible for Phase 2, which provides up to $1,000,000 for two years to translate proof of concept into a prototype. At least for NSF vs. NIH-sponsored SBIRs, applicants proposing novel drug discovery platforms, as opposed to drug discovery products, are highly encouraged. To be eligible for SBIR funds, all you really have to is incorporate somewhere – and not necessarily in Delaware. In fact, you should probably incorporate in the state where you envision the startup taking root.

For a great synopsis of the summit, please see Storifies of #lsvs2012 conference tweets compiled by MedCity News editor Chris Seper here.

Building off that momentum, last week I attended a crowdfunding teach-in led by patent lawyer Kiran Lingam and Gust.com CEO David S. Rose in a classroom at General Assembly, which is an awesome multicity, educational space with a focus on entrepreneurship. I heard about this event from my cousin, who forwarded me the event Evite.

I attempted to livetweet the proceedings but was thwarted by a lousy AT&T 3G signal. (Granted the antenna on the iPhone 4G sucks, but still). So I resorted to taking notes on my iPhone.

Lingam explained that one of the major provisions of the recently enacted JOBS Act is a reversal of the 70-year ban on “general solicitation,” which was enshrined into law by the Securities Act of 1933, and given regulatory muscle by the Securities Exchange Act of 1934, which created the Securities and Exchange Commission (SEC). The rationale for the ban was that in the financial ruin of the Great Depression, Congress needed to protect unsophisticated investors from unscrupulous securities issuers who helped fuel speculative bubbles during the Roaring Twenties. Exemptions to the ban are given to accredited investors, basically rich people who are sophisticated enough to invest without supervision.

Lingam discussed the gory details of the law and to be honest much of it went over my head. (Say “broker dealer” ten times fast). The upshot is once the SEC issues formal rules later this year, a hypothetical entrepreneur will be able solicit investments from the general public using social and traditional media, but there will still be restrictions on the number of unaccredited investors that can participate. And there are other restrictions in place on the total amount of money that can be raised by crowdfunding, so it’s not a total opening of the floodgates.

David Rose echoed many of Lingam’s points. Most of what he said also went over my head but I grasped his vision of a crowdfunding utopia, which went something like this. An entrepreneur starts by issuing debt (as opposed to equity) to the crowd. Once the venture begins to bring in some revenues, a few percentage points are scraped off to pay back the crowd. Investors may get paid up to 3x or 5x the value of the loan. This modest return is in exchange for the slim chance that the venture becomes the next Google.

If all of this feels a bit ethereal, you’re not alone. Next time I’ll dig into some real-world examples of science crowdfunding.

What do you think? Please annotate in the comment thread below.

( reproduced from Figure 1 of Choy & Thomas, Molecular Cell 4: 143-152 )

To begin, Choy & Thomas assayed whether nose contraction requires serotonin using two mutant strains that have serotonin production or packaging defects. Turns out the noses of those mutants contracted just fine in response to a fluoxetine pulse. Specific phenotype in hand, they exploited it in a genetic screen for nose resistant to fluoxetine (nrf) mutants, i.e., mutants that are resistant to fluoxetine-induced nose contraction.

Next they performed two important controls. First, they showed that nrf mutants exhibit cross-resistance to two other antidepressants, including a brethren SSRI paroxetine/Paxil^® and the relatively serotonin-selective, older tricyclic clomipramine/Anafranil^®. Second, they showed that nrf mutants respond normally to non-antidepressant pharmacological agents that induce nose contraction by a distinct mechanism, namely inhibitors of excitatory ion channels.

Case closed! Well, not exactly. Serotonin-deficient mutants respond to fluoxetine, but confoundingly the putative worm homolog of the human serotonin reuptake transporter hSERT – the conventional protein target of SSRIs – is still present. As luck would have it, two years later Bob Horvitz’s lab independently replicated the general observation of serotonin-independent activity of fluoxetine in worms. In Ranganathan et al., they performed a clever fluorescence microscopy-based genetic screen for mutants whose neurons fail to uptake serotonin, thus genetically triangulating the location of the worm homolog of hSERT, mod-5. In a display of pharmacological specificity, they found that fluoxetine is still active in behavioral assays of strains carrying a null allele of mod-5.

Nothing much seemed to happen with the story for several years. Fast forward to 2006, when Thomas’ lab published a follow-up paper in which they genetically characterized one of the NRF genes in detail. They found by sequence alignment that it is homologous to mammalian lipid binding proteins, and proposed the following model: NRF proteins function as non-specific hydrophobic drug carriers. However, I think there’s an alternative and simpler interpretation.

Cationic amphipaths like fluoxetine don’t require active transport in order to accumulate in cells. Obviously, bioavailability is not uniform across cell types or environments, but thermodynamics demand that cationic amphipaths partition into lipid bilayers. Even a brief exposure to millimolar fluoxetine would result in copious membrane accumulation. NRF genes might regulate basal fluoxetine uptake capacity in a tissue-specific manner, such that in their absence specific cellular membranes accumulate less fluoxetine. A more baroque version of my model states that NRF genes are induced in response to fluoxetine, leading to a compensatory response that augments drug accumulation. The function of NRF genes remains a mystery but they contain an acyltransferase domain, which is consistent with the idea that they remodel membranes.

SSRIs are clearly active in both yeast and worms in the complete absence of a functional serotonin transporter. May we finally assume that psychoactive cationic amphipaths accumulate in all eukaryotic cellular membranes, especially in the tissues of organisms that have evolved serotonergic neurotransmission?

In my first first-author paper, I showed that drug responses – resistance or hypersensitivity – of naturally recombinant yeast strains are genetically complex, i.e., modified by two or more genes. In other words, “fungal pharmacogenetics” was born. (The original Chemistry & Biology cover image for this paper is reproduced above, and was designed by my good friend Ben de Bivort).

That initial publication was extended a year later by a genetic linkage study that I published in collaboration with Leonid Kruglyak’s lab. I identified nucleotide polymorphisms that broadly modify “drug overdose” in yeast cells, including substitution of a deeply conserved proline residue of an inorganic phosphate transporter protein PHO84 that confers resistance to polyphenolic uncouplers of oxidative phosphorylation. My collaboration with the Kruglyak lab continues to this day, and we are currently preparing a manuscript describing the application of the X-QTL method to identifying psychoactive drug overdose genetic determinants in yeast.

In parallel with those quantitative genetic analyses, I undertook a yeast cell growth-based high-throughput screen to identify novel chemical modifiers of the TOR kinase inhibitor rapamycin. In an email that I dashed off in 2004 to Prof. David Rubinsztein at the University of Cambridge, I proposed a collaboration to see if any of the Small-Molecule Enhancers of the cytostatic effects of Rapamycin, or SMERs, which I had identified in a primary screen in yeast were active in secondary assays of autophagy in metazoa. The rationale was simple: rapamycin induces autophagy, so a subset of SMERs may induce autophagy, too. As fate would have it, Prof. Rubinsztein responded favorably to my invitation to collaborate, and in 2007 we published a paper describing three SMERs, and analogs thereof, that were effective in a Drosophila melanogaster (fruit fly) model of neurodegeneration.

We still don’t know the molecular targets of those SMERs. But last year Paul Greengard’s lab at Rockefeller University independently showed that SMER28, which is commercially available, promotes clearance of amyloid-b in mammalian cells by a mechanism that depends on autophagy. When a Nobel laureate replicates your findings, it’s a good day in Academia…

(1) supplementary data, e.g., high-resolution electron microscopy images, associated with my lab’s published papers, primarily our recent PLoS ONE paper on membrane accumulation of the SSRI antidepressant sertraline/Zoloft^® in yeast cells;

(2) unpublished, icebreaker data intended to be a catalyst for online scientific discussions and new collaborations.

One of the cool features of Figshare is the altmetrics data, specifically the number of times a research object has been viewed or shared, as shown here in a screenshot of a representative single post page (above).

As of this writing, the 19 research objects I’ve uploaded have amassed 3,425 total views and 101 total shares. 70% of shares are tweets, with the remainder split between Facebook and G+. I was particularly curious about the relationship between views and shares, which I graphed here:

Two things struck me. First, nine objects received zero shares. None of them individually garnered more than 60 views, suggesting that sharing is required for large viewership, at least for my data. Second, there is a nice positive relationship between sharing and viewing.

A few qualitative observations might be useful for interpreting this static snapshot. Viewership wanes quickly, with most of the buzz lasting only a few days. However, research objects generously shared continue to amass views at a non-negligible rate compared to the zero-share class, though there is high variance in growth trajectories.

All in all, I’m impressed with the usability of Figshare, and I look forward to its wider adoption. It’s the principal reason why I leveraged the Figshare API on my Open Science platform. To boot, Mark Hahnel, the founder of Figshare, is helpful and timely with technical site assistance, and very generous with retweets!

As a Lewis-Sigler Fellow at Princeton, I’ve had the unique opportunity, in addition to overseeing an independent research program, to be an instructor for Princeton’s Integrated Sciences Curriculum, or ISC. While traditional postdoctoral fellows have limited if any teaching responsibilities, the ISC requires an unusually strong commitment not only from the dedicated students it attracts, but also from its instructors. Part of my learning process was to find my pedagogical voice. The teaching philosophy that informs the voice I’ve found developed and coalesced during my recent two-year stint as an instructor for ISC235/236, the yearlong sophomore introductory biology course. However, the core of my teaching philosophy dates all the way back to age 16, when I adopted a largely self-directed, rather than classroom-centered, approach to my own scientific education.

As an intern at a local Florida biotech venture, I first accessed scientific journals in the company library. Those high school years coincided with major technological leaps, such as the fledgling Internet enabling direct email contact with corresponding authors of research articles. One such early contact, with immunologist Ron Germain led me to three highly formative summer internships in his lab at the NIH in between college semesters at Columbia.

When I started teaching ISC235/236 two years ago, encouraged by ISC’s emphasis on innovation in science pedagogy, I drew upon my own early, full immersion in the primary scientific literature, and the vistas it had opened. So I embarked on a series of paper-based teaching experiments, designed to recapitulate, facilitate, and accelerate, in a classroom setting, my own autodidactic explorations. As a result, I am leaving my mark on the sophomore biology course in two innovative respects, and I openly confess to reconfiguring elements of a first-year graduate seminar for an undergraduate audience. First, I introduced my students to the custom of a post publication review, in which I present data from seminal papers in precept that goes along with weekly lecture material. I then moderate free-flowing, vigorous, largely student-sustained discussion of a paper’s methodology, data and models. Second, I initiated the practice of using papers as templates for problems that comprise weekly problem sets as well as midterm and final exams. My goals: to pave another avenue of exposure to the primary scientific literature, and to show students that general principles can be distilled from ostensibly disparate model systems or experimental approaches.

A paper-based, Socratic strategy has several practical advantages, coinciding with the emerging post-textbook, Udacity age. For example, a department over time will amass a distinct archive of pedagogically useful papers and problem sets that can ultimately be shared with other academic institutions by an open-source model. Most important, students are empowered to develop proactive critical thinking skills sooner than they otherwise might on their own, reading papers like mature scientists. Likewise, instructors are challenged to fulfill the promise of living curricula, which might otherwise lapse due to inertia or “limited bandwidth.” This modular approach can be infused into existing or new undergraduate courses, and already is a natural fit at the graduate level.

My journey began on Wednesday April 18^th 2012, when Chen et al., my lab’s paper on the bioaccumulation of Zoloft in yeast cells first appeared in the online Open Access journal PLoS ONE. I will present data on the first month of paper pageviews, and then discuss inferences regarding readership and reach.

The first plot (right) shows the growth rate of total pageviews, defined as HTML views + PDF downloads + XML downloads. Two obvious features jump out at me. First, site traffic came in spurts. My paper averaged 175 pageviews per day over the observation period, but this number was inflated by three amplifications; a better representation of site traffic is the median, 72 pageviews/day. The other neat result was that the amplifications were not made equal .

To explore those differences more closely, I re-plotted the data as daily pageviews (left), and color-coded Saturdays and Sundays in red to highlight the fact that site traffic almost always slowed to a trickle on weekends. An initial surge occurred in the first 48 hours, petering out by the first weekend. Unexpectedly, a second, equally sized surge began on Sunday May 6^th (Day 18), but it exhibited a slower decay than the first surge: 3 days vs. 2 days. A third, smaller surge began on Thursday May 17^th (Day 29) and faded by the fifth weekend.

The first and third surges, although differing in amplitude, exhibited comparable decay kinetics. “Newness” and press release syndication drove the first surge as adduced by an initial EurekAlert, and several popular science news aggregators that propagated it, e.g., PsychCentral. The third surge was caused by a review of my paper by Derek Lowe on his blog “In the Pipeline,” which is appointment reading for a diverse group of scientists from academia and the pharmaceutical industry. To wit: several colleagues independently congratulated me by email within 24 hrs of the blog post, and attested to Lowe’s wide sphere of influence.

The slower decaying second surge is more complex. I think that multiple independent and potentially self-reinforcing catalysts sustained site traffic buzz for an extra 24 hours. The spark appears to have been my tweeting a link to Chen et al at the hashtag #APAAM12 on the morning of May 6^th. Attendees and online followers of the 2012 annual meeting of the American Psychiatric Association in Philadelphia would have seen that tweet. A Phoenix-based mental health advocate and blogger in fact did, and she retweeted my tweet, #APAAM12 hashtag and all. She also alerted me to a blurb about my paper on the psychiatry news aggregator “Mad in America,” and she claimed that it was the source of the buzz. But the timing of that blurb preceded the second surge by a few days, which led me to investigate, i.e., to Google search, other sources of amplification.

Turns out there were several good candidates. John Timmer, whom I know from Science Online NYC (@S_O_NYC on Twitter), writes for Ars Technica, and he penned another blurb about my paper but it also appeared days before the second surge. Still not entirely satisfied by the lack of direct cause and effect, I searched more and uncovered yet another blurb about my paper, this time by a Phoenix-based lawyer. His firm was apparently using Chen et al (without my knowledge, btw) to help solicit plaintiffs to a class action lawsuit claiming that Zoloft caused birth defects, which is apparently sweeping the nation. All in all, it appeared to be a perfect storm.

Some of you may be miffed by my apparent conflation of site traffic and readership. Without more sophisticated analytics, I confess that it’s difficult to gauge the background of readers (scientists vs. non-scientists), or how much of the paper they’re actually reading (abstract vs. full text). However, the ratio of HTML views to PDF downloads (HTML/PDF) may be informative here. After the initial surge, HTML/PDF was 1 in 20 and remained there until the second surge, after which it fell to 1 in 30. If we assume that PDF downloads are a proxy for “expert” readership, then the second surge diluted quality readership. Conversely, the third surge lifted the ratio to 1 in 15, with as many as 20% of readers, many of whom were presumably academics, on Day 29 choosing to download a PDF version of the paper.

In the next post in the Open Science series, I will present results of a sociological experiment in post publication review, and discuss my recipe for vibrant online commentary.

The drug resistance approach works spectacularly well with pharmacologically simple compounds, e.g., high-affinity binders to a single protein target.  However, the number of mutational paths to resistance appears to scale with the complexity of molecular interactions with drug targets at or above the lethal drug dose. Two specific cases studies from the annals of yeast genetics provide rich justification for a yeast-based “psych drug overdose” resistance approach, which my lab debuted in a 2010 Genetics paper, hereafter Rainey et al. Both examples involve the work of unsung heroine and second-wave yeast geneticist named Norma Neff.

One of the major results of Rainey et al was our isolation of sertraline-resistant mutations in three subunits of the vacuolar ATPase complex (V-ATPase): VMA1, VMA3 and VMA9. But chemical screens of yeast deletion collections in the early 2000s showed that V-ATPase mutants were non-specifically hypersensitive to drugs, the opposite of what we found. To make sense of this confusion, we need to go back to 1988. Norma Neff’s group at Sloan-Kettering was interested in the mode of action of trifluoperazine,  a psychoactive cationic amphipath, and so chemical relative of sertraline. Specifically, trifluoperazine is a phenothiazine antipsychotic that descends from chlorpromazine (Thorazine^®), the first popularized antipsychotic drug. Neff’s group published a paper in which they cloned a novel trifluoperazine-resistance gene, TFP1. Based on sequence homology to a subunit of the energy-producing  mitochondrial F1/F0 ATPase, Neff’s lab correctly concluded that TFP1 was (a component of) a proton transporter, as shown in a reproduction of Figure 7 from the paper: (right). Therefore, sertraline resistance and trifluoperazine resistance share a common genetic modifier, the V-ATPase.

The only criticism I have with this otherwise well-executed paper is the interpretation of why mutations in TFP1/VMA1 resulted in trifluoperazine resistance. The paper concludes:

“The regulation of the activity of such ion pumps may contribute to intracellular traffic and organelle function, because local metabolic activity and protein targeting may be influenced by changes in the different electrochemical potentials across the different membranes which enclose cellular compartments.”

This gets it half right, I think. Neff clearly understood that acidification was the salient physiological process underlying trifluoperazine resistance. But the most parsimonious explanation is lysosomotropism, as I argued in Rainey et al. For more recent literature on lysosomotropism in a mammalian cell model, see the excellent work of Jeff Krise’s lab at the University of Kansas.

The second, totally unexpected finding of Rainey et al was the observation that the reduced fitness caused by a point mutation in the heavy chain of clathrin (CHC1^E292K) is suppressed by low doses of sertraline. In other words, the same drug used to select for drug-resistant mutants at high doses enhances the growth of one particular sick drug-resistant mutant at low doses, as originally shown in our paper: (right)

Turns out Neff and associates, working in the lab of genetics impresario David Botstein (now at Princeton, then at MIT), published in 1985 a lovely Genetics paper describing almost exactly the same phenomenon, except in their case they selected for mutants resistant to the microtubule depolymerizer benomyl. Benomyl insinuates itself into tubulin and prevents yeast cells from segregating their chromosomes, a lethal blow at high doses. One benomyl-resistant mutant in particular, tub2-150, exhibited “benomyl dependence,” that is this benomyl-resistant mutant actually grew better in the presence of benomyl, especially in the cold. They rationalized benomyl dependence as follows:

“The benomyl-dependent mutations described here fit this model well. They could have a defect that renders their microtubules more stable than wild type, as has been demonstrated for a tubulin mutation in A. nidulans. Such a defect could be compensated for by a chemical destabilizing agent, such as benomyl, or by a physical one, such as low temperatures.”

If you substitute clathrin for microtubules and sertraline for benomyl, one can begin to understand why sub-lethal doses of sertraline rescue the fitness defect caused by altered clathrin function. But how exactly sertraline accumulation in vesiculogenic membranes affects clathrin function in yeast cells is an ongoing focus of research efforts in the lab.

“Yeast don’t have a brain or a serotonin transporter or tissue of any type! Why are you giving antidepressants to yeast cells?”

Their skepticism was not without reason. At first blush, a drug that was pharmaceutically honed to be more selective than the overdose-prone, preceding generation of antidepressants known as tricyclics had no business being biologically active in the simple yeast cell. So imagine the look of consternation when I proceeded to explain that the inaugural experiment of my lab was a selection for mutant yeast that are resistant to sertraline-induced “overdose.”

I wasn’t the first to observe the unexpected antifungal properties of SSRIs. That honor belongs to a group of Austrian scientists who published back-to-back papers on the subject in 2001. The first paper recounted a serendipitous clinical observation: three women suffering from premenstrual syndrome (PMS) and a Candida yeast infection were given Zoloft. During the course of Zoloft treatment the yeast infection cleared, but the Candida returned after discontinuation of the drug. The second paper went on to show that another yeast species (Aspergillus) was sensitive not only to sertraline but to other SSRIs, too. However, the authors cautioned that the concentration of SSRIs required to kill Candida cells in the lab were much higher than their serum concentrations in patients taking the drug. They went on to speculate that yeast cells might be hypersensitive to Zoloft inside the human body, and they implied that the concentration of Zoloft varies across tissues. The straightforward therapeutic take-home message was: let’s repurpose antidepressants into antifungal agents.

Two years later in 2003, scientists at Pfizer, the pharmaceutical company that originally developed Zoloft, responded to the Austrian studies. The Pfizer group first dutifully observed that yeast cells lack a serotonin system entirely, including therefore the conventional drug target of all SSRIs – the human serotonin transporter, or hSERT. Next, they made several key observations about the underlying structure-activity relationships. First, they found no correlation between hSERT binding and antifungal activity by examining a panel of Zoloft analogs, including diastereomers of Zoloft. Second, they found a strong, Overton-Meyeresque correlation between hydrophobicity and antifungal potency. They concluded the paper by stating:

“The present study indicates that the antifungal effect of SSRIs is mediated through a non-specific mechanism related to the lipophilicity of the agents, and is associated with significant toxic effects on human cells at concentrations required for activity against fungal pathogens.”

In the final analysis I believe that the Pfizer group’s conclusions were shortsighted, because they didn’t actually provide evidence to support the claim that the antifungal effect is “non-specific.” They reached that conclusion because the antifungal activity of SSRIs is not stereo-specific, merely suggesting that the putative antifungal drug target (a) is not a protein and (b) is very hydrophobic. Sounds a lot like cellular membrane to me…

Skip ahead to 2010 when my lab published in the journal Genetics the first installment of the Zoloft saga, describing the isolation and genetic characterization of Zoloft-resistant yeast mutants. For more on that study, please check out this more technical companion Featured post, which begins to chronicle the history of drug resistance screening in yeast.