We Are All Obsolete

I've tried to keep this blog focused on academic science, but I've got an idea pinging around my head.  I know it's not original (for one, my cousin mentioned it in a car ten years ago), but here it is: we are all obsolete.  Every "successful" person today - whether they are a musician, scientist, programmer, or athlete - is going to be surpassed in our lifetime.

Historically, this is obvious.  In athletics, records constantly fall.  The average IQ scores which constantly go up. Groundbreaking papers are trivially reproducible.

Like dying, obsolescence happens slowly, so we can ignore it in our daily life. But it catches up to all of us.

There are a lot of factors pushing us towards obsolescence.  We all lose intelligence as we age.  There is the glacial force of natural selection (if those still work in an age of medicine and a social safety net).  Today I'm going to focus on two factors that feed into this: the skyrocketing of the effective population; and the punctuated improvement of education.

Effective Population


I've had a fortunate education.  I started at a Montessori elementary school, where play time motivated me to work hard.  I went to a private high school that let me go to college a year early.  Made up my own major, computational neuroscience, at Case Western.  At Duke I was the first graduate student in a lab that eventually became a Howard Hughes lab.  And now I'm in Geneva with free reign to do taste research in an effectively olfactory lab.


I got lucky.  Lucky I was effectively an only child; that my parents were educated and valued education; that I generally had good teachers along the way.  I'd guess only one in twenty Americans were as lucky as I was, but that could just be hubris.


When you think about the demographics of the world, it's easy to say that the US has 300 million people with the same opportunities, but that's not really the case. Some people estimate 25% of US children are in poverty.  If you take inverse of that number, only about 225 million people in the US have ample opportunities; this is the US's effective population.


I got the idea of effective population from Information Processing.  The idea is most applicable to a country like China, which has 1.3 billion people, but only about 300 million of them are able to compete in the global marketplace.  That is, only 300 million have the nutrition, education, and financial stability to go to college, get educated, and try to create something in this world.  The rest suffer from malnutrition, families that need the income, or simply from a lack of teachers necessary to educate a billion people.

If China's effective population is only 300 million, what about the rest of the world?  I already estimated that the US's effective population is around 225 million. Rather than type this out, I'll just estimate the effective population of the world (gotta love a table that intimates billions of people don't exist; all of these numbers are pulled out of my ass.  For example, how do I estimate Europe, which combines the well-developed West, and the still developing East?):

Country/Continent	Population	Effective percentage	Effective Population
USA	300 million	75%	225 million
China	1.3 billion	25%	300 million
India	1.2 billion	10%	120 million
Latin America	600 million	25%	150 million
Africa	1.4 billion	10%	140 million
Europe	700 million	60%	420 million
Asia ex-China/India	1.5 billion	10%	150 million

In total, about 1.2 billion, give our take a few hundred million.  And this number is always going up.

A few pundits have made waves recently pointing this out (Hot, Flat, and Crowded; Post-American World).  As a scientist this is both scary and exhilirating: the competition is going to get MUCH tougher; and hopefully the achievements will as well.  But unfortunately, it means my effective place in the world will go down.

Educational Improvement


Some systems are so vast and hard to measure accurately that it's easy for anyone to have an opinion on how they should be run: health care; taxation; and for this post, education.  Everyone has an idea how the education system should be run.  School system funding should be ample, teachers should be held accountable, parents should read to their children, and the students themselves need to be measured (but we shouldn't teach to the test).  We should teach people how to work together in groups, but not ignore basic skills.  The subjects should include the three R's, but also newer things like psychology and computer programming.  In the end, most people imprint on their own education, and have ideas about what did and did not work in theirs.

I have no idea how to improve education.  But I do know the way we educate people now is vestigial and will be improved upon.

Right now, most education treats students like cogs in a factory (I generally sneer at those RSA videos as middlebrow, but holy cow that drawing struck a chord).  We group people in classes because that's all we could do a century ago, if we wanted to educate as many people as possible.  We continue doing so due to the inertia of institutions.  And educational opportunities are almost non-existent for lower-class people, both in the US and around the world.

As I said, I don't know how to improve education, but there are lots of people trying different things.  For example a multitude of individualized education programs are sprouting up (I am biased in favour of this, having started in Montessori school).  There's the School of One in Brooklyn (more press here).  The Gates Foundation is trying a lot of different models, supporting guys like Salman Khan.  If one of these works, we can copy the model and disseminate it.

At the top of the post, I mentioned I had a good education, and that only one in twenty Americans might have had access to something like it.  But the world is getting wealthier all the time, increasing the number of people who will get educated.  And the education they're going to get continues to improve.  It's easy to imagine thirty years from now, when I'm sixty, there will be a whole new generation of scientists, from around the world, that have a better education than me.  I'm going to have to pit my ideas against theirs, hoping my experience can compensate.  And eventually, I'm not going to even going to be proven wrong, I'm not even going to be able to compete.

A Walk Along the Paper Trail: Katzogenesis

While the last few walks have covered taste receptors, I'm more interested in the central representation of taste.  When you taste something the information is relayed from the taste receptors by three facial nerves to the brainstem (NST), then to the thalamus (VPMpc), and from there to gustatory cortex (GC).  The NST also projects to the amygdala and lateral hypothalamus, sending reward and feeding intake information.

There aren't a lot of labs that study taste coding in GC, but one of the best labs is  Donald Katz's at Brandeis.  He's done some interesting work on ensemble representations of taste, but today I will cover his neuroscience origin story.  As a post-doc he worked with Sid Simon at Duke, and put out two nice papers about taste coding in GC, which I will review here.

GC neurons respond in three phases


In the first paper, they simply recorded from GC (and oral SC) neurons in awake rats while they licked four basic tastants: sweet, sour, salty, and bitter.  When they calculated the taste responses over the 2.5s following a lick, 14% of the neurons responded to at least one taste, which was in line with previous reports.  However, if they instead binned the responses into 500ms chunks, they found that the number of responsive neurons increased to 33%.  The dynamics of the responses were different for each neuron, with some changing their "preferred tastants" in each time bin.

The responses of two neurons in 500ms bins. Neuron 1 has a similar response, scaled, to each tastants. Neuron 2 has different dynamics for each tastant.
From Katz et al 2001

For example in the figure above, neurons 1 and 2 have different preferred tastants in different bins (counting inhibition as "preferred" since it's still information).  And when they further smoothed their analysis by using a moving average of the response, 40% of the neurons responded to tastants.

Given the complex dynamics of the response, they next asked if different information was represented at different times.  To do this they identified "modulations" in the continuous response, either inhibitory or excitatory.  Then they made a histogram of the modulation onset times, and found they were bimodally distributed (panel A).

A. GC neurons modulations are bimodally distributed <0.5s and 1-1.5s.  B.  A closer analysis of the early responses shows there is another division of responses around ~250ms.
From Katz et al 2001.

Some modulations started early, with an onset time of <0.5s, while others started later, >1s.  They then looked even closer at the early onset times, and plotted just those within the first 0.8s (right panel above).  Here they found that a set of onset times seemed to cluster <0.25s, while the rest were distributed >0.25s.  Thus there were three modulation windows: 0-250ms, 0.25-1s, and >1s.

They hypothesized that the early and late onset times were due to orofacial movements like licking, or facial gestures made in response to palatable (or not) tastes.  If this were true, since licking occurs at 5-10Hz, one would expect the early and late responses to have information at 5-10Hz range. To see this, they performed FFT on the responses, and looked at the power spectra of the early, middle, and late responses.  And indeed the early and late responses had more power in the 5-10Hz range.  Given this they concluded that GC neurons encoded different information at different times of the response.

GC neurons encode three types of responses over different time scales. Early, somatosensory licking info, "middle" taste info, and finally palatability.
From Katz et al 2001.

This is probably hindsight bias, but I can't believe it took so long for GC scientists to look at the temporal dynamics of taste coding.  In the discussion section, they cite research in the olfactory bulb, motor cortex, and visual cortex that all investigated temporal dynamics years earlier (frankly, mid-90s seems late).  And using temporal dynamics completely changes the picture of GC: the number of taste-responsive neurons jumped from 10 to 40%!  GC went from being a tangentially taste-related cortex to being obviously taste specific.  All the papers since then have confirmed that indeed, 40% of neurons are taste responsive.

CC: GC neurons


To follow this work on single neuron representations, Katz Simon, and Nicolelis next turned to the population response.  This was done by simply calculating the cross-correlation (CC) between the firing rates of pairs of neurons (the paper includes more sophisticated analyses like linear discriminant analysis, but the CC result is cleanest).  They recorded 237 pairs of neurons, sometimes in both hemispheres, in 12 rats.  Of the 237 pairs, 85 had changes in CC due to specific tastants.

Three example neuron pairs. On left are CCs for the pairs (how often neuron 2 fired following neuron 1); on right are the individual neurons' responses to tastants. A. This pair of neurons had different PSTHs, but an increase in CC. B. An example of inhibitory CC. C. An example of neurons that had sharp increase in CC within a short window.
From Katz et al 2002.

For example, panel A shows a neuron pair that had increased correlation during NaCl licking (mauve).  While this pair had a change in CC, you can see that neuron 2 was not taste-selective.  Of the 85 pairs of interactions identified, 67%of them included non-responsive or non-taste selective neurons.

Panel B shows a loss of correlation between the two neurons, perhaps due to inhibition. Panel B also shows a pair that was responsive to two tastants; of the 87 neuron pairs, 50 showed significant CC for more than one tastant.  Panel C shows a short timescale interaction of a few ms, overlaid on top of a broader CC of hundreds of milliseconds.  These short timescale interactions occurred 17% of the time.  Finally, they showed that neurons in different hemispheres could have CC.

In the discussion they consider a few different sources for the CC, including common sources, coupled latency, or orofacial behaviours, but discard them due to the analyses I did not present. They mention that you could get changes in CC between neurons with different PSTHs (e.g. panel A), which shows this is due to CCs in single trials.  Overall, they concluded that these CCs showed there was a population representation of GC information.

I'm curious about the identity of these pairs of neurons.  The recordings were performed blind to the cortical layer being recorded, and whether the neurons were pyramidal or interneurons.  This information would be hugely useful.  For example, if neuron B lagged neuron A by 50ms, it would mean entirely different things if neuron A was in layer 4 and neuron B was in layer 2/3, or vice versa.  In the latter case (neuron A in layer 4), you could simply chalk the delay up to normal circuit function; if neuron A was in layer 2/3 though, this would imply some more complicated feedback processing.  Similarly, when you see a reduction in CC, one might guess that the pair includes an interneuron. To get at this information, we're going to need more sophisticated tools to record from identified cortical neurons.

In any case, those are the two papers from Donald Katz's postdoc in the Simon lab at Duke.  He's revisited the theme of population coding many times (I'd recommend Jones et al 2007 for a Hidden Markov Model version of the story).  Jusqu'à la prochaine fois.

Katz DB, Simon SA, & Nicolelis MA (2001). Dynamic and multimodal responses of gustatory cortical neurons in awake rats. The Journal of neuroscience : the official journal of the Society for Neuroscience, 21 (12), 4478-89 PMID: 11404435

Katz DB, Simon SA, & Nicolelis MA (2002). Taste-specific neuronal ensembles in the gustatory cortex of awake rats. The Journal of neuroscience : the official journal of the Society for Neuroscience, 22 (5), 1850-7 PMID: 11880514

The Mechanisms Underlying Grantsmanship are Not Fully Understood

I was editing one of the lab's papers today, and came across the classic grant/paper sentence, "The mechanisms underlying ... are not fully understood."  Do you ever see that sentence outside of science? So I went to Google Scholar and searched for "mechanisms underlying" and "not fully understood"  to find the first usage of it.

If you search for each term individually, you will find hundreds of references dating to the nineteenth century.  They were both common scientific phrases, but it took time for them to be combined.

If you search for the two phrases combined, the earliest link is to a book review from 1920, but skimming the document, I could not find either phrase.

The next reference comes from a 1950 paper, "THE SIGNIFICANCE OF THE "ONE-MINUTE" (PROMPT DIRECT REACTING) BILIRUBIN IN SERUM'," although they use each fragment in different sentences:

"The mechanisms underlying the renal excretion of bilirubin are still obscure." (I like that twist, I'm going to steal it.)

"The factors governing the speed of diazotization of bilirubin in serum are not fully understood."

It was not until 1962 that the full power of the phrase was unlocked almost simultaneously by two papers, "Physiology of acclimation to low temperature in poikilotherms:"

The degree of compensa- tion is different in different groups of animals (2, 3) and the mechanisms underlying this compensation are not fully understood.

 and "The inflammatory response to a foreign body within transplantable tumors."

This response seemingly lies in the stroma and, although mechanisms underlying the inflammatory reaction in normal tissues are not fully understood...

The science world would never be the same.

Do Whatcha Wanna*

While some PIs eventually learn to take pride in their "grantsmanship," I doubt anyone is happy with the grant system.  Nominal scientists spend their time trying to raise money rather than doing actual science.  We award grants based on people's paper trail, and then go tell them to teach, train, proselytize, and, oh yeah, publish.

I don't have a well thought out solution to the problem, but I do have a half-baked one: treat scientists like start-up companies.  My idea comes from two strains.

Cause it makes you smile if it sounds dope
When I read The Double Helix, the biggest surprise to me was that Watson and Crick discovered the structure of DNA as a side project.  They both were working on other projects - I can't remember what, but I think it had to do with invertebrates - and would sneak off together to try and piece together the crystallography data.  And Watson kept having to appease his advisor that his main project was indeed moving along, and apply for fellowships.

The lessons I took from this (and this is simplistic) are that people work best on things their interested in, and trying to make them work on a specific project is counterproductive. This may be my experience, but I know many people who toil away on mediocre projects when they yearn to do something else.** Yet, when we apply for grants, we make people write up specific projects that by definition may not yield interesting results. So what do we do if we stop writing grant proposals?

Scientists and startups
One of my favourite essayists is Paul Graham.  He's an angel investor (venture capitalist) who biannually runs a startup bootcamp to identify and train tech entrepeneurs.  When he decides whether to invest in a company, he almost ignores their business plan, because nascent companies constantly change plans.  What he focuses on are the founders, and he looks for specific traits: determination, flexibility, imagination, naughtiness, and friendship. Founding companies is extremely demanding, with a high failure rate.

Entrepreneurs and scientists share a lot of similarities.  They're both trying to do something new, which means exploring a lot of idea space, and modifying the plan as results come in.  They both have to overcome failure, whether it's experiments not working, or users not signing up. The rewards are asymmetric, with the best projects doing orders of magnitude better than the average. The best scientists and founders are not necessarily those that are the smartest, but the best hackers and hustlers.  And both groups waste a lot of time trying to raise money.

Ten years ago, venture capitalists evaluated startup companies the same way we evaluate grants today: they'd ask for a business plan, and then fund based on that.  But they've realized another model has better yield: ignore the business plan, and fund the founder.  My proposal is that scientists do the same.

Rather than have people spend weeks writing a fellowship, filled with scientific justification and wedged-in hyphotheses, let's run a scientist boot camp.  Take a month, send people off to Woods Hole (or wherever), and have them slap together a project.  See who stays up late.  See who tries something spectacular, fails, then whittles it down to something manageable.  See who hacks together a solution to a problem.  And fund them, for whatever the want to do, proposal unseen.

In the end, I don't think the cost is that high: some flights, a month's pay for the students, and a couple supervisors.  You'd save a lot of grant reviewers' time.  You'd build camaraderie between the students that may last as they venture back to their home institutes.  And you might end up funding successful scientists rather than people with good credentials.

(Having slept on this, I am downplaying the logistics  here.  For example, working with mouse models would be difficult in a one month course. But there are pretty common, useful mouse lines like Thy1-GFP/ChR2, and you could even try a BYOM system if the mouse quarantines were modified for the unique situation.)

*In honor of Treme.

** I realize even mediocre projects need resolution.  Sometimes its better, though, to just pull the plug.

A Walk Along the Paper Trail: A Cannabinoid Trail Mix

It's grant writing time here at the Paper Trail, which means reading lots of papers to cite in the background section of the grant.  I'm going to cover my favourite paper that I've discovered, which shows that endocannabinoids can directly modulate taste receptors.

More than meets the tongue


Flavour is a tricky perception.  It's obviously dominated by how things taste, but also influenced by olfaction, and internal states like hunger.  Recordings from rat gustatory cortex show that other sensory modalities are represented as well, like sensorimotor information from the tongue, or temperature.

While some of these modalities are directly encoded in cortex, others are represented indirectly, through hormones and neuromodulators.  The most famous of these is leptin.  Leptin is released by fat cells (adipose tissue), and is bound by leptin receptors in the hypothalamus and sweet taste bud cells (TBCs).  Leptin is an anorexigenic mediator, which means it suppresses appetite.  What's really cool is that leptin doesn't just act centrally:  if you record from TBCs in mice, you'll find that leptin decreases the firing of sweet TBCs.

The Munchies


In contrast to leptin, endocannabinoids are orexigenic mediators (appetite stimulants) that were known to work through CB₁ receptors in the hypothalamus and forebrain.  In the paper I'm covering today, Yoshida and colleagues showed that endocannabinoids (henceforward ECBs) can act orexigenically directly on sweet receptors themselves.


They started by recording from the taste nerve innervating the anterior tongue.  In wild type mice, the taste nerve responded to a variety of tastants, including NaCl, sucrose, quinine, etc (panel A, below).  To see the effects of ECBs, they injected the endocannabinoid 2-AG i.p., and again recorded from the taste nerve and found that 2-AG increased the response to sweet tastants  (panels A, B). They also tested the dose-dependence of 2-AG, and found it saturated at approximately 1mg/kg body weight.

Endoannabinoids increase sweet response; CB₁ -/- mice have no increase.
From Yoshida et. al. 2010.

Next they repeated the experiment in CB₁ knockout mice, and found that the knockout mice had normal responses to all tastants.  However, when they injected the ECBs, there was no increase in the sweet response. This shows that ECBs can enhance the sweet response, and that the CB₁ receptors are essential for that modulation.

To see the behavioural effects of ECBs, they measured how often the mice licked a liquid source.  To make the task more interesting, they mixed quinine (a bitter tastant) with sucrose at different concentrations.  At all the concentrations tested, the lick rate increased in mice injected with the ECBs.  CB₁ knockout mice, however, had no difference in lick rate.

Next, to verify that ECBs work directly on TBCs, they isolated TBCs and recorded from them directly.  They used a transgenic line that expressed GFP in sweet cells, under the promoter for T1r3 (expressed in umami cells as well). Using a glass microelectrode, they recorded extracellular action potentials from the isolated TBCs in response to sweet tastants (see below).  Then they bath applied 2-AG and found that the firing rate increased in response to sweet tastants.

2-AG enhances TBC response to sweet tastants.
From Yoshida et. al. 2010.

They tested the response over a variety of concentrations, and found the EC50 for 2-AG was 0.1 ug/mL.  They also verified the ECBs worked through the CB₁ receptor by applying antagonists against CB₁ and CB₂.  Only the CB₁ antagonists were able to block the ECB enhancement. In the final figure of the paper, they performed RT-PCR and immunostaining to verify CB₁ was present in sweet TBCs.

It's amazing to me how often the brain seems to modulate in depth.  There are endocannabinoid receptors on the tongue, and in the hypothalamus and forebrain.  And it occurs across modulators as well, as leptin is expressed in all these places.

While I jokingly titled the review "the munchies," the body expresses endogenous endocannabinoids, and these levels inversely correlate with leptin levels in the blood. And while the effects of endocannabinoids are obvious on the tongue, I don't think it is quite as clear in the brain.  It would be interesting to record from gustatory cortex while animals were under the influence of endocannabinoids to see how the representation changes.  You could sell it to the NIH under the drug addiction program.

Yoshida R, Ohkuri T, Jyotaki M, Yasuo T, Horio N, Yasumatsu K, Sanematsu K, Shigemura N, Yamamoto T, Margolskee RF, & Ninomiya Y (2010). Endocannabinoids selectively enhance sweet taste. Proceedings of the National Academy of Sciences of the United States of America, 107 (2), 935-9 PMID: 20080779