Olfactory bulb activity in the absence of direct olfactory input

OR: The dumbest experiment I've ever done that kindaworked

Over the last few months, I have written about my ongoing research. In brief, the odor code in the olfactory bulb is known to be highly dynamic over the course of a single breathing cycle. I have found that there are inter-breath dynamics as well: that is, the odor code during the first sniff of an odor differs from subsequent sniffs. Furthermore, I found that odor-specific information persisted in the olfactory bulb after the odor was no longer present.

I thought this was a decent finding, and wanted to push it out, but mon chef wanted to beef the paper up by identifying the (or a) mechanism. The two most obvious mechanisms for inter-breath and post-odor dynamics are dynamic input from olfactory receptor neurons, and feedback from "higher" olfactory areas. Olfactory receptor neuron dynamics have been reported in flies before, but not well observed in mammals. However, we are not well equipped to record from ORNs at the moment. Feedback, on the other hand, is more virgin territory, and easier (for us) to manipulate.

I'm working on a few experiments regarding feedback. The first one involves dissociating the feedback and olfactory input, by removing the direct olfactory input; in other words, I blocked the nose, leaving only feedback..

Responses of individual neurons

I recorded from the left olfactory bulb of anesthetized mice using tetrode arrays. Prior to each recording, I blocked the left nose using a liquid plastic, liberally lathering the ipsilateral nostril. I am confident I got near complate blockage, for: 1.) when I removed the plastic, it outlined the inside of the nasal cavity; 2.) the nasal pressure in the contralateral nostril doubled after blockade; and 3.) there was almost no odor information in the ipsilateral olfactory bulb. With the nostril  blocked, I recorded the mitral cells' response to a set of odors, then removed the block, and recorded from the same neurons with the nostril open.

Blocking the nostril significantly reduced the activity in the olfactory bulb. With an open nostril, most (putative) mitral cells fire at 10-20Hz. With the nostril blocked, the firing rate was ~4-5Hz. There were some cells which were almost silent in the blocked condition, but which fired vigorously once the nostril was opened.

Raster plot of one cells' response to Amyl Acetate. Spikes shown as black lines, breaths as blue lines. During the first twelve trials, the nostril was blocked, and the cell fired infrequently. Once the nostril was open, the cell fired vigorously, in phase with the breathing cycle.

I recorded from 91 cells in 3 animals at ten recording sites to generate 400 cell-odor pairs. Of the 91 cells, only one cell displayed phasic firing during the breathing cycle in the absence of odor. Of the 400 cell-odor pairs, only 5 responded to odor.

One of five cell-odor pairs to respond to odor. While the nostril is blocked (first twelve trials), the neuron has a broad, phasic response. With the nostril open, the neuron refined its timing. Of note, the neuron responded to 1/3 odors when blocked, but 3/3 odors when open.

How does one interpret these sparse responses? One might claim that they are due to a leaky blockade, and that this subset of cells are highly sensitive. I would argue against this, because more neurons recorded in parallel had no response. Furthermore, I think I did a competent job. Ultimately, to show that this is not a leaky blockage, one would need to record from these neurons while presenting very low concentrations of odorant, to compare to the blocked condition. This would be a lot of work for relatively little gain.

The take-home message here is simply that mitral cells in the olfactory bulb receive a vast majority of their excitatory input from ORNs. Whatever feedback may be present must modulate the signal, rather than drive it.

LFP changes in the blocked nostril

While the feedback in the blocked nostril condition is not strong enough to drive spiking, that does not mean there's no feedback. To examine sub-threshold input, we looked at the LFP on electrodes which contained spikes, and saw that the odor presentation was causing oscillations in the LFP.

LFP (top) and nostril air pressure (bottom) during presentation of odor (black bar from 0-4s). Before the odor, the LFP has little power except at ~1Hz. During and after the odor, the LFP is moderately active.

To determine whether this was real, I calculated the LFP power using FFTs, looking at the periods 3-4 seconds before, during, and after the odor. For ~7/10 recording sites, the LFP power was higher during the odor and post-odor period, for frequencies in a range from 5-30Hz (see below, left; eye-test). For comparison, I also calculated the LFP power with the nostril open. With the nostril open, the baseline power is higher, reflecting the increased spiking. During the odor, there is an increase in power at all frequencies, and especially in the gamma range.

Log power for one recording site with the nostril blocked (left) and open (right). In the blocked condition, during odor presentation, the LFP power increased in the 5-35Hz range, and remained elevated during the post-odor period. For the open nostril, the baseline power is higher. During odor presentation, there is a large gamma frequency band.

To test whether this was significant, I pooled together the results of the ten recording sites, and performed an ANOVA by calculating the power for each experiment at 5,10,...30Hz. There was a statistically significant effect of the odor presentation on the LFP power (p<0.05).

Average LFP power for blocked nostril (left) and open nostril (right).

Discussion

The main conclusion from these experiments is that during odor presentation, in the absence of direct olfactory input, the OB still receives neural input, presumably from higher processing areas.

These are the first LFP measurements I have done, so I am not expert at interpreting them. However, my understanding is that higher frequency (30+ Hz) LFP oscillations reflect local processing, while low frequency oscillations reflect longer distance connections. You can see that in the final figure abovewhen the nostril is open, the OB receives direct olfactory input, is quite active, and there is power in the gamma frequency; with the nostril blocked, the gamma frequency disappears. The fact that the LFP power in the blocked nostril condition is at low frequencies is in accordance with the idea that it is feedback from higher (more distant) areas.

I am not the first person to record from olfactory neurons while blocking the nostril. The Mori lab recorded from the olfactory bulb and anterior olfactory nucleus while blocking either the ipsilateral or contralateral nostril. They found that all mitral/tufted cells responded only to ipsilateral odor inputs, while AON neurons could respond to both contralateral and ipsilateral inputs. Of AON recordings, ~40% of cells were ipsilateral-exclusive, while 60% responded to bilateral inputs. Interestingly, the AON cells had highly overlapping odor "receptive fields" for ipsilateral and contralateral odorants.

Mitral cells respond exclusively to ipsilateral odorants, while AON neurons can respond bilaterally. A. Diagram of recording apparatus. The odors were applied focally via an external septum, and chitin membrane. B. No mitral/tufted cells responded to odorants from the contralateral olfactory bulb. C. Example AON neuron responds to both ipsilaterl and contralateral odor presentation.

They also found that ipsilateral blockade removed the phasicness of the baseline response, almost silencing the AON neurons. Interestingly, after a few hundred seconds, this phasicness was restored, presumably from contralateral inputs.

My experiment confirms their basic result in the olfactory bulb: almost no mitral/tufted cells spike in response to contralateral odor input. However, I do find that there is some sort of sub-threshold, contralateral input in the olfactory bulb, as evidenced by the LFP. After reading their paper, I wish I had also recorded from the olfactory bulb while blocking the contrlateral nostril, to see if that modified the odor receptive field.

PI Notes - Hiring, Virtual Seminars, and Lab Meetings

I'm at home for Easter, which means copious free time, and a limited science network. So here's a soft-science post.

In sports, as in science, you progress from performance to management. Smart, forward-looking players often write journals during their career, noting what their coaches do that's effective (or counterproductive), and gradually develop a model for how they will run things. Since grad school, I've kept a notebook (rather, a Google Doc) with observations of how labs function. Whenever I meet a successful, thoughtful PI, I ask them how they run their lab (two of the most thoughtful are Hillel Chiel and Rick Huganir). None of this is revelatory, but here are some notes:

Hiring

I've never hired anyone, so like all battle plans, these guidelines will probably change once I actually do hire someone. And, of course, if I had special insight into hiring, I'd be a millionaire business speaker rather than a junior post-doc. One book I'd recommend is, How Would You Move Mount Fuji, which gives a concise overview of hiring generally, and the puzzle interview specifically.

1. The goal of hiring is to avoid bad hires
I  learned this from ... Mount Fuji. Science productivity follows the Pareto principle. If you're a starting PI, you probably can't attract outstanding grad students and post-docs, but you can try to avoid bad hires. And if you hire someone who is unproductive? Firing them is the fastest way to improve efficiency.

2. Hire hackers (or makers)
I've already written two posts that state neuroscience is computer science, so I won't elaborate here. Besides wanting people who can code proficiently, I want people who, in their spare time, can't help but make things (machines, programs, blogs, whatever). Running experiments and analyzing data is straightforward - but you need to set things up first. You can teach people neurobiology, but you can't train people to hack.

3. Actively recruit
Nobody does this, and I don't know why. You're much more likely to get outstanding people if you look for them, rather than hope they find you. The two best opportunities for recruitment are conferences and invited seminars. At conferences, grad students and junior post-docs present posters, which is a perfect opportunity for a five minute, informal interview. You can figure out if the presenter is productive, thinks critically, and has adequate social skill. For invited seminars, the opportunity here is the student lunch, where you can micro-interview a dozen people at once (how to be the PI at a student lunch is another post). As a bonus, if a student stands out, you can easily ask their PI about them later.

4. Recruiting trips
This is my craziest idea, but I love crazy ideas. There are over two billion people in India and China who want to get out, and thousands of them apply to grad schools. Grad schools, however, are unfamiliar with the market, and have a hard time identifying the best applicants. So you, as a PI, spend a few thousand dollars to go to India, and give seminars at various IITs. They would probably be happy to cover room, board, and intra-India travel. But what you're really doing is recruiting: going to lunches with students, interviewing PIs about their students, talking to people about their coworkers. Then you hire the best people you've identified. Two weeks, and a few thousand dollars is a bargain to identify talent. Plus, in between seminars/interviews, you get a cool vacation.

Virtual Seminars


Famous PIs are invited to more seminars than they can schedule. Usually they filter by prestige, so they go to the schools that will appreciate them the least. But this is the 21st century, and we have the internet, so PIs can present virtual seminars to "lesser" schools.

The seminar is the centerpiece of visits, but is usually the same presentation every time. The speaker is often tired from travel during these presentations, reducing quality. So rather than fly a person around the country/world to give the same presentation, I suggest PIs simply record their presentation once (or annually). This would let the presenter polish their talk, and record it when they're well rested, in a familiar environment. Then when the PI receives a seminar invitation from a remote place like Johannesburg, they could send the video, and then make themselves available for questions via Skype afterwards.

This misses some of the ancillary benefits of university visits - meeting other PIs, talking to students - but keeps the core. PIs with bad presentation skills might want to avoid this. Since the presentation is the product, you would want to avoid already published results that could just as easily be read, and highlight the most recent, unpublished findings.

Some places do this in a roundabout way, recording invited seminars and posting them on the web. The Broad Foundation sponsored a series of seminars at Duke, many of which are online. The speakers include Tonegawa, Koch, MacKinnon and Deisseroth.

Services like Khan Academy and Udacity may revolutionize teaching via video, I don't see why academia can't follow.

Lab Meetings

Lab meetings are easy to screw up. The PI, who (hopefully) knows the field best, can dominate the discussion. Presenters can mail in their presentations. Non-presenting members can come to lab meeting unprepared, or otherwise not contribute due to shyness or lack of seniority. And lab meetings can last forever. Given that, here are my keys to a good lab meeting.

1. Call on people
Participation has two goals. First, it keeps people engaged in the meeting, and hopefully the lab. Second, given that everyone in the lab is sharp and diligent enough to contribute, I, as a PI ,want to hear their opinions.

It may seem school-marmy, but one way to encourage participation is to simply ask people by name what they think. If people know they might be publicly called on, they may preempt it by participating. Second, many junior people, and non-Americans, aren't comfortable speaking up, but often have something to say.

2. Record them
In science, people are judged by their writing, speaking, and experimenting, and I cannot abide presentations with poor fundamentals (too much text, unclear aims, etc.). Lab meeting is a perfect place to hone presentation skills in a low pressure situation. After every lab meeting, I would take five minutes to review presenters' slide stacks. For each member, once a year, I suggest recording lab meeting on video, and rewatching it. If I knew I would have to go through the excruciating experience of watching myself for 30-60 min with my PI, I would certainly want to put some work into it.

3. Take notes
You gather together ten people. You present your data. People ask, "Did you think of this or that?" No, you haven't. You (and your PI) don't write anything down, and have just wasted ten hours of time (a mythical man-day!).

The presenter should be taking notes about potential experiments, controls, discussion points. That's their remuneration for their presentation. The PI does not necessarily need to take notes, depending on whether they have private meetings with the presenter. But lab meetings provide a great opportunity for anyone, as a proto-PI, to practice project management. Whenever someone else in my lab presents, I take notes, and compare them to the notes I had for their previous presentation. This lets me ask more probing questions, about followups to previous experiments, or about the general direction of the project.

4. Journal clubs
In the worst journal clubs, no one has read the article, and no one participates. My simple solution to this is to ditch the powerpoint format for journal clubs. Everyone prints out the article for themselves. You wouldn't even need a "presenter," someone could just pick the paper, then the lab as a group could go through it. If someone was not engaged, it would be obvious.

5. Technician attendance is optional
Technicians are often obligated to attend lab meeting, but often don't contribute. While seeing data presentations may help them understand what's going on in the lab, that ain't necessarily so. If they're not getting something out of lab meeting, let them do something productive.

Paper trail day trip: Genomic systems neuroscience

Theoretically, each animal's taste repertoire is determined by the food it eats. For herbivores, the important tastes are sweet and bitter, which lets animals distinguish between calories and poison. For carnivores, they are umami and sour, which help identify whether meat is fresh. Flies, for whatever reason, detect carbonation in water. As omnivores, humans combine the taste repertoires of herbivores and carnivores. A recent paper in PNAS looked at the relationship between feeding behaviour and taste receptors by sequencing the genes for taste receptors in a variety of species.

They started by sequencing the Tas1r2 gene (sweet receptor) from twelve species in the order Carnivora. Of these species, seven had pseudogenized versions of Tas1r2, via a wide variety of ORF-disrupting mechanisms including false start codons, frame-shifts, and premature stop-codons. (Genomic papers' figures are quite staid, which is why this is a day trip.)

The Tas1r2 gene for seven animals has been disrupted. *s denote points where the ORF has been ruined. (Below) The coding sequence of two of these mutations.

After establishing that the sweet receptor had been pseudogenized, they performed an evolutionary analysis that showed that many species had lost the sweet receptor individually, rather than via a shared ancestor (the genomic analysis here was a bit beyond my ken). To verify that these genes were indeed non-functional, they selected two species to perform a taste preference task for sweet. The asian otter, which has a pseudogenized Tas1r2 had no preference; while the spectacled bear, which had an intact Tas1r2, preferred sweet tastes. (I would pay five francs to see these videos. I bet they were adorable.)

Asian otters, which have a pseudogenized Tas1r2, have no sweet preference. Spectacled bears, which have an intact Tas1r2, prefer sweet tastes.

Dolphins and sea lions swallow their food whole, and have sparse taste receptors on their "lingual epithelium" (tongue). Jiang et al sequenced the umami and bitter receptors of these two species, and found that the sea lion lacks both sweet and umami receptors, while the dolphin seems to lack sweet, umami, and bitter receptors.

Genomic Neuroscience

I love how this paper was able to directly connect genes to behaviour. As the cost of sequencing has plummeted - a whole genome now costs ~$1,000, or will soon - I've been trying to think of systems neuroscience experiments (viz. electrodes in the brain) that can leverage genomic information. In particular, I am interested in using genomics to understand individual differences, whereas neuroscience until now has utilized genetics and mutant models to study disease.

The two key parts to a genomic analysis (to me as an outsider) are having genes with diverse alleles, and having a readout that you can correlate with the genes. On the genetic front, neuroscience is lucky to rely on receptors, which can tolerate small mutations and still function. For example, the Serotonin transporter has more than dozen variants in humans.

On the readout front, neuroscience is suboptimal. Since the influence of a single gene can be small, to be able to detect the influence, one wants to measure many traits from many animals. In contrast, in neuroscience, one thoroughly characterizes a small number of cells in a small number of animals. For example, I record from the olfactory bulb, and determine the odorant receptive field of ~50 cells in one animal; for a set of experiments, I might record 5-10 animals. This is time intensive, and not specific to any particular gene of interest. To get a data set large enough to correlate with genes, I would need to record from fifty animals - a year's work - with no sure outcome. Simple forward genomics will not work for systems neuroscience.

Ultimately, I think the best genomics-of-individual-differences experiments would work similarly to the disease paradigm: you perform a behavioural screen on a large number of mice, sequence them all, and identify the alleles of interest. Then you record from selected brain areas of the interesting mice.

Given this framework, I tried to think of an interesting experiment. The first set of genes I considered were the chemosensory receptors, whether for olfaction or taste. But correlating receptor mutations with function is really a biophysics question.

So this is my best idea: there are a large number of signaling molecules involved in feeding behaviour, including leptin, insulin, cannabinoids, AgRP, etc. Most studies have concentrated on knocking these genes (the fuck) out. However, feeding behaviour is quite complex. Individuals can vary not only in outcome (weight), but in how they feed (sporadically or continuously) or how body state effects their behaviour. So I propose devising a set of simple behavioural experiments to tease out differences in feeding behaviour. Body weight is the simplest. You could quantify how quickly satiety sets in during feeding; or how quickly satiety fades. Then for all the subjects, you sequence the genes of interest (or the whole genome if it's cheaper), and hope to find a strong correlation between some alleles and feeding behaviour. Then you take mice with specific alleles, and record from their hypothalamus during feeding to see if their brain is acting different.

In the end, I'm not entirely satisfied with this experiment. It's a fishing expedition, although I guess most genomic experiments are so. Perhaps the key is fishing in a well-stocked lake. In any case, genomics is getting ever cheaper, and there are great rewards (i.e. prestige) for whoever can figure out how to combine genomics with more sophisticated assays like in vivo physiology. They can even coin a term like geneuromics, or neurosomics.


References


Jiang P, Josue J, Li X, Glaser D, Li W, Brand JG, Margolskee RF, Reed DR, & Beauchamp GK (2012). Major taste loss in carnivorous mammals. Proceedings of the National Academy of Sciences of the United States of America, 109 (13), 4956-61 PMID: 22411809