Striatal time cells, transgenic birdsongs, stuttering mice and more – Birdsong4 Sattelite and #SFN14 Notes
Joe Paton from the Saltzman lab – Time encoding cells in the rodent striatum.
You can use an operant conditioning train animals to press a lever and get a reward. Using a fixed interval paradigm, you then do not reward the animal for lever presses until a certain time interval is passed. Animals will learn roughly learn this time association, and pause from lever presses until a point some intermediate time before the interval will expire, and then they begin pressing the lever again.
If you record in the striatum of rodents that have learned this task, you see neurons that fire at every point in the fixed interval and rescale with the fixed interval, if the interval changes. They saw that the rescaling was always slightly subproportional, and also that striatal cells multiplex information about action and time.
I think this is a really great clear electrophysiological link between striatal activity and the task being performed. I wonder if labs have tested mutant mice such as FOXP2 KOs or Shank3B knockouts with proposed striatal defects in tasks like this.
Also, it’s harder to learn the association between a Conditioned Stimulus (CS) and Unconditioned Stimulus (US), if you stretch out the time delay between the CS and US, while keeping the inter-trial interval the same. However, for some reason if you increase the inter-trial interval proportionally with the delay between CS and US, then that increased difficulty of learning the CS/US association doesn’t occur.
Maybe, it’s more difficult to make the association between CS/US with a longer delay, and therefore it takes the brain longer to process the association. But, if you leave the brain to process the old trial offline without interference from new tasks, then it can form the association. To restate that, when the trials happen too quickly, processing the subsequent trial interferes with the ongoing processing of the previous trial and interferes with learning.
Carlos Lois – Transgenic Songbirds – Genetic tools to investigate brain circuit assembly and cellular basis of behavior.
Lois mentioned that HVCà RA neurogenesis occurs during song learning. Neurons migrate into the HVC after making soma-soma contact with resident neurons. A thought I had never really had was how much is going on during windows of time when fetuses are learning about speech. A lot of developmental psychology work has shown that babies learn to identify prosaic cues of language, their mother’s voice, etc. while in the womb. I wonder to what extent these process are concurrent with and depend on neurogenesis. More generally, do we have a good sense from autopsy studies and/or radiation exposure studies when neurogenesis ends in normal human development for cortical areas, the striatum, etc.
Collaborating with the Gardner lab, Lois designed an adeno-associated virus (AAV) to express GCAMP6 in a small population of neurons in the HVC. (For some reason with the current viruses and promoters they’re getting much smaller yields of infected cells than people normally do with mice. ~2,000 neurons in birds compared to ~20 million in mice. Of course this isn’t always bad–sparse labeling can be good for measuring morphology or separating cell-autonomous effects from emergent effects). By mounting a low weight CMOS camera on top of the birds head, they could record activation of many cells simultaneously in a singing behaving bird (and know where these cells lay relative to each other in the rostro-caudal and medial-lateral axes).
Lois’s group also infected HVC with a virus driving expression of a bacterial Na+ channel ‘NaChBaC.’ Neurons in this channel went from firing single discrete spikes to prolonged depolarizations. As the channel starts being expressed, song gets completely distorted, but remarkably a few days later the bird has found a way to compensate and get song back to normal. They said that they did histology to confirm the infected neurons were still alive and expressing the channel. However, maybe the bird simply down-regulated all of the synaptic strengths of the infected neurons, and there is enough redundancy to produce the song with the remaining neurons.
Finally, Lois showed data from transgenic zebra finches that have been germline transfected with RNAi to knock down CNTNAP2 expression, which is associated with developmental language disorders and autism. They showed that the CNTNAP2 knockdown birds showed normal learning of simple syllables, but impaired learning of complex syllables. He also mentioned that they had developed a line of GCAMP6 transgenic animals which is pretty exciting. I believe viral infection of germline cells cann be combined with the CRISPR-CAS system to cause deletions and premature stops at target genes, and theoretically even induce homologous recombination (by also providing a template strand complementary to the region where the DNAase has cut the DNA.
Someone asked a question about CRE lines in birds, and Lois said that economically it probably wasn’t viable. Not enough people do research on birds to justify that kind of investment, and making such animals would be a largely thankless job. However, there are still a ton of experiments that could be done on simpler transgenics such as: plain KOs, KDs, transgenics that express fluorescent proteins and optogenetic channels. I think just having that level of genetic tools combined with the song system’s anatomical modularity, the strength of song as a behavior, and the strong phenotypes seen in FOXP2 KDs and CNTNAP2 KDs show that genetic songbirds could be an extremely powerful models for diseases of speech, social behavior, and motor learning. Further, birds have similar reproductive cycle lengths as mice and live longer, so stocks of transgenics can be bred in reasonable amounts of time. I wanted to ask Lois what he thought about creating inbred strains of zebra finches, but I didn’t have the chance. (I’m almost tempted to start this as a side project in my apartment, but I’m afraid of then bringing an infection into the lab.)
Lois had the impression that the NIH was not interested in funding zebra finch transgenics. His grants came from mouse projects and work he did on zebra finches were side projects.
A member of the Simons Foundation “SFARI program” said that they have wide agreement that rodents aren’t “sophisticated.” However, the program is most concerned with investigating the rare genetic causes of autism and generating high-throughput screening models for autism. Personally, I think this is short sighted and the basic science just isn’t there yet. We really need to investigate the basic mechanisms of communication, imitative learning, etc., and I think songbirds are one of the best model organisms for these behaviors—and without a doubt better than rodents. So, while using genetics we’ve done amazing work on specific etiologies like Rhett syndrome, as someone in the audience rightly pointed out there are still almost no examples of rational drug design in neuroscience and neurology, except for the new example of orexin antagonists for insomnia and how good of a drug they’ll be in the long term is unclear.
Dennis Drayna – genetics of stuttering and mouse models
Stuttering affects 4% of the population at some point in their life, but only .5-1% of adults are persistently affected and at that point there is a 4:1 ratio of males to females.
Most human genetics has focused on this persistent stuttering population. They’ve shown that it’s about 85% heritable and adoption studies have shown no evidence that stuttering is learned. (So if a child is adopted by a parent who stutters, they are no more likely to stutter themselves.) Early linkage studies attempting to locate mendelian genes involved in stuttering in the United states were not successful. However, studies looking at linkage in a consangiunous family from Pakistan and a large polygamous family in Cameroon identified two proteins involved in intracellular targeting to the lysosome. These mutations cause ~50% of the normal enzymatic activity, and homozygous mutations of these genes been previously identified to cause sever health problems such as mucolipodosis. Also interestingly, in line with the 4:1 ratio of males to females, there are many females in those families carrying the stuttering genes that do not show persistent stuttering. When they then screened completely unrelated populations of stutters from other continents 10-20% cases showed mutations to these genes. This is an astoundingly high number compared to similar screens for mutations related to autism, MR, etc.
Persistant stutters with the heterozygous mutations seem otherwise completely normal in normal and neurological screenings. Drayna suspects that there may be a unique group of neurons that is both important for human speech production and uniquely sensitive to this metabolic deficit. Someone in the audience also suggested that maybe there is a developmental time window that is important as well and corresponds with a time period in speech learning.
Drayna created KIs of the familial mutations and studied how they influenced mouse ultrasonic vocalizations. Mouse vocalizations are not generated by vibration of vocal folds and may therefore be more analogous to whistles. Different inbred strains produce different calls, and cross-fostering experiments have shown that these differences are innate and not learned. Crossbreeding studies also show that they segregate as near Mendelian traits. The stuttering gene KI mice showed some differences in vocalizations such as increased intrabout intervals. Drayna wants to use conditional KOs with CRE lines to test different cell populations and try to narrow down what population is producing this change. I’m curious if people have worked out the circuits involved in producing these vocaizations. I also wonder how analogous mouse vocalizations will be to speech, especially since it is unclear if they have vocal learning abilities. It would definitely be interesting to KD these genes in neuronal populations of songbirds, though there may be difficulties since they are housekeeping genes probably necessary for survival. Depending on the gene length perhaps one could produce cells that both knocked down the homolgous protein and introduced a mutant version? However, getting the right level of decreased expression might be difficult, so maybe they aren’t so great of genes to study in birds after all…
Jesse Goldberg – Trial and Error Learning in songbirds and mice.
Gave a really interesting presentation which featured an impressive zebra finch song impersonation.
They trained birds using a white-noise (WN) conditioning paradigm that target a single syllable and randomly on 50% of trials delivered a pulse of WN. After days of training, they simultaneously they recorded from the VTA. Most neurons showed activity locked to movement, but they also found a small proportion of anti and found a small population of antidromically identified VTAàX neurons that increased firing on the 50% of syllable renditions when WN is absent and decreased on the renditions where WN was present. This finding suggests a 2-way learning rule where the bird is not comparing song to a copy of the template, but rather to it’s own expectations of song. In other words, on the trials where no WN distortion is present the bird interprets its song as unexpectedly good and increases DA release, on the renditions where the WN blasts occurred, it interprets the song as unexpectedly bad and shows a decreased DA release. This type of DA firing has been reported with striatal reward-based learning in the past. Goldberg also suggested a system where errors are processed in the AIV, which then projects to the VTA, and the VTA then sends dopaminergic projections to Area X.
He also created a behavioral task using joysticks for mice which replicates the “center-out” task in primates. Mice are trained with food rewards to move the joystick to push a dot outside of a circle. By putting the joysticks in the mice’s home cages they can get the mice to produce ~400 trials per day. Although the mice are rewarded for moving the joystick in a way that moves the cursor outside of any part of the circle, the mice will converge on a motor plan that moves the cursor in a particular direction and edge of the circle. Experimenter’s can then switch the task so that the reward is only given if the animal moves the cursor to one corner of the circle. So if the mice are targeting the bottom right corner of the circle, the experimenter can switch the task so only the top right corner is rewarded. The mouse will eventually stumble upon the new correct solution and then gradually update their motor plan. They can measure and quantify the trajectory as it switches from trial to trial.
LMAN shell and AId may be important for song learning. If you label LMAN with anterograde tracers, you see projections that project continuously to RA and the adjacent AId. LMAN shell shows a dramatic increase in size that corresponds with the start of song learning and then decreases in size as song learning crystalizes. LMAN lesions disrupt song learning, and in learned song they disrupt syllable sequence but not syllable production. Further when you do eletrophysiologcal measurements from LMAN shell, you see some neurons that are specific for the bird’s own song (BOS) and others that are specific for tutor song. In contrast, recordings from LMAN core show cells that respond to both BOS and tutor song, but do not significantly differentiate between the two. During song learning there is a convergence between core and shell pathways at the AId, because core LMAN appears to send transient projections to Aid specifically during the early stages of learning.
Olveczky – Motor skill learning in rodents
Olvecsky set out to reproduce some of the benefits of bird song in rodents: self-generated learned behavior, complex and stereotyped behavior, continuous and quantifiable behavior that is amenable to neural recording. Hey came up with a lever pressing task for mice, and like Jesse Goldberg, used home cage training to streamline the behavioral process. The task involves the mice pressing a lever twice with a precise delay in between the presses. Mice randomly stumble on complicated behaviors that involve continuous movements, first pressing the lever while rearing, pawing at the cage walls (and in one case licking the wall), before returning back down to hit the lever again,with the rewarded inter-press time window. This task really highlights the motor circuitry’s ability to explore motor space and discover a solution for a task, and then learn to repeat it, once it’s rewarded. They paint the mouse’s paw and can record that paw’s trajectory as it completes the task, and presumably quantify how movement variability is decreasing across the task. (I wonder if they have some way of constraining the approach angle to ensure a real difference in the movement itself and not just of changes in the projection onto a 2d screen.) They used a counter-weighted tetrode array to simultaneously record from animals.
They also performed both M1 and M2 lesions on animals and showed surprisingly (to me) that while these regions are necessary of learning the task, and animals are paralyzed shortly after the lesions, once they recover the mice lesioned in both M1 and M2 could perform the complicated learned motor task without problems. (Perhaps the motor learning was transferred down to lower areas such as spinal cord or brainstem motor circuits? He brought up the “Lashley puzzle box experiments” where primates were trained to open puzzle boxes, then their motor cortex was lesioned, and two weeks later they could open the puzzle boxes again. I wonder how much inter-species differences and age-related effects contribute to this as well. For example do humans just have a lot more direct control of motor programs from their cortex than rodents and even primates? Also, in young adult humans that get strokes affecting motor cortex do they show complete motor recovery?
In slice pallidal neurons show spontaneous EPSCs that appear regularly around every 50 miliseconds. It seems like each one is innervated by a single cell firing at about 20 hz. They could block this signal with NBQX and showed when they did this, firing of the pallidal neurons decreased variability. (If this finding is true in vivo as well then this regular 20 hz current may be involved in generating the variability in pallidal firing that contributes to motor exploration.) In an attempt to identify the cell type that was generating these EPSCs they stained for Vglut2, a vesicular glutamate transporter, and found a sparse population of neurons in area X that expresses it. (They suggested this population could be analgous to STN neurons in the mammalian basal ganglia, though I don’t know enough about STN to comment.) They also wondered if the glutamate could be noncanonical glutamate release from cholinergic INs, so they did an IHC stain for ChAT (a cholinergic marker) and Vglut2 and did not see colocalization. Someone suggested that cholinergic neurons could make separate cholinergic and glutamatergic synapses and Perkel seemed skeptical. However, I saw a poster at SFN showing that Vglut2 and TH expressing neurons from the VTA and substantia nigra make separate glutamatergic and dopaminergic synapses and that those two markers show no colocalization. Therefore I think it’s reasonable to try to test explicitly for glutamate release from the cholinergic neurons though I’m not sure exactly how you’d do that in songbirds. Use a virus that drives a diffusible fluorescent protein under the ChAT promoter in the striatum. Then you could stain form Vglut2 and look for colocalization with fluorescence. Additionally, in slice you could identify the cholinergic neurons and record from them to see if they spiked at 20hz.
Brawn in Margoliash lab poster
Trained animals in two classification tasks and looked for interference effects. Then recorded animals sleeping patterns with 4 silver wires. Used video camera to confirm whether awake or asleep (because REM resembles awake activity) and then used a clustering algorithm to categorize sleep as REM, slow wave sleep, or intermediate level sleep. They showed that SWS increased during learning of discrimination tasks. Evidently the REM-like sleep in birds also involves eye movements, which is interesting since I believe it’s thought to have evolved independently. Sleep featuring both REM and SWS has only been demonstrated in birds and mammals, with the outgroups such as platypi and emus showing proto-sleep.
Contingent social feedback facilitates vocal learning—vocal learning is increased when tutors respond to infant vocalizations. Goldstein had the idea that contingent social feedback is rewarding. To test this he decided to do a conditioned place preference experiment where he manipulated contingency in two spaces, separated the barrier between them and tracked where the infants spent their time. These results seemed to indicate that contingent social feedback may act as a reward.
He also performed experiments exposing young birds to vasotocin (a bird analogue of oxytocin/vasopressin) and showed that it increased tutor similarity matching. Similarly, vasotocin antagonists decreased tutor matching. I’ve always wondered about the apparent connection between oxytocin’s influences on maternal behavior and also milk letdown in mammals. I wonder if there are similar relations between hormones that increase behavior for singing to young birds and also parental feeding behavior.
Dana Lipkind in Hahnloser lab
They use a box-learning experimental paradigm, that is based on the fact that birds can change tutors early in development. If you have the bird learn a recorded song and then modify the recorded song, then the bird will shift it’s song to match the new recording.
Lipkind showed that if you rearrange the sequence and slightly shift the pitch, so if the bird is singing ABC and you switch it to ACB+ (where B+ indicates a version of B shifted up one semitone), the bird will first correct the pitch error and sing AB+C, and then it will learn the two new transitions AàC and CàB+ individually at later points in time. She pointed out that this suggests Zebra Finches have local template matching mechanisms that are context independent and also a larger time scale transition/syllable order song learning mechanism.
They then performed experiments shifting from the song ABAB to AB-AB+ (where B- is shifted down one semitone and B+ is again shifted up one semitone). One possibility from this experiment is that opposing error signals would keep the pitch for B at one place. What actually happens is that the Bs shift one way or the other AB-AB- or AB+AB+, and then bird finds another way to make the missing syllable. For example the bird might sing AB+AB+, and you see the bird change it’s long call’s pitch and shape to become the B- and eventually it becomes AB+AB-. (Someone else’s poster session I believe showed that sometimes The bird will continue to sing AB+AB+ but the long call still shifts to B- even though it doesn’t get incorporated in the song.)
According to this Lipkind proposed a “musical chair” theory of syllables, where the bird tries to template match tutor syllables with syllables that are nearby in acoustic distance and also whether or not that template site is “occupied” with a matching syllable.
From another poster using this method, I wrote that past day 75 they do not acquire new shifts up or down, but sometimes they can switch spontaneously back to old previously learned songs. Also, when the bird doesn’t completely learn the song, say AB-AB+ gets learned as AB+AB+, you will sometimes see the call shift up into the exact pitch of B-, despite the fact that it doesn’t actually get incorporated into the song. I should check that the call shift really does get incorporated in the song with the Lipkind story, but that was my understanding.
You see song-like activation of syringeal muscles during sleep, but it does not directly recapitulate daytime experience. He proposed that perhaps it represents nighttime “trial and error” to help maintain song during adulthood without making errors. He also proposed that possibly sleep activation is required in maintaining the superfast muscles of the syrinx. Testing these hypotheses directly–through night-time inactivation of the nerve–is difficult because it is hard to avoid inactivating the vagus muscle and therefore sleep.
Editor of Birdsong, Speech, and Language along with Martin Everaert.
Proposed that the NCM may be analogous to Wernicke’s area, as NCM lesions in adults impair recognition of the tutor song but don’t impair song production. When you play tutor songs to animals you see greater ZENK activity in the left NCM. You also see lateralized NCM activation during sleep in juveniles. Bolhuis wanted to make the point that he is distrusful however of attempts to make parallels between birdsong and language.
Was also interested in NCM lateralization of birdsong, specifically if birds that are more NCM L-lateralized learn better, or if birds that learn better become more L-lateralized. She used MEK inhibitors (which block ERK but not synaptic transmission), and administered them to the different hemispheres either during the morning when they were exposed to the tutor or in the afternoon when they did not hear song. In the birds that were injected in the L hemisphere, during the morning when they interacted with the tutor, they show 50% of the normal amount of song learning.
Andires Ter Maat
They came up with an ultralight wireless system for juvenile birds to carry around a microphone in a backpack configuration. Each microphone lets you resolve the sound of that individual animal and wirelessly transfer it to be recorded with 22 or 44khz sampling. The backpack weighs <5%. They can have many birds carrying around microphones and track their calls and songs, and where they are when they vocalize.
Looked microscopically at recordings from white noise conditioning experiments–which trials were escapes and which trials were hits. He was interested in whether birds were using escape learning or hit learning.
Hahnloser created a model of sources of variance: some which can be used for learning (efferent noise) and some which cannot be used for learning (inaccessible noise). He saw that birds sing higher pitch in the morning and lower pitch later in the day, so if birds were being punished for low pitch renditions they actually can increase errors throughout the night. He therefore added an additional element of inaccessible noise to reflect this pattern. In addition there is a term for pitch bias, which is the result of shifts through the white noise conditioning and then disappears over time if the white noise conditioning ends.
This work built on their previous finding showing that bilateral lesions to NCM did not hinder WN-mediated learning, but once the WN was stopped animals with NCM struggled with finding their way back to baseline pitch.
He showed a finding I had not heard of where if you induce a pitch change through white noise conditioning and shift the pitch high enough, you reach a point beyond which you can no longer learn. This point is proportional to the variance of pitch of that syllable. This seems to have some interesting similarities with the headphone experiments done by Sam Sober. I wonder if this effect is also mediated by the speed of step transitions.
Hahnloser also suggested that we need to standardize the way we do similarity scores. For example select 10 songs randomly from two animals, and do the 100 pairwise comparisons for similarity scores.
There was a debate this year about whether or not HVC is encoding time or muscle-related activity (like gesture extrema such as the onsets and offsets of syllables).
They have developed a method to compare song similarity when the bird has multiple tutors by breaking up the song into smaller pieces and comparing the similarity to each of the possible tutors, which I believe is written about here: http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0096484 (An Automated Procedure for Evaluating Song Imitation).
Has developed a machine-learning technique for optimizing white noise templates using exemplars and distractors and a gradient descent algorithm.
There seems to be more connectivity on the antero-posterior axis of HVC based on electrophysiological approaches where they stimulate in one point and look which directions the signal propogates. This is also consistent with some of the activity shown by Gardner and Lois with GCAMP in HVC. Additionally, creating a microlesion between medial and lateral HVC does not seem to have a large effect. If you ablate LMAN and then ablate medial HVC the bird will omit some syllables, usually those from the end of the song. To me this was reminiscent of Ashmore in the Schmidt lab where they show microlesions of RA cause dropped syllables from songs. Followup work seemed to indicate that microlesions to dorsal RA were more likely to cause syllable drop-offs while ventral RA were more likely to change spectral parameters of syllables. (I wonder if medial HVC projects to dorsal RA and both of these lesions are disrupting the same pathway.) The Frank lab also showed that if you ablate LMAN and then medial ablation, the bird sings all of the syllables but out of order.
Is working with an interesting behavioral paradigm, where one tutor sings to 10 juveniles. One thing that surprised me while looking at this data is how much more the young birds sing than the tutor. I really need to take the time to observe these birds more and gain a better intuition about their behavior.
Lab members also showed that if you show the birds videos, they are more likely to copy the song of a tutor that is facing them compared to a tutor that is facing away from them.
Did in vivo DA cyclic voltometry in urethane-anesthetized birds (sparrows?). Used glass coated carbon fibers and then played different calls to elicit dopamine release.
Eva Ilhe Used a microdyalis HPLC method to analyze DA in the striatum during directed song. Showed increased DA in the presense of a female. Also showed a sharp increase in the number of song sung shortly after interacting with the female.
Jeff Knowles trained female zebra finches in a Go/No-go auditory discrimination task. They had to discriminate between six syllable songs where the first three syllables (ABC) were all the same, and the last three were all different (XYZ, X2Y2Z2, etc.). He trains the animals in the task and then does acute recordings in Field L and CM.
He found neurons that selectively respond to songs (more for the go song, but also some selective for some of the distractor songs) that show a peak in activity after the X. This makes sense as the X is really the first information bearing syllable that determines whether the bird should ‘go’ or ‘not go.’ In this sense the first three syllables are not information bearing for the task, however they do provide context: the syllable you are looking for may be coming up. These selective cells cannot be model well with simple linear additive models of their response to the syllables in isolation. Also, if the contextual syllables A,B, or especially C are removed, the animals are worse at the behavior and the selective cells show greatly decreased response. Using this method, Knowles can also play all of the syllables in random sequence and look for reverse correlations of patterns of syllables in a method similar to early STRF work. I was curious if Jeff knew how this plasticity was taking place and he said he was interested in investigating the avian equivalent to the nucleus basalis and cholinergic projections to the auditory cortex. Could such cholinergic projections could also be involved with the bird learning and interpreting it’s own song.
I’m curious whether similar selective cells will be observed in more naturalistic conditions: a bird learning from a tutor. My guess is you would also see non-additive syllables that effectively therefore have song-order selective responses. Development of such cells may be important for filtering for conspecific songs in a noisy environment, and discriminating between songs of different species.
Science of Scientific Writing – George Gopen and Judath Swan
Someone recommended this as a primer on scientific writing. There are a lot of these floating out there, but this one seems
Life history of a species versus genetic relationship
This is the concept that vocal learning is a solution to certain challenges a species faces. Unrelated species facing certain challenges
Chronic recordings in birds, rodents, and monkeys
Zebra finches can live 10-12 years which gives them an advantage over rodents as a small model organism for studying very long term chronic electrophysiological recordings. When you start looking at recording quality on the timescale of years, rodent recordings are confounded by the aging process.
Soft polymer probes where you insert them with a stiffener and then pull that out.
California thrashers have >100 syllables (compared to the ~12 for Bengalese finches).
In rodents, there may be fast kinetics for dopamine release and reuptakein the DL striatum and slow release and reuptake in the ML striatum. (Experiments where they stimulate the mFB and then record from the striatum with cyclic voltometry.
Exploration induces double strand breaks (DSBs) in rodent cerebral cortex, which may be repaired during sleep.
If you lengthen the syllable of a song, over progressive generations they’ll lower the syllable length each generation.
Seasonal versus Non-seasonal breeding has evolved many times and depends on the environment. Long seasons from far North/South climates and high elevation make birds more likely to be seasonal breeders.
There are two P.I.s in the songbird field named Sarah Wooley
Filed under: Uncategorized | 2 Comments