Blog

A Modern Classic, Part I

Today I’ll be revisiting a paper that had a massive impact both on what we know about control of energy balance, but also how I think about and approach my experiments. I’m talking about Atasoy’s 2012 Nature paper with the pretentious title1, in the study that first introduced me to optogenetics. Or at least to the possibility of controlling awake mouse behaviour using optogenetics.

This paper came out not long after the seminal work by Krashes et al., where he used DREADDs to drive to activity of AgRP neurones in vivo, and show the direct effect on feeding when these neurones are activated2. These papers were published towards the end of my PhD, and I was very keen to use these exciting new tools in my own research.

In fact, one of the first things I did in my postdoc was to help set up the use of targeted nanoinjections of AAV DREADD’s in our transgenic mice. It was only after a couple of successful experiments with DREADD’s that I even began to think about using optogenetics – I really wanted to develop the easier stages before jumping straight into the more advanced stuff.

Anyway, back to Atasoy’s paper. After some initial testing to make sure they can express ChR2 in AgRP neurones, and to demonstrate inhibitory input onto POMC neurones with electrophysiology, they take the optogenetic stimulation in vivo. They show, firstly, that you get increased food intake with coincident stimulation of AgRP and POMC neurones, demonstrating for the first time that the feeding drive for AgRP neurones is outside the Arcuate nucleus, ie. that the acute feeding action of AgRP neurones was not mediated by the suppression of POMC neurone activity (Figure 1).

But, if the acute feeding effects of AgRP neurones are not mediated by action on POMC neurones in the Arcuate, where are they mediated? Atasoy demonstrates the power of optogenetics for investigating neuronal circuits, by activating AgRP neurone terminals in awake behaving animals (Figure 2). Picking 2 areas with dense AgRP neurone terminals and known for controlling food intake, they target the PVH and the PBN. The results speak for themselves, with such a drastic response from the PVH, and nothing at all from the PBN.

There are more figures in this paper, but for me these were the most important findings. I think the power of optogenetics comes down to several factors, allowing us to overcome a number of the most challenging aspects of studying the brain:

  • High temporal precision – can get physiological responses instantly, and influence behaviours that rely on millisecond response times
  • Circuits – optogenetics allows us to investigate the neuronal circuitry involved in complex behaviours by stimulating neurotransmitter release in target areas
  • Stimulation patterns – the flashing light can be patterned to mimic neuronal firing patterns, which can produce differing behaviours even from an otherwise identical experiment

For those interested to read more, there is another early paper that really influenced my thoughts on optogenetics, which was a 2013 paper from Betley et al.3 For Part II, I’ll be investigating the next big advance in in vivo technology which has changed my approach to understanding energy balance.

1. Atasoy et al., Nature 488, 172-177 (2012) Deconstruction of a neural circuit for hunger.

2. Krashes et al., J Clin Invest 121(4), 1424-1428 (2011) Rapid, reversible activation of AgRP neurones drives feeding behaviour in mice.

3. Betley et al., Cell 155, 1337-1350 (2013) Parallel, redundant circuit organisation for homeostatic control of feeding behaviour.

Of Mice and the Internet of Things

Ever since the long lost time of my PhD (about a decade ago), I have been excited by telemetry. More specifically, the use of telemetry and wireless technology to obtain high quality physiological data from mice.

During my PhD, I used telemetry to record ECG in transgenic mice, using DSI’s transmitters (www.datasci.com). ECG was actually my second choice for investigating cardiovascular control in our knockout strain, but the blood pressure transmitters were too challenging for me to be confident in spending that much of our limited grant funding on.

The reason we wanted blood pressure recordings is that it is a much more reliable readout for the stimulation of the cardiovascular system, as there are many reflex control on heart rate that make it tricky to understand exactly what is going on (for example, if you stimulate cardiac output, you might well increase heart rate along with blood pressure, but then your baroreflex kicks in and the heart rate drops). As it turns out, I was able to delve into heart rate variability analyses using the ECG transmitters, which formed a large part of my thesis, so all turned out fine.

Anyway, I have more recently been using DSI telemetry to investigate body temperature and locomotor activity in mice, but have found myself getting annoyed. Between the surgery and singly housing animals, crappy battery life and expensive refurbs, short range recording and signal dropouts, it’s been getting on my nerves. And it’s 2021, why are we still using transmitters and recording technology that was developed 20 years ago?

After some time niggling away at the back of my mind that there must be a better way, I had a conversation with my dad about something he’s been working on (he’s technically retired, but is working with an old friend from the oil drilling business) about uses of the Internet of Things. This is one of those terms I’d heard, but thought it was a bit of a gimmick, like amphibious cars, or smartphones.

In fact, it turns out technology has reached the point where everything can be connected. For example, from the industrial sector that he was talking about, they can monitor the temperature of a certain piece of machinery, the pressure inside the system, performance indicators, pollution levels etc. Really, anything that might possibly want monitoring can have a sensor placed inside, which will be quiescent until certain parameters are met, and then it pings out a signal. This means that the battery drain is negligible, and the sensors can remain in place for years. Hearing this, I was excited to check out the state of the technology for my experiments; as researchers we are prone to just use the same as we always have. Here’s what I envisaged:

  • A transmitter that is small enough to implant through a (fat) needle, negating the need for pesky surgery
  • The signal is long-range enough, and includes identifying information, that you can have a single receiver for a number of group-housed mice
  • Implants are single-use – cheap(er) than DSI and disposable, so no faffing with refurbs and sterilisation

It’s possible that my desired were too restrictive, particularly with regards to the maximum size, because after much internet scouring, the best I was able to find was implantable ID chips like this from Unified Information Devices (UID – www.uidevices.com):

These are injectable RFID chips that are primarily used for mouse identification. Apparently, such things are fairly common in industry, where you would subcutaneously implant every mouse with an RFID chip, allowing you to essentially scan a mouse like a barcode and it brings up all the relevant information about that animal.

The UID implants take the identification a step further, also providing a temperature readout along with the animal ID. Unfortunately, this normally requires you to “beep” the mouse with your reader at very close (likely skin contact) range. However, UID also produce a “mouse matrix”, that can read the info (including body temp and track movements) from outside the cage. They are quite pricey though.

The reason this is needed is that the RFID chips don’t have a battery, instead the tiny microchip is temporarily powered by the electromagnetic waves from the reader itself (same as the chips in most modern cash/credit cards).

So, I’ve been thinking, couldn’t you put a tiny battery inside, and low-powered, infrequent data transmission? You would only have to transmit every 5 or 10 minutes, so surely even a tiny battery could manage that? There is a product called Anipill, which takes a similar approach (www.animals-monitoring.com). Their 1.7g implant sends out data at various timeframes from 1 min to 1 hour, and you can record data from a number of animals (up to 8) simultaneously from a single receiver, which improves the animal welfare by allowing group housing. This seems to be exactly what I had been thinking of, but with a capsule size of around 18 x 9 mm, it is far bigger than I had wanted.

Sadly then, if this company can’t make transmitters anywhere near small enough for injecting, then it’s probably not possible, at least with current technology. But, this is an area that is only improving from the advance of technology, so I will not lose my interest so easily.

Forsaking cre

I read an interesting paper that came out recently, by Garau et al. about an inhibitory AgRP to Orexin connection that mediates exploratory behaviours1. They use optogenetics and photometry to investigate the role of Orexin neurones in some exploratory behaviours, with particular relevance to valence and anxiety.

I’m particularly interested in their use of AAV’s to investigate the Orexin neurones in this paper. I have had serious issue trying to investigate Orexin neurones in my own work, because of the lack of suitable cre-driver lines (we tried an Orexin-cre in the past, but were unable to get any detectable cre recombination using AAV’s or crossing with reporter lines). However, Garau et al. forsake the usual cre-dependent AAV’s when it comes to investigating the Orexin neurones, and instead use Orexin promoter-driven expression of GCaMP6 or inhibitory ArchT.

Trying to copy the promoter region for a gene can be troublesome, particularly if you want to then package it inside an AAV (which can only take around 5-6kb, whereas promoter regions can extend for 10’s of kb). However, it seems to work well for the Orexin gene, and has been done previously by Saito et al. to drive tdTomato in a reportor AAV2. Here, Garau et al. report that their Orexin-ArchT AAV tags around 60% of Orexin neurones, with an extremely high fidelity of over 99% (Figure 1). This is important, because it means that, while they might miss some number of the Orexin neurones, they don’t have off-target expression.

They also demonstrate using current-clamp electrophysiology that the ArchT functions as expected in the Orexin neurones (Figure 1C; always an important control to check in any new optogenetics paradigm you use). They then proceed to show a definitive effect of silencing the Orexin neurones in vivo on real-time place preference (Figure 1E), which is a test I have used as well, and I find very useful in its simplicity.

I won’t go into more detail here, but it is an interesting paper, and I would recommend anyone reading this to go and check it out.

1 Garau et al. (2020) J Physiol 598.19 pp 4371-4383 Orexin neurons and inhibitory AgRP -> orexin circuits guide spatial exploration in mice.

2 Saito et al. (2013) Front Neural Circ 7: 192 GABAergic neurons in the preoptic area send direct inhibitory projections to orexin neurons.

Going Fibre-Free

Optogenetics is a fantastic technique, enabling the control of mouse behaviour with a high degree of temporal and neuron-specific precision. However, due to the high levels of light power needed to activate channelrhodopsin and its variants, a typical system will use a high intensity laser or LED connected via fibre-optics.

In my experience of doing opto’s, the fibres have proved to be the biggest technical issue, due to the high level of stress they incur in the mice. This comes as a result of a number of factors, including the requirement for housing the mice in an open cage, with a tether on the head that places some amount of torque on them at all times, and requires them to have their head upright. Also, because of the fragility of the optic fibres, the mice will often be housed alone and in a sterile environment. All this amounts to both unnecessary suffering for the animals, which as researchers we a morally bound to reduce whenever possible, but also to stress that will inevitably impact on any behavioural measure you are investigating.

This brings me to a paper that was published recently by Anpilov et al. in Neuron1, where they developed a wireless optogenetic stimulator to overcome these issues with fibre-connected opto’s and investigate social behaviours in a “semi-natural” setup.

From a technological standpoint, I am very interested in the device they developed, which is my primary interest in this paper. Their device is almost ludicrously simple – it’s just an LED connected to 2 button batteries via a magnetic switch, which results in a total weight of about 1g (Figure 1A). They connect this to the implanted optic fibre and cement the whole lot onto the mouse’s head. Then the LED can be switched on externally by proximity of a magnet (Figure 1B). This means that the mice can be kept in a complex group-housed environment and the opto’s switched on at will remotely (Figure 1C).

In fact, the researchers placed the magnet above the feeder, so they don’t ever need to disturb the animals. This enabled Anpilov et al. to influence aggression, grooming and other social behaviours in response to oxytocin activation, which would otherwise be extremely challenging to investigate using classical fibre-connected opto’s.

Figure 1. A Wireless Device for Prolonged Optogenetic Manipulation in a Semi-natural
Setup
(A) Schematic illustration of the wireless device. Two batteries are connected in series to an LED through a magnetic-field dependent reed-switch. The LED is attached on top of an optic cannula positioned above the dorsal part of the PVN.
(B) A device mounted on a freely behaving mouse activated by a magnet.
(C) Schematic illustration of the semi-ethological arena and software-controlled electromagnet installed on the feeder. The arena consists of an open 70 x 50 cm box containing a nest, feeders, water, elevated areas, and barriers.
(D) Light power emitted at the tip of the optic fiber as a function of the number of 2 s light pulses. Battery capacity is sufficient for over 215 pulses.
(E) Section through a 3D map of blue light intensity along the axis of an illuminating fiber in graymatter. The slice was imaged from below as the fiber was lowered through. The section is superimposed with a contour map of iso-intensity lines corresponding to light intensity levels. Light intensity >= 8microwatt/mm2 is sufficient for effective SSFO photoactivation. Taken from Anpilov et al. 2020 Neuron.

However, I do have some concerns to voice about the device they used, related to the actual light output they can achieve. Essentially, you can’t expect high light power through an optic fibre from a battery-connected LED – the light scatter doesn’t allow it. The upshot being that Anpilov only got 3.2mW/mm2, which is much dimmer than a typical opto system which will deliver 100-300mW/mm2 (our current system delivers around 175mW/mm2). Furthermore, their system has a maximum on-time of around 400s, which is why they cleverly used the SSFO to provide long-term activation after a very short stimulation pulse.

So, while the system developed in this paper clearly works well for their application of oxytocin-mediated social behaviours, I can imagine the further applications being relatively limited due to the following reasons:

  • Need to use SSFO or equivalent long-term responsive rhodopsin loses the high temporal precision you need for many optogenetic applications
  • Very low light output will likely mean further applications will be limited to those with very strong/sensitive behavioural responses, and probably with very dense neuron populations
  • The LED is only ON in close proximity to a magnet, which limits the design of environments that allow for activation without disturbing the animals.

Overall, I was impressed by the paper and am very interested in their approach to negate the drawbacks of fibre-connected optogenetics. But, I think their device has its own drawbacks, which if overcome would produce some very powerful tools with wide-ranging applications for in vivo optogenetics.

1 Anpilov et al. 2020 Neuron Wireless Optogenetic stimulation of Oxytocin neurons in a semi-natural setup dynamically elevates both pro-social and agonistic behaviours.

Thoughts on genotyping.

First week back after the Christmas shutdown, and all my mouse strains have had litters weaned. The couple of hours it took to earpunch the 70-odd animals was inevitable, if tedious, and can be chalked up as a necessary requirement for working with transgenic mice. However, after extracting DNA from all those samples and then at least one PCR for each, I started to wonder how necessary the endless hours of pipetting might be.

Our University has an agreement with an external paid genotyping service, which some researchers in the animal facility have been using. However, our lab as a rule haven’t used it, because it is expensive, and we have always thought it would cost too much to get all our mice genotyped this way. Also, I’ve always maintained that genotyping builds character, which is why I often allow new students and technicians the opportunity to learn using my samples.

Anyway, given the time investment needed to successfully genotype the mice, I wanted to compare the actual cost per sample of using the automated service compared to doing it ourselves, including the cost of reagents and postdoc salary time.

Looking back through my genotyping lab book, my usual batch of DNA extractions involves around 35.85 samples, and my PCR’s average 26.35 samples. In the table below, I have estimated the various costs associated with genotyping this typical batch, including the likely time spent by yours truly, that could otherwise be spent on more productive lab activities. I have estimated my time to be worth 22.9 £/hour, based on my salary and working 37.5 hours per week for 47 weeks a year (which is my contracted time, minus some of my annual leave that I never end up taking anyway).

Obviously, this calculation is heavily dependent upon the number of samples run at a time, so delaying the genotyping to do bigger batches would save on time/costs (see graph below for estimated drop in cost per sample with increased run sizes). But, it is possible that we would end up paying more for the mouse costs, so that needs to be kept in mind.

Now, comparing to the genotyping service, they charge $7.85 per sample (at current exchange approx. £5.77). This costs about double my calculation for doing it myself, and correlating on the cost/sample graph shows that this lines up slightly below 15 samples, so unless the genotyping batches decrease in size, I think it is unlikely I will be able to persuade the boss to fork over the extra cash to pay for external genotyping services.

However, there is another factor to take into account, which is the current *situation* the world finds itself in, namely Covid19. We (in the UK) are currently in our third national lockdown due to rampant infections, so it might make sense from a health and safety perspective to spend a bit extra on contracting the genotyping externally, if it would save workers from having to come in to the lab and mix in a possibly Covid-transmissable manner. Just something to think about.