See my Science: Laminin:LARGE glycan structure reveals the molecular basis of muscular dystrophy.

Laminins are interesting proteins, which is just as well, because they are what I am working on at the moment.

Laminins are large heterotrimeric¹ proteins that reside in the extracellular matrix (ECM – the stuff outside cells). If you were to google Laminin, you’d find some stuff about the protein, and a whole mess of stuff about it looking like a crucifix, and how Laminin somehow therefore proves the existence of god and/or intelligent design. It is hopefully clear to you, dear reader, that this is utter nonsense, for 2 reasons:

1. The diagrams of laminin are exactly that – DIAGRAMS. Cartoons drawn by scientists to make describing these large complex molecules easier. Cartoons of Bugs Bunny are not proof of walking, talking, wise-cracking rabbits;
2. The pictures (mostly rotary shadowed electron micrographs) of intact laminin are 2 dimensional representations of flexible 3-dimensional molecules, (see fig 1). They don’t look like crucifixes, because they have 3 dimensions – they probably look more like palm trees.

Fig 1. A Laminin electron micrograph.

This is first protein I’ve worked on that has its own page at Snopes.com, which is quite an achievement.
Right, now I’ve got this curious case of molecular pareidolia off my chest, let’s crack on with why Laminin is pretty damn interesting all by itself.

What Laminins do.

Laminin is a principal component in extracellular structures called ‘basement membranes’ (not the same as lipid/plasma membranes, which you might have heard about in high school biology). Basement membranes are important structures that incorporate Laminins, collagens and other ECM proteins to form protein barriers that separate different tissues layers.

Each Laminin molecule is actually 3 different protein chains attached to each other. Because scientists are exceptionally inventive and imaginative, these three proteins are called Laminin alpha, beta and gamma chains respectively.

They fit together a bit like this (fig 2):

Fig 2: Schematic of a Laminin molecule.
Taken from Hohenester & Yurchenco, 2013 http://www.ncbi.nlm.nih.gov/pubmed/23076216

One end of each of the three proteins (the palm tree fronds / the animo-terminus) has an LN-domain (marked LN), and this is involved in forming the basement membrane itself. An LN domain from 3 different Laminin molecules come together to form a sheet like network as shown in fig 3.

Fig 3: How Laminins form a sheet like structure… Taken from Hohenester & Yurchenco, 2013 http://www.ncbi.nlm.nih.gov/pubmed/23076216

This sheet needs to be anchored to the surface of cells. Happily, at the other end of the Laminin, (the base of the palm tree – marked LG1-3/LG4-5 in Fig 2, and marked with a double arrow in Fig 1) there are two functional units that anchor to the surface of cells in two different ways:

1. By binding to proteins on the surface of the cell called integrins. This happens through the LG1-3 region.
2. By binding to a specific sugar which is attached to a protein called Dystroglycan. This happens through the LG4-5 region.
   THIS IS WHAT THIS BLOG POST IS ALL ABOUT

Dystroglycan

Dystroglycan is an integral membrane protein – it sits within the plasma membrane (not basement membrane) and has bits that protrude into both the extracellular and intracellular spaces. It links the basement membrane (outside the cell) to the cytoskeleton (a protein framework inside the cell) – see Fig 4.

Fig 4: Dystrophin complex links intracellular (bott0m) and extracellular (top) structures. From Baresi & Campbell. http://jcs.biologists.org/content/119/2/199

 

The extracellular side of Dystroglycan has a unique sugar polymer covalently attached to it. This sugar (called the ‘LARGE glycan’ or ‘Matriglycan’) has so far only been found attached to Dystroglycan until last week had only been detected on one protein – alpha-dystrogylcan. A recent paper by our collaborators – as demonstrated that this sugar can also be found on other proteins, attached via the same sites known to be occupied by other sugar modifications of proteins (such as the glycosaminoglycans, Chrondroitin Sulphate and Heparan Sulphate). The synthesis of the LARGE glycan is also complex, and the whole linkage region has only recently been mapped. Figure 5 shows a summary of what we know so far about the structure of the LARGE glycan and lists the enzymes that are involved in synthesising it.

Fig 5: How to make LARGE glycan. By the author, information from various papers.

The business end of the LARGE glycan is the -[Glucuronic acid – β1,3 – Xylose α1,3 ] disaccharide polymer – the repeating orange stars and blue/white diamonds on the right of Fig 5. So what the hell is that then?

It is a long chain of alternating Glucuronic acid (a sugar acid similar to Glucose) and Xylose (a sugar that was first isolated from wood, hence Xylose, derived from the greek for wood) units (see Fig 6). The β1,3 & α1,3 refer to how they are connected.  We don’t know quite how long it is yet, or how many modifications there are on each dystroglycan molecule (either 1 or 2). More research required!

Fig 6: The LARGE glycan repeating unit. By the author, created using chemdraw 14.0

So why do we care about this?

In a nutshell, muscular dystrophy.

If you cannot make:

• Laminins
• Dystroglycan
• Any of the enzymes involved in LARGE glycan synthesis

you get muscular dystrophy. Loss of the genes that encode LARGE glycan synthesis (i.e. the genes shown in Fig 5) lead to a subset of extremely severe muscular dystrophies called “dystroglycanopathies“. They are congenital (inherited) diseases that prevent correct formation of basement membranes. This means that tissues do not form properly, and gives rise to developmental disorders. If we can understand this system, maybe we can fix it?

Our recent work

So, we know what the LARGE glycan is made up of, and we know that it binds to the Laminin G-type (LG) domain 4 of Laminin alpha 2.

What we want to know is how they interact with each other at an atomic level – happily the best tool for this sort of study is X-ray crystallography, which is what I do for a living!

So I expressed, purified² and crystallised (Fig 7) a fragment of Laminin alpha 2 which contained the LG4 and LG5 domains. I then soaked a tiny (and I mean tiny) amount of the LARGE glycan that our collaborators had synthesised for us into my crystals. Given that when I did this, we had no idea of tightly Laminin alpha 2 bound to the LARGE glycan, this was a bit of a long shot. I setup around 20 different experiments, with different soaking times and concentrations to try and get some ligand into the protein.

Fig 7: Crystals!

 

I cryocooled the crystals in liquid nitrogen and sent them off to Diamond Light source, where we collect our diffraction data. My remote access data collection shift started at 4am. Oh joy. An early start, plenty of coffee and a taxi to work and I was good to go. However – a drawback of the ligand soaking approach I had to take was that I would have no idea whether or not my crystals had the ligand in until I had processed all the data. I didn’t have the time to process the data as I went along, so the data collection was done using ‘the American method’ – Shoot first, ask questions later.

After my 5-hour shift was up, I grabbed breakfast and more coffee and started to process all my data.

About 5 datasets into my haul of data, I hit the jackpot.

Fig 8a : What the Calcium binding site in LG4 looks like.

 

Fig 8b: Difference density map of SOMETHING.

Above the calcium binding site in LG4, right where it was predicted to be, was a big green blob in my electron density difference map. If you look at Fig 8a, you can see what the empty binding site looks like – this is a very much simplified representation as there is an awful lot going on. The cyan sphere is a calcium ion. The red/green/blue sticks around the calcium show the orientation of the amino acids in the protein that bind to the calcium. The loops and whirls represent other bits of the protein.

I generated an electron density map that highlights the differences between an empty crystal and a soaked crystals (fig 8b). To generate this map, I simply subtract data from an empty crystal (apo form) from data of a soaked crystals (bound form) – the difference between the two datasets should equal the ligand. Happily, it does! After a bit of building and refining, we end up with the final refined structure of the Laminin:LARGE glycan complex.

Fig 9: THAR SHE BLOWS!

Figure 9 shows this final structure of the ligand with the same density shown in Fig 8, just made transparent.

 

Fig 10: A schematic of important interactions in-between Laminin (red) and the LARGE glycan (black).

When we dive into the binding site and take a look around (Fig 10), we see that all the interactions are between a single glucuronic acid – xylose repeat, even though there are 3 repeats in the sample of LARGE glycan we used. The link between the two sugar rings straddles the calcium ion, forming a sort of chelating or clathrate-type interaction. Oxygen atoms (carrying a net negative charge) interact strongly with the positively charged calcium ion . The carboxylic acid group in the glucuronic acid ring (The O=C=O bit) also pokes into a little positively charged pocket formed by two backbone amino groups (NH). All these interactions combine to form quite a strong interaction for (~0.2µM KD, if anyone is interested) for an interaction with a relatively small interaction surface.

So what have we learned?

We now know exactly how Laminins bind to the LARGE glycan. Let’s be clear – this is not a drug target – we definitely do not want to inhibit this interaction! But this shows us a crucial link in the chain between the cytoskeleton (the internal skeleton of a cell) and the outside world.

We have also seen a really interesting and novel mode of protein-carbohydrate interaction. The Laminin does not recognise individual sugars – it recognises the unique linkage found in the LARGE glycan. Given that there are over 100 proteins in the human genome that contain LG domains, and some of those are also known to bind to the LARGE glycan, our structure provides a paradigm for LG domain – LARGE glycan interactions.

We know that a single disaccharide is sufficient for Laminin to engage (although longer stretches of sugar bind more tightly) and we show that the sugar forms a pseudo-clathrate cage over the calcium in the protein. This is probably why the Laminin-LARGE interaction is ~10x tighter than interactions between proteins and other similar sugars.

Anyway – the paper should be out NOW in Nature Chemical Biology.

A PDF is available here.

The pdb files are here

The author’s right forearm.

 

 

 

---

1. Hetero – different, trimer – three. Made up of 3 different proteins.
2. Made in mammalian cells – we give mammalian cells DNA that encodes our protein.

Multi-contoured electron Density maps

When wandering around the department, I am struck by how few crystallographers use multiple contoured electron density maps* whilst building. I really don’t understand this: YOU’RE THROWING AWAY INFORMATION PEOPLE!

Even at moderate resolution, the information gained can be invaluable:

1. Precisely locating a heavy atom in a big blob of electron density. The Atom will sit at the highest point of the map. IF you are using multiple contours, this will be obvious! (figure 1)
2. Resolving His/Asn/Gln sidechain flips. A bit more prone to noise here – but in terms of electrons, O > N > C. You can easily decide which way around Asn and Gln side chains should point, and often get some help with His side chains as well. (figure 2)

Figure 1a: THAR SHE BLOWS!

This calcium ion sits RIGHT on the peak in the electron density map. No ambiguity where it lies. A single map contoured at 1 sigma is not helpful.

Figure 1b: Nope. Not helpful.

 

Figure 2: Gln sidechain flips, made easy…

The increased electron density  on the right hand side here indicate that this Gln residue is the correct way around. Again, a single map contoured at 1 sigma is no use here.

I realise that there are other ways to achieve what I have described, but when you are building your models, saving time and making things easier is hugely helpful. I find using multi-contouring incredibly helpful.

* multi-contours shown here are made using the “Multi-chicken” command in COOT (extensions>maps>multi-chicken). Multi-chicken creates 10 maps contoured at 1,1.5,2, 2.5, etc sigma. I find the default setting is a tad dark so I use “brighten maps” (extensions>maps>brighten maps) a couple of times to sort that out. I then contour the original map at 0.7sigma (depending upon noise) and make it really deep purple.

All screenshots made with COOT.

 

[PODCAST] – Me on Dessert Lionel Discs

Damn good coffee, and hot! (Photo credit: photojenni)

Self promotion, dead ahead!

---

I have been fortunate enough to have been interviewed by @Astrondrew of  Sound of Science for episode 5 of the Dessert Lionel Discs podcast 😀

In it I talk about how I got into science, who inspired me to do science, and what exactly I do as part of my day job. I also talk Pink Floyd, Crystals, Twin Peaks and damn fine coffee.

Please listen, and I hope you enjoy!

E.coli can make Disulphide bonds if you ask it nicely.

Human protein, expressed in NEB SHuffle cells, correctly folded, crystallised, solved, etc.

---

Electron density around a Disulphide bond – which E.coli cannot make in their cytoplasm normally – unless you tinker with a few genes here and there.

The Disulphide is the bond between the two green sulphur atoms in the centre of the image

EDIT: Click to see the picture more clearly – I don’t know what WP has done to it, but it should look crisper than that.

On protein crystallisation…

Image via Wikipedia

A bit of a departure from the usual quack-bashing…

---

Crystallography is great. I love it. The reason that I love it is that it is to my mind one of the most powerful techniques we have at our disposal for getting mechanistic detail about how life works at a molecular level. Biological crystallography allows us to determine the precise three dimensional structure of biologically relevant macromolecules – like proteins and DNA, but also small molecules such as ligands and metabolites necessary for life to thrive and survive.

As the name suggests, crystallography requires the growth of crystals. I think that one of the more interesting facets of crystallography is the dichotomy between the very rigorous and precise methods of data processing and model refinement, and the act of crystallising a protein, which is often described as ‘black magic’.

Crystallisation is essentially controlled precipitation – the protein comes out of solution in an ordered manner, and forms a crystal lattice. In order to get the protein to fall out of solution in this way, we have to alter the chemical environment of protein by mixing it with different concentrations of buffers, salt and precipitating agents.

The complex nature of both the process of crystallisation and the chemical nature of proteins means that we cannot currently predict the chemical conditions in which protein might crystalize. So, we just try stuff that has worked previously. We take our hard won protein and mix it with random solutions, and let it concentrate slowly using a process called vapour diffusion*

Crystallisation is now routinely carried out using crystallisation robots – liquid handling machines that make the whole process easier, faster and more repeatable. They also can dispense nanolitre volumes of protein, making the meager amounts of  proteins that we have struggled to produce and purify go much further. Initial screening is carried out using screens that can be brought in deep well block format from companies such as Molecular Dimensions and Hampton Research.

Labs will have their preferred range of screens and preferred suppliers. FWIW, my first line of attack is to use the JCSG+, Pact Premier, Morpheus and Clear Strategy I & II screens from molecular dimensions. It is the follow-up optimisation of any crystallization ‘hits’ that I want to discuss here.

In days of old (~10 years ago) crystallisation was setup on a micro/millilitre scale. We would setup 24-well trays with 0.5-1 ml of well solution, and drop sizes of 1-10µl. Now, with the robots, we setup 96 well trays with ~80µl of well solution and drop sizes of 200-600 nanolitres. When optimising crystallisation hits, one might suppose that bigger drops might be more likely to yield bigger crystals – which is probably true, but the crystallisation conditions do not necessarily scale up in a simple fashion – this is due to changes in the crystallisation setup, such as the ratio of well solution volume to total volume within the experiment and the drop surface area to volume ratio .

I will admit that I had previously struggled to make the transition from robot-setup nanolitre-scale screens to hand made microlitre-scale optimisation screens. However, I now would argue that hand-setup drops are no-longer required in routine cases of structure solution. Cases where ligand or heavy atoms soaks are required may still need 24-well plate style setups and whathaveyou.

Rather than screen around potential crystallisation hits, I now setup bespoke deep well blocks to screen around them as follows.

I setup 16 15ml falcon tubes. A-H (low) and A-H (high).  Each set of falcon tubes will contain crystallisation conditions with different extremes of one variable, for a given condition.

Let us suppose that I get a hit in 100mM MES pH 6, 20% PEG 3350, 0.2M CaCl2. The first set of variables that we might want to screen are PEG concentration, pH and salt concentration. I would setup 3 sets of falcon tubes as follows.

A (low) 100mM MES pH 6, 10% PEG 3350, 0.2M CaCl2

A (high) 100mM MES pH 6, 30% PEG 3350, 0.2M CaCl2

B (low) 100mM MES pH 5.5, 20% PEG3350, 0.2M CaCl2

B (high) 100mM MES pH 6.5, 20% PEG3350, 0.2M CaCl2

C (low) 100mM MES pH 6, 20% PEG3350,

C (high) 100mM MES pH 6, 20% PEG3350, 0.4M CaCl2

I would then setup a gradient within rows running from positions 1-11 as follows:

1. 1ml of ‘low’
2. 0.9ml of ‘low’, 0.1ml of ‘high’
3. 0.8ml of ‘low’, 0.2ml of ‘high’
4. and so on up to 11, which contains 1ml of ‘high’

This is a relatively straight forward setup for fine screening of crystallisation hits for those of us without really big liquid handling robots that make up your own screens from stock solutions**. You can also incorporate increasing amounts of additives such as cryoprotective agents or perhaps ligands into screens as well, although I tend to use column 12 for screening a couple of low concentrations of cryoprotective additives – e.g. 1% and 5% (v/v) of Ethylene Glycol, Glycerol, MPD and PEG400 – in the base conditions.

I find that this setup is extremely useful as the deep well block can be stored in the fridge for several months (if properly sealed) and used to screen successive batches of material or perhaps to screen point mutants, (that may or may not crystalize in the same conditions, depending upon how lucky you are 😉 ). You get the benefit of bespoke optimisation screens with fine gradients AND the reproducibility, ease and low protein use of a robot. This obviously works best for a system like the Art Robbins Pheonix robot (my weapon of choice) where the crystallisation screen is dispensed into the crystallisation tray by the robot.

Another benefit of this system is the ease at which one can obtain rudimentary phase diagrams, if you use a crystallisation plate with 3 drops per well (I use Art Robbins 3-well intelliplates) and set up screens with 3 different protein concentrations, you will generate data which allows you to create such a phase diagram.

More often than not, this optimisation can yield crystals of the sizes required for routine synchrotron data collection (i.e. >100µm along at least one edge), thus rendering the stock of 24-well plate gathering dust in the corner of the lab pretty much obsolete. Clearly, there will always be occasions when a larger crystallisation drop is necessary, but on the whole, I believe that simple fine screening done like this is sufficient for most structure determination efforts.

*other crystallisation methods are available.

** pH might not necessarily scale linearly with different buffer mixes, especially if buffer types change – but this can be checked with a pH meter if a particularly successful mix of buffers is found

---

This is the first time I have posted about work and technical aspects of it on my blog – I didn’t think that this method would merit a write up as a technical note in a journal, and as far as I know, most labs might be doing this already. However, I have found this technique to be very successful and easy, and so thought it might be beneficial for some to post it online. If you either already use this setup, or you have used this setup after reading it here and it was successful or not, please post below and give me feedback!