A Blog by

Building anti-flu drugs on a computer

You have a sculpture, an intricate piece of modern art, covered in bulges and blisters. Your task is to weave a cover for it. The fit must be exact. You have to fill in every dent and wrap around every lump. Here’s the catch: you have to make this faultless shroud from a single piece of string that must automatically weave itself into the right three-dimensional shape.

This is the challenge that Sarel Fleishman, Timothy Whitehead and Damian Ekiert from the University of Washington have just overcome. Their “sculpture” is a protein called haemagglutinin, or HA, which sits on the surface of flu viruses. Their “shroud” is another protein designed to perfectly fit onto the contours of HA and neutralise it. They have found a way of fashioning these designer proteins on a computer – a feat that could make it easier to create the next generation of anti-flu drugs.

Under the microscope, flu viruses look like spherical pincushions. The “pins” consist of two proteins: haemagglutinin, which allows the virus to stick to a host cell, and neuraminidase, which allows it to eventually escape. The different versions of these proteins give flu viruses their names. For example, a virus with a haemagglutinin from group 1 and a neuraminidase from group 1 would be called H1N1 – the same type that went pandemic in 2009.

These two proteins are crucial to the virus’s infective abilities, and they are also its Achilles heel. Neutralise them, and you cripple the virus. It can’t infect or spread. These proteins, like all of them, have complex three-dimensional shapes that allow them to interact with their targets. To deactivate them, you need to design designing molecules that perfectly fit into their many nooks and crannies, like sticking gum in a lock.

That is easier said than done. The standard method is to create a large “library” of molecules with different shapes, and identify those with the closest fit. This is hard, but it’s even harder to design the right shape from scratch. Proteins are long chains of amino acids that naturally fold into complicated shapes. Creating a protein is like creating origami that folds itself. Some amino acids attract one another; others repel. To get the right shape, your chain must be just right. A single misplaced amino acid can throw the contours of the entire protein out of kilter.

To create their designer proteins, Fleishman, Whitehead and Ekiert relied upon state-of-the-art software that took around 20 international groups of scientists to create. It soaked up over 100,000 hours of parallel computing time.

Other groups have tried to achieve the same thing, but with far more modest results. They mostly started by taking an entire protein and gradually tweaking its structure to get the best possible shape. Fleishman, Whitehead and Ekert took a different approach. They identified individual amino acids that would strongly interact with HA, and created a scaffold that would join these dots together.

From virtual thin air, Fleishman, Whitehead and Ekert conjured up 73 designs, which they brought to life using yeast. They engineered the fungus to manufacture the designer proteins and shunt them to their surface, where they could be easily tested against HA. In this way, the team could test their designs quickly without having to actually purify the proteins (a time-consuming and technically demanding task in itself).

Two of the 73 designs stuck to HA. One of them fit in a way that almost exactly matched the predictions of the team’s software. By tweaking the amino acids in their designer proteins, Fleishman, Whitehead and Ekert managed to improve the fit even further. This proved the principle – virtual protein design can work, even though it’s still a bit inefficient.

The two proteins were designed to stick to version of HA carried by the 1918 H1N1 flu, the one that killed millions of people around the world. However, the designer proteins also target a part of HA – the ‘stem’ – which is unsually stable. It doesn’t change a lot over time, and It looks very similar from strain to strain. This means that Fleishman, Whitehead and Ekert’s proteins should be able to target all H1 flu (such as the 2009 pandemic strain), if not all flu viruses.

Do the proteins actually stop the virus? Sadly, that’s the only missing part of the puzzle. It’s a reasonable expectation, given that other antibodies that target the HA stem can stop the viruses from entering a host cell.

But that’s a matter for a future study. The point of this one was not to create tomorrow’s flu drugs. It was to show that such drugs could, in principle, be designed from scratch on a computer. This is a fiendishly difficult puzzle, and solving it, even just once, is impressive enough.

Reference: Fleishman, Whitehead, Ekiert, Dreyfus, Corn, Strauch, Wilson & Baker. 2011. Computational Design of Proteins Targeting the Conserved Stem Region of Influenza Hemagglutinin. Science http://dx.doi.org/10.1126/science.1202617

More on protein structures:

8 thoughts on “Building anti-flu drugs on a computer

  1. One need only bind to the small patch of protein responsible for host recognition. A reactive (suicide inhibitor) polysaccharide would do it – and be immensely cheaper (synthesis and equivalent weight) than a long polypeptide chain.

    Management obsesses on what is measurable instead of promoting what is important. Management is rewarded for enforcing process not creating product.

  2. This is interesting given that we don’t have many drugs to treat viruses.

    I also think it’s interesting from the standpoint of being an application of basic research. Basic research is often harder to sell the public on, but this is a case where basic research in protein folding could lead to something that the public would find of great benefit.

  3. @ Seriously!
    Keep in mind this isn’t a treatment. Presumably that’s what you meant when you asked your question. They haven’t even verified that this stops H1 flu yet. This is promising, but it’s not an anti-viral treatment. Yet.
    And if it does work on H1 flu, then they have to find out if they can use the same approach on HIV and a range of other viruses. It’s too early to tell if it will be effective at all, let alone how broadly applicable it will be toward the range of viruses we wish to control.

  4. 100,000 hours of parallel computing time might seem like a lot but if we can get a virus like H1 to implode itself or just smother it, they may be onto something huge, especially if this is a designer/targeted solution.

Leave a Reply

Your email address will not be published. Required fields are marked *