The history of venoms is a wonderful example of an evolutionary process. We're all familiar with the idea of venomous snakes, but the cool thing is that when we examine exactly what it is they're injecting into their prey, it's a collection of proteins that show a nested hierarchy of descent. Ancient reptiles had a small and nasty set of poisons they would use, and to improve their efficacy, more and more have been added to the cocktail; so some lizards produce venomous proteins, while the really dangerous members of the Serpentes produce those same proteins, plus a large array of others.
So something like CRISP (Cystein RIch Secretory Protein) is common to all, but only the most refined predators add PLA2 (Phosopholipase A2) to the mix.
Now lethally poisonous snakes are nice and cute and all, but we all know where the interesting action really is: cephalopods. Let's leave the vertebrates altogether and look at a venomous protostome clade to see what they do.
Relative glandular arrangements of a cuttlefish and b octopus. Posterior gland is shown in green; anterior, in blue. Orange structure is the beak.
Brian Fry, who did all that excellent work characterizing and cataloging the pharmacy of venoms secreted by poisonous snakes, has also turned his hand to the cephalopods. He examined the products of the venom glands of octopus, squid, and cuttlefish, and found a range of proteins, some unique, and others familiar: CAP (a CRISP protein), chitinase, peptidase S1, PLA2 and others. There are a couple of interesting lessons in that list.
First, evolution doesn't just invent something brand new on the spot to fill a function — what we find instead is that existing proteins are repurposed to do a job. This is how evolution generally operates, taking what already exists and tinkering and reshaping it to better fulfill a useful function. Phospholipase A2, for instance, is a perfectly harmless and extremely useful non-venomous protein in many organisms — we non-toxic humans also make it. We use it as a regulatory signal to control the inflammation response to infection and injury — in moderation, it's a good thing. What venomous animals can do, though, is inject us with an overdose of this regulator to send our local repair and recovery systems berserk, producing swelling that can incapacitate a tissue. Similarly, a peptidase is a useful enzyme for breaking down proteins in the digestive system…but a poisonous snake or cephalopod biting your hand can squirt it into the tissues, and now it's being used to digest your muscles and connective tissue. Some effective venoms are simply common proteins used inappropriately (from the perspective of the target).
Another interesting observation is that cephalopods and vertebrates have independently converged in using some of the same venoms. In part, this is a consequence of historical availability — all animals have phospholipases,, since they are important general signalling molecules, so it's part of the collection of widgets in the metazoan toolbox from which evolution can draw. It's also part of an inflammation pathway that can be exploited by predators, in the same way that we have shared proteins used in the operation of the nervous system that can be targeted by neurotoxins. So there is independent convergence on a specific use of these proteins as toxins, but one of the things that facilitates the convergence is a shared ancestry.
In fact, some very diverse groups seem to consistently settle on the same likely suspects in their venoms.
But finally, there must also be physical and chemical proteins of these particular proteins that must also predispose them to use as toxins. After all, animals aren't coopting just any protein for venoms — they aren't injecting large quantities of tubulin or heat shock proteins into their prey. There must be something about each of the standard suspects in venoms that make them particularly dangerous. What the comparative evolutionary approach allows us to do is identify the common molecular properties that make for a good venom. As Fry explains it,
Typically the proteins chosen are from widely dispersed multigene secretory protein families with extensive cysteine cross-linking. These proteins are collectively much more numerous than globular enzymes, transmembrane proteins, or intracellular protein. Although the relative abundance of these protein types in animal venoms may reflect stochastic recruitment processes, there has not been a single reported case of a signal peptide added onto a transmembrane or intracellular protein or a hybrid protein expressed in a venom gland. A strong bias is also evident for all of the protein-scaffold types, whether from peptides or enzymes. Although the protein scaffolds present in venoms represent functionally and structurally versatile kinds, they share an underlying biochemistry that would produce toxic effects when delivered as an "overdose". Toxic effects include taking advantage of a universally present substrate to cause physical damage or causing changes in physiological chemistry though agonistic or antagonistic targeting. This allows the new venom gland protein to have an immediate effect based on overexpression of the original bioactivity. Furthermore, the features of widely dispersed body proteins, particularly the presence of a molecular scaffold amenable to functional diversification, are features that make a protein suitable for accelerated gene duplication and diversification in the venom gland.
To simplify, killing something with a secreted poison typically involves reusing an extant protein, but not just any protein — only a subset of the proteins in an animal's proteome has just the right properties to make for a good venom. Therefore, we see the same small set of proteins get independently coopted into the venom glands of various creatures.
Fry BG, Roelants K, Norman JA (2009) Tentacles of venom: toxic protein convergence in the Kingdom Animalia. J Mol Evol Mar 18. [Epub ahead of print].