Eight Thousand Kilometres on a Raft of Weeds: How Iguanas Reached Fiji

A Lizard Where No Lizard Ought to Be

Go and look at a Fiji banded iguana, which runs about the length of your forearm, brilliant green, banded in pale stripes that come and go with its mood, watching you from a branch with a steady black eye. The animal looks Caribbean. It sits in the South Pacific instead, two and a half thousand kilometres from Australia and nine thousand from the nearest wild iguana in the Americas, with no business being there at all.

Iguanas belong to the New World, and almost the whole family sits between the southern United States and Paraguay, with outriggers on the Caribbean islands and the Galápagos, a distribution that has been stable and boring for as long as anyone has had a map. Then Fiji breaks it: four living species of Brachylophus, all of them rare, one of them down to a few hundred animals on a single dry island [6], plus a fifth, a giant called Brachylophus gibbonsi, which lived in Tonga and was eaten out of existence after people arrived [7]. Nothing else in the entire Pacific basin has iguanas. Not New Guinea. Not New Caledonia. Not Vanuatu, Samoa, the Marquesas, or the Solomons [5].

So how did they get there?

Two answers were acceptable for most of the twentieth century, and the shared merit of both, though nobody said so, was that they let you avoid the third. The first was vicariance, which says the animals never travelled at all and the land did: iguanas were spread across the ancient southern supercontinent of Gondwana, Gondwana broke up, and the Fijian population simply rode its fragment out into the Pacific. The second was island hopping, a chain of short crossings from South America through a Pacific that had more islands in it then than it does now, and neither story requires you to picture a lizard floating on a log for a year.

The third answer requires exactly that, and it carries a name George Gaylord Simpson gave it: sweepstakes dispersal, which is the bleakest phrase in biogeography. You buy a ticket you will almost certainly lose. The prize is a continent.

The Scarpetta paper is an argument that Fiji's iguanas hold a winning ticket, and that the draw was held somewhere off the coast of what is now California, about 34 million years ago.

What Four Thousand Loci Said

The result rests on a phylogeny, so a reader deserves to know what a phylogeny is and where this one could go wrong.

A phylogeny is a family tree for species, built by comparing DNA sequences on the principle that lineages which split recently have had less time to accumulate differences and so still look alike. Any single gene can lie to you. Genes keep their own histories, and a gene can pass between lineages, or be caught on the wrong side of an ancestral polymorphism, or be dragged into shape by selection that has nothing to do with the species carrying it, which makes a tree built from one gene a tree of that gene rather than of the organism. The defence is volume.

Sequence hundreds or thousands of independent stretches of the genome, build a tree from each, and see what the majority says, which is what Scarpetta and colleagues did across more than 4,000 loci and over two million base pairs [1]. They used four classes of marker, chosen because they fail in different ways: rapidly evolving long exons, which carry a lot of signal over short timescales; ultraconserved elements, stretches of the genome so unchanged across vertebrates that their flanking regions make reliable anchors; anchored hybrid enrichment loci; and a large set of exons from the X chromosome. Then they analysed the whole pile twice, once by concatenation, which glues everything into one enormous alignment, and once under the multispecies coalescent, which builds a tree per locus and reconciles them while explicitly allowing gene trees to disagree with the species tree.

Both methods gave the same answer with strong support: Brachylophus, the Fijian genus, is sister to Dipsosaurus, the desert iguana of the Mojave and the Sonoran. Not sister to anything Caribbean. Not sister to anything South American. Sister to a lizard that lives among creosote bushes in Death Valley.

The dates falling out of the analysis put the split near or under 34 million years ago, with credible intervals reaching back towards 40 and in some runs 42 [1], which is late Paleogene, roughly the moment the Earth began tipping into its long cold phase.

Then comes the biogeography, which is where a tree stops being a diagram and starts being a claim about places: given a tree, a set of dates, and the present-day ranges of everything on it, a program called BioGeoBEARS asks which sequence of range changes best explains what you see. Among the models it compares, selection was not close: the version permitting founder-event speciation, where a few individuals jump to a new area and become a new lineage, beat every alternative with an Akaike weight of 1.00 [1, 8]. North America came out as the most probable ancestral range for the Brachylophus and Dipsosaurus pair.

Crossing Off the Comfortable Answers

Take the Gondwana story first, because most people were taught it and it has the comfortable shape of an explanation that requires nobody to get wet.

It fails on geology. Fiji is not an old continental fragment carrying passengers away from a broken supercontinent but an oceanic island arc, built by volcanism above a subduction zone, with the islands themselves late Paleogene at the oldest [18]. No Fijian land surface existed for a Cretaceous iguana to be marooned on, and the rocks are simply younger than the lineage would need them to be. No amount of argument fixes that.

Take the southern route next, iguanas moving through South America, then Antarctica, then across to the Pacific islands, a route that dies on climate: from about 34 million years ago the Antarctic ice sheets begin building in earnest, and the window during which a large-bodied ectothermic herbivore could have walked across Antarctica closes exactly when you need it open [1].

The Eurasian route, westward across Asia, runs into cooling and drying along the whole way, with no plausible corridor for a heat-loving lizard [1], and nobody has ever pressed it very hard.

That leaves the sea.

Overwater dispersal stopped being disreputable some time ago, though through the middle of the twentieth century, suggesting that an animal had floated somewhere marked you out as unserious, because plate tectonics had just arrived and vicariance explained everything so beautifully. Alan de Queiroz named the reversal "the resurrection of oceanic dispersal", and molecular dating drove it: again and again, lineages turned out far younger than the continental splits supposed to have separated them [3, 4]. Madagascar's mammals are the standard example, and modelled Eocene ocean currents can carry a raft from Africa to Madagascar in a few weeks [19], which is an unremarkable length of time to be adrift.

Direct evidence exists for the short hauls, and it is the kind a court would accept. In October 1995, after two hurricanes crossed the Lesser Antilles, at least fifteen green iguanas came ashore on Anguilla riding a mat of logs and uprooted trees, having crossed about 300 km from Guadeloupe; some were still alive a month later, and the population established itself [2].

Three hundred kilometres is not eight thousand. But the mechanism is real, and the animals doing it are iguanas.

The Arithmetic of the Crossing

From here on, the numbers are ours, since the paper reports a crossing without ever costing one out. We did, because a claim about an 8,000 km drift is a claim about evaporation and metabolism and the calorific value of leaves, all of which can be checked by anybody with a calculator and a free afternoon.

Start with distance, since everything else in this section is a rate applied to it. The paper's figure is more than 8,000 km measured as a great circle, and a raft does not travel on a great circle: it meanders, sits in eddies, gets pushed north and south by the wind, and now and then goes backwards for a week. We apply a path sinuosity factor of 1.35. Call the water covered 10,800 km.

Now speed. The westward conveyors across the tropical Pacific are the North Equatorial Current and the South Equatorial Current, both driven by the trade winds and both flowing towards Asia and Australia [14, 15]. Surface speeds vary with season and with latitude and shift hard during an El Niño, so we take three cases: 0.15 m/s for a weak northern limb, 0.35 m/s for a typical speed, and 0.70 m/s for the core of the South Equatorial Current with a windage credit for a raft carrying a canopy that acts as a sail.

Divide.

Those speeds come to 13.0, 30.2 and 60.5 kilometres a day, which is roughly the pace of a slow walk sustained without pause for months. The fastest case is just under six months. The median case is just under a year. The slow case is two years and three months.

Figure 1 puts those three bars against the two deadlines we are about to derive, and the whole shape of the problem sits in that picture, so look at it before reading another word.

We looked for a second, independent way to sanity-check these numbers and did not fully find one. Modelled drift trajectories for sweet potato seeds crossing the tropical Pacific give transit times of 80 to 120 days for comparable distances in the fastest water [16], and drift pumice from volcanic eruptions has been tracked across the Pacific at broadly similar rates [17]. Our fast case sits at 179 days, roughly one and a half times the top of that range, not inside it. A tuber and a raft of vegetation are not the same object on the same route, so this comparison tells us the two estimates sit in the same rough neighbourhood rather than confirming ours. It is a plausibility check, not a validation, and we would rather say so than round the gap away.

Fuel

A 70 g lizard has a fuel tank, and we would like to know how far that tank carries it across open water.

Metabolic rate in reptiles scales with body mass as a power law, an observation that holds across squamates over four orders of magnitude of mass [11], and for reptiles as a group the field metabolic rate is close to \(0.196\,M^{0.889}\) kilojoules per day, with mass in grams [12], which at 70 g comes to 8.56 kJ per day. An animal on a raft is not defending territory or chasing food, so we scale that to 40%, a resting rate, and then correct for temperature, because an ectotherm's metabolism doubles and a bit for every ten degrees: with a \(Q_{10}\) of 2.5 and a body temperature of 28.5 °C against a 30 °C reference, the correction is 0.8716.

Now the tank itself: lizards store fat in abdominal fat bodies and in the tail, and a well-fed animal going into a dormant season can carry around a tenth of its mass as fat. Assume 10% of body mass, of which 80% is actually mobilisable before the animal dies of something else, at 39.3 kJ per gram, and the store comes to 220.1 kJ.

Seventy-three days, against a voyage of 179 days in the best case. The animal starves with 105 days still to run, and on the median current it starves with 283 days still to run, and on the slow current it dies before it has covered a tenth of the distance.

Which would be the end of the story. The raft, though, is made of salad.

Iguanas are herbivores, so a raft of uprooted vegetation is less transport for an iguana than a floating meal that happens to be going the right way, which is the single most important biological fact in the whole argument and the easiest one to walk past. To cover the median crossing our animal needs 1,066 kJ, and at a gross energy of about 7 kJ per gram of dry foliage with half of that assimilable, it needs 305 grams of dry leaf spread across the eleven months at sea. Four times its own body mass, spread across a year. A single uprooted tree carries kilograms.

A carnivore on the same raft has nothing, an asymmetry that probably explains why the long-distance records in this business keep going to herbivorous reptiles rather than to anything that has to hunt.

A catch turned up, and we found it by asking the wrong question first: mass of foliage is not the constraint; freshness is. So we modelled the raft as supplying a fraction \(\phi(t) = e^{-t/\tau}\) of the daily requirement, with the leaves dying and the food getting poorer as the voyage goes on, and solved for the value of \(\tau\) at which the fat store runs out exactly at landfall.

On the fast current, \(\tau\) has to be at least 151 days. On the median current, 741 days. On the slow current, 4,427 days, which describes not a raft any more but a small floating island carrying a working ecosystem, and that is the most awkward number in our whole model, which we will come back to when we make the case against ourselves.

Thirst

Water is worse, and worse by a wide margin.

Surrounded by it, and none to drink. A lizard cannot process seawater; iguanids have nasal salt glands that let them shed excess sodium and potassium in a concentrated crust, which is why marine iguanas sneeze, but those glands manage a salt load rather than manufacture fresh water [13]. Our animal loses water continuously and has almost no way to get any back.

So we built the loss from physics instead of guessing at it, counting the four routes by which water leaves a lizard, one of which runs backwards.

Through the skin. Reptile skin is not waterproof, merely very resistant, so treat it as a diffusion barrier with a resistance \(r\) in seconds per centimetre, across which the flux is the vapour density difference across it divided by that resistance, times the surface area. At 27 °C air, 80% humidity, and skin at 28.5 °C, saturated air at the skin holds 27.91 g of water per cubic metre and the marine air holds 20.55, so the deficit driving the loss is 7.35 g/m³. Surface area from the standard lizard allometry \(A = 10M^{2/3}\) is 169.8 cm², and with \(r = 300\) s/cm, the leaky end of what has been measured for desert lizards, the skin loses 0.3596 g/day.

Through the lungs. Exhaled air comes out saturated at body temperature, so tie ventilation to oxygen consumption, assume 15% extraction from each breath, and the respiratory loss is 0.0354 g/day, a trivial amount, because reptiles breathe slowly and shallowly and that thrift is one of the reasons they survive deserts.

Through urine and faeces. Reptiles excrete nitrogen as uric acid, a paste rather than a solution, which is spectacularly water-thrifty compared to the urea route a mammal uses [13], and charging 0.045 g of water per kilojoule metabolised gives 0.1343 g/day.

And back the other way, burning fat produces water: every gram of fat oxidised yields about 1.07 g of metabolic water, so the animal gets 0.0813 g/day back for free.

Total body water in a lizard runs around 70% of mass, so 49.00 g here, and a xeric lizard can lose something like a quarter of that before the circulation gives out, which leaves twelve and a quarter grams of margin.

Divide and you get 27.3 days.

Twenty-seven days. The fastest crossing is 179. Without rain the voyage is not difficult but flatly impossible, at every current speed we modelled, by a factor of six or more, and no amount of care with the other parameters rescues it.

Figure 2 draws both clocks against one median crossing, along with what happens when the sky is allowed to open, which is the only version of this voyage anybody survives.

Two details are worth sitting with, and the first is that skin carries 68% of the gross loss, which rests the whole question of survival on a single hard-to-measure number, the skin resistance, for which we chose a pessimistic value at the leaky end of the measured range rather than a flattering one. The second is that warmth kills: push the air from 27 °C to 36 °C and the dry window falls from 27.3 days to 14.9, and it falls because the vapour deficit widens, rather than because the metabolism speeds up. A raft in the doldrums under heavy cloud is a far better vehicle than the same raft under a clear-sky high, which inverts the intuition that a voyage wants good weather.

Bench Notes: The Afternoon We Argued About Rain

Working notes, transcribed, lightly tidied.

First version of the water model had the animal dying on day 31, and we all agreed that killed the paper, until somebody pointed out we had not put any rain in it. Long pause. Then a slightly embarrassed rush to the whiteboard.

The route goes through the Intertropical Convergence Zone, the band where the trade winds of the two hemispheres collide and shove air upwards, which makes it the rainiest belt on the planet, with the South Pacific Convergence Zone hanging off it diagonally towards Fiji. We had spent two hours modelling a crossing of a desert when the actual route crosses something closer to a permanent rainstorm, which nobody enjoyed admitting out loud.

So: rain as a daily coin flip, probability \(p\), where a wet day lets the animal drink from foliage and refill up to 12% of its body mass, a dry day costs it 0.4480 g, and death arrives when the accumulated deficit hits 12.25 g. Run the whole thing twenty thousand times per cell.

Argument one, and it took a while. Is a Bernoulli draw honest? No. Real tropical rain is clustered, with wet spells and dry spells, so the true distribution of dry runs has a fatter tail than ours does, and our model therefore understates the chance of a fatal dry spell. We have left it in, with the flag on it, because the alternative is inventing a persistence parameter we have no way to calibrate for the Oligocene.

Argument two. Can a lizard actually drink off wet leaves? Yes, and this is one of the few places where the literature is unambiguous, because desert lizards drink from rain-wetted surfaces routinely, and some of them have skin architecture that channels water towards the mouth; the desert iguana in particular lives in country with no permanent fresh water and gets almost everything it needs from its food and from the occasional soaking [9, 10].

Argument three, which nobody won. Somebody asked about sea spray: how much of the fresh water sitting on a raft's surface goes salty, and how fast, and whether a heavy downpour rinses the leaves clean before the animal gets to them. We had no number. We still have no number, and it goes into the limits section as L3, an honest hole in the middle of the water budget.

What came out the other side is the table below, and its shape surprised us, because we had expected current speed to dominate on the obvious grounds that a faster crossing is a shorter exposure. It does matter, but rainfall matters far more. Going from a 10% chance of rain to a 20% chance moves the slow-current survival from 0.0034 to 0.7170, a factor of two hundred, while going from the slow current to the fast current at fixed 10% rain moves it from 0.0034 to 0.3411, a factor of a hundred, so the two levers are comparable, and the difference between some rain and none dominates both of them.

The Survivable Window

Twenty thousand simulated voyages per cell, seeded, raft treated as edible throughout, and the bottom block gives the rainfall frequency at which survival first reaches one half.

Between 44 and 66 rainy days a year, and that is the whole requirement, the entire price of the voyage stated in weather. The tropical Pacific along this route delivers rain on a far larger fraction of days than that, year in and year out, and did so in the Oligocene as well, because the convergence zones are a consequence of the planet's rotation and the distribution of heat rather than of any particular climate state.

Figure 3 draws the same grid as filled bars, because a cliff is easier to see than to read off a column of four-decimal numbers.

So the answer to the question we started with is yes, with conditions: the crossing is survivable if the raft stays edible and the sky behaves the way the tropical sky normally behaves, and flatly unsurvivable otherwise.

The Strongest Case Against

Now we should try to knock the whole thing down, and do it properly rather than politely, since a club that only ever agrees with the paper it is reading is not much use to anybody.

The best objection against all of this has nothing at all to do with the genetics. The phylogeny looks solid: four marker classes, two analytical philosophies, strong support, and a sister relationship that is not subtle, so if anybody overturns Brachylophus plus Dipsosaurus we will be very surprised. Our quarrel is with what a sister relationship licenses anybody to say about a journey, which turns out to be a good deal less than the headlines suggested.

Objection one: the tree does not locate the voyage in time. The paper says so itself, in a sentence most of the coverage skipped: the split between the two genera is dated to around 34 million years ago, but the dispersal could have happened anywhere along the long Brachylophus stem, which is the stretch of tree between that split and the first branching within the Fijian group [1]. A lineage could have sat on the American mainland for twenty million years after diverging from Dipsosaurus and only then gone to sea, so everything we have computed concerns a crossing whose date is genuinely uncertain, which matters because the current system and the island geography of the Pacific were not the same at 34 Ma as at 14 Ma.

Objection two: stepping stones we cannot see. The Pacific is littered with the drowned remains of islands, seamounts and guyots being volcanoes that stood above water once and have since subsided beneath it, and the authors acknowledge directly that arrival by island hopping across a chain of now-vanished stepping stones cannot be excluded [1, 18]. If that is what happened, the record-breaking single crossing becomes a series of shorter ones, and our whole physiological argument becomes a calculation about a journey that never took place in one go. Nothing in the genome separates one 8,000 km hop from six of 1,300 km.

Objection three, and this one is ours rather than the paper's: the raft has to last. Our own model says the raft must go on supplying food on a timescale of 151 days in the fast case and 741 days in the median case, and we have no evidence that vegetation rafts do anything of the kind. Observed rafts break up, waterlog and get bored through by shipworm over weeks to months; the Anguilla iguanas crossed in about a month [2], and stretching that to a year is not a small extrapolation. Our model handles raft integrity by not modelling it at all. One guessed probability, in the section after this one, carries the entire question. If rafts reliably fall apart at ninety days, the story requires either a much faster current than any we used or a mechanism we have not thought of.

Objection four: our physiology is a sketch. Skin resistance carries 68% of the water budget and we picked one number for it, the lethal dehydration threshold of 25% is an educated guess spanning a real range, and body mass is a free parameter the Paleogene animal may have been nothing like. We average over day and night and over storms, we do not model salt spray at all, and salt spray is exactly the kind of thing that could turn a drinkable raft into an undrinkable one overnight.

Objection five: the fossil record is silent, and silence cuts both ways. An extinct Fijian iguanid, Lapitiguana, has never had its relationship to Brachylophus tested with modern methods [1], and if it turns out to be a separate arrival, then Fiji was colonised more than once, which would make overwater arrival look easier and would also mean the tree we are reasoning from is missing branches.

Here is the honest summary, arrived at after some argument: descent from a North American ancestor is very well supported, and we would defend it against anybody. The claim that they got there by a single unbroken 8,000 km drift is a reasonable inference that the data underdetermine, and the strongest version of the doubt says nothing about whether a lizard can survive at sea. It says we cannot tell how many legs the journey had.

The Lottery That Always Pays

Grant all of it. Suppose, for the sake of the arithmetic, that the crossing really is close to impossible, and then explain how an almost-impossible event ends up on the map with four living species sitting on it.

Set it up as a rate problem, with two quantities and a division. Let \(\lambda\) be the number of rafts per century that leave western North America carrying an iguana and enter the westward current system, and let \(q\) be the probability that any one of those voyages ends with a breeding population on Fiji. Successful colonisations then arrive as a Poisson process with rate \(\lambda q\) per century, so the expected wait between them is \(1/(\lambda q)\), and nothing in the rest of this section is harder than that.

Our \(q\) is a product of four terms, three of which are guesses rather than calculations, and we would rather say so plainly than bury it in a footnote.

One in a hundred and eighty thousand. Put a number like that in front of anyone and they will tell you it does not happen, and on the scale of one lifetime they are right.

Now divide by deep time, which is the step that turns a number nobody believes into a number nobody can argue with.

At ten rafts per century, over the 34 million years the phylogeny allows, 3,400,000 rafts set out, and multiplying by \(q\) gives an expected 18.85 successful colonisations, which puts the probability of at least one arrival at 1.0000 to four decimal places. Even at one raft per century, an extremely stingy figure for a coastline that gets hurricanes, the expected number is 1.885 and the chance of at least one arrival is 0.8482.

The waiting times tell the same story from the other side, and they are worth reading slowly. At one launch per century you wait 18.03 million years for a success. At ten, 1.80 million years. At fifty, 360,000 years. Absurd waits on any human scale. Unremarkable on a geological one.

Figure 4 is that calculation drawn as a curve, with the pessimistic variants beside it and both axes logarithmic, because nothing in this argument fits on a linear page.

We also asked how fragile the conclusion is, because anything built on three guessed probabilities deserves interrogation: make \(q\) ten times worse and the probability of at least one arrival in 34 Myr is still 0.8482. Make it a hundred times worse, one voyage in eighteen million, and it falls to 0.1718, so the argument survives an order of magnitude of pessimism comfortably and breaks somewhere in the second. Solid, then, and not infinitely so.

A Very Long Time, Seen From a Raft

Here is the part the arithmetic will not tell you, and the part that kept three of us at the table long after the rest had gone home.

Somewhere off the coast of a continent that did not yet have the shape we would recognise, a river came down in flood and took a section of bank with it. Trees, soil, roots, the whole mat. On it, by accident, an animal. Nothing about the journey was chosen: the lizard was asleep in a burrow, or sitting in a branch, and then the ground was moving and then the ground was water.

The current took it west, as the current does. We can put a number on the rest: eleven months, give or take, on our median current, and three hundred and fifty-seven mornings of open water in every direction. Fifty-three days of rain on the crossing our model happened to draw, which is a drink and a wash and nothing else; three hundred and five days without. The longest dry stretch was twenty-three days, and the model says it dies at twenty-seven, which means that at some point in the middle of the Pacific this animal came within four days of the end and had no way of knowing it.

It arrived with just over half its water margin and 14.8% of its fat.

And then, presumably, almost every other one of them died, without issue and without witness. For every raft that reached Fiji, \(q\) says there were something like a hundred and eighty thousand that came apart, or drifted into empty water, or carried an animal that starved, or landed a single male on a beach where he lived out a perfectly successful and entirely pointless life. The lottery is not kind, only patient.

What gets us, sitting around a table arguing about skin resistance, is the mismatch of scales: the voyage is an animal-sized thing, eleven months, one lizard, a specific number of rainy days. The process that made it inevitable is a geology-sized thing, thirty-four million years wide, and it does not care about the individual at all, nor is it required to. Both of those are true at once and neither reduces to the other. The arithmetic in Section 10 says a success was essentially guaranteed, and it says nothing whatsoever about the one that actually happened, which was a particular animal on a particular mat of drowned vegetation, for whom the outcome was in doubt every single morning for a year.

About four thousand Fiji crested iguanas remain, and the number is falling, mostly because of goats, rats, cats and cleared forest [6], with every animal on that list having arrived by boat. Whatever arrived on that beach carried the whole future of the lineage in one body, and it has taken us about two centuries to put that lineage back in serious danger.

Thirty-four million years of holding on, and a hundred and eighty thousand failed crossings to buy the one that worked. We should probably be more careful with it than we are.

Sections 4 through 8 and 10, and Figures 1 through 4, are the club's own simplified model rather than the paper's analysis: Scarpetta and colleagues did the sequencing, the tree, the dating and the biogeographic model selection. We did the arithmetic a reader can check by hand, and we have listed twelve assumptions and six limits in the script so that you can break them.

References