Habitat tracking, stasis and survival in Neogene large mammals

Abstract

Species response to environmental change may vary from adaptation to the new conditions, to dispersal towards territories with better ecological settings (known as habitat tracking), and to extinction. A phylogenetically explicit analysis of habitat tracking in Caenozoic large mammals shows that species moving over longer distances during their existence survived longer. By partitioning the fossil record into equal time intervals, we showed that the longest distance was preferentially covered just before extinction. This supports the idea that habitat tracking is a key reaction to environmental change, and confirms that tracking causally prolongs species survival. Species covering longer distances also have morphologically less variable cheek teeth. Given the tight relationship between cheek teeth form and habitat selection in large mammals, this supports the well-known, yet little tested, idea that habitat tracking bolsters morphological stasis.

1. Introduction

Environmental change modifies habitats thereby driving species into sub-optimal conditions. Species responses may vary from adaptation to the new conditions, to dispersal towards territories with conditions more similar to their original habitat (habitat tracking), and to extinction.

Palaeontological accounts suggest that most species maintain nearly constant environmental tolerances over their existence [1,2]. This is especially true of larger organisms, which might be unable to evolve quickly enough to keep up with changing environments because of their long generation time [3]. Habitat tracking and extinction are therefore two very important reactions to extensive environmental change for large-sized species. Habitat tracking has been extensively reported in Caenozoic mammals facing large-scale, repeated events of global cooling [3,4]. It is famously apparent in that Late-Pleistocene, mammals moved from the North to the South and the other way around in the wake of glaciations [5].

Species vulnerability to extinction depends, to a large extent, on their biological and ecological traits, most notably body size [6], geographical range size [7] and ecological specialization [8]. Habitat tracking and species survival intermingle. In the face of environmental change, species able to disperse over long distances to track their preferred habitats may be more likely to survive than those that can only disperse over short distances [9,10].

Despite the likely importance of this possible correlation between survival and habitat tracking, this relationship has been little explored in fossil species. Here, we provide a palaeobiological investigation of the effect of habitat tracking on species duration and their tendency towards morphological stasis. We used the portion of the range where species are most common over successive time intervals to test whether species moving over longer distances survived for longer. We then tested for a correlation between the distance moved and morphological variation within clades. We predict a negative correlation if species track optimal ecological settings during their existence.

2. Material and methods

To take into account possible phylogenetic effects, we prepared a phylogenetic tree including 72 extinct ungulates, carnivores and proboscideans. The details of the phylogenetic tree are available in the electronic supplementary material. Species were selected because of their good and time-continuous fossil record. We compiled a database of fossil occurrences of these mammals from the Paleodb (http://www.paleodb.org/) and NOW (http://www.helsinki.fi/science/now/) databases. We divided the record into 1 million year (Myr)-long time bins according to the age estimates of the fossil localities. In order to calculate the distances covered by a species during its existence, we identified the position of the weighted centre of its distribution at successive time intervals. This weighted centre (central feature, CF) identifies the fossil locality minimizing the summed distance to all other localities where the species occurred in any given time bin. The distance between successive CFs would represent the distance covered by the species over successive time bins. Yet, it is not a good representation of their actual displacement, because the geographical distribution of fossil localities across successive time bins is uneven. To account for this, we computed the geometric centre (GC) of the whole fossil record for each time bin. The distance between GCs represents the displacement of the record owing to the difference in spatial sampling between successive time intervals. This should be subtracted from the CF–CF distance. To achieve this aim, we translated the CF–GC vectors of all time bins to a single GC (figure 1). This is geometrically equivalent to placing all CFs in a single-reference system.

Figure 1.

The procedure for calculating species movement through time involves four steps. Step 1: the central features (CFs) of the species fossil sites distribution in all time bins are computed. Step 2: the geometric centres (GCs) of all the fossil sites (either including or not focal species) are computed for each time bin when the species lived. Step 3: GC–CF vectors of all time bins are translated to a single, reference GC. Step 4: the geodesic distance between the translated CFs can now be computed.

The species stratigraphic duration is the difference between the species’ first and last occurrence in the record (electronic supplementary material, table S1). To test for possible correlation between duration and body size [11], we took extinct species' body sizes either from the literature or estimated it by allometric equations (electronic supplementary material, table S2).

We regressed the total distance covered by each species during its existence on its duration, both using the raw data and in a phylogenetically informed context. Phylogenetic regression was performed twice by using phylogenetic generalized least squares (PGLS), first under the Brownian motion model of evolution, and then correcting branch lengths by using Pagel's λ transform (see the electronic supplementary material for details). We used PGLS regressions because stratigraphic duration is phylogenetically conserved [11]. Hence, a significant relationship between distance and duration might be an artefact of shared ancestry.

We used a binomial test to determine if the maximum distance covered by a species falls in the last time bin more often than expected by chance (see the electronic supplementary material for details). The rationale is that if habitat tracking really prolongs survival, species may have actively tried to cover larger distances under the worst conditions, which presumably occur just before they went extinct. This is expected if extinction is not so rapid that species cannot track their preferred habitat [1].

The fossil record is biased, particularly when the number of localities (hence sampling) is small and uneven. The number of localities occupied by a species is expected to be small just before species extinction. As we were particularly interested in the last intervals, we devised a test to examine if sampling inequality among successive time bins affects the computation of distances. We created dissimilarity matrices for distance per million years for all pairs of CFs and for sampling intensity, and tested their correlation, using a Mantel test of randomizations. The test revealed that sampling inequality does not influence our results (see the electronic supplementary material for details).

To test for the relationship between stasis and species duration, we retrieved information from the NOW database on species molar shape and size. These are both attributes highly indicative of the type of food consumed, and therefore of the habitat exploited. Data on carnivores were supplemented by us as explained in the electronic supplementary material. Tooth shape variables were reduced by principal component (PC) analysis and the first two PC scores (amounting to 87% of variance explained) were used for computing morphological disparity for all clades in the tree. We examined the correlation between disparity values and the average of the distances covered by species in all clades in the tree. We predict that the correlation is negative, indicating that in clades with low morphological disparity, species travel long distances to track their habitat.

3. Results

The mean distance between successive CFs is 3258.1 km. As the mean stratigraphic duration is 6.154 Myr, the location moves, on average, 529.4 km Myr⁻¹. The maximum distance covered between any pair of successive intervals is 1898.6 km, averaged over all species, which is some 3.5 times the mean.

The relationship between the total distance covered by species during their existence and their stratigraphic duration is significant and positive (table 1), both by using raw data (p = 0.022), and under a phylogenetically explicit context (p_pagel = 0.017, p_brownian = 0.084, table 1). As home ranges and dispersal distance increase with body size [12], we tested if the distance covered in geological time by species is influenced by body size (table 1), and found that it is not (p_{raw data} = 0.454, p_pagel = 0.577, p_brownian = 0.844).

Table 1.

Regression statistics between log stratigraphic duration (in Myr) and total distance (in log km), maximum distance (in log₁₀ km) and body size (in log g). (AIC, Akaike information criterion score; log Lik, log likelihood.)

	raw data	Brownian motion	Pagel's transform
total distance
slope	0.253	0.165	0.246
p	0.022	0.084	0.017
AIC		22.074	−0.666
log Lik		−8.037	4.333
maximum distance
slope	0.130	0.016	0.139
p	0.269	0.879	0.221
AIC		24.894	3.416
log Lik		−9.447	2.292
body size
slope	0.026	−0.008	0.019
p	0.454	0.844	0.577
AIC		26.797	6.991
log Lik		−10.340	0.504

The largest distance per million years is covered by 41 species out of 72 in their last interval. This is more often than expected by chance (12 occurrences; p_binomial ≪ 0.001). The cladewise correlation between the total distance and morphological disparity is significant and negative. As the tree contains polytomies, we randomly resolved the tree 100 times, and computed the correlation each time. The average correlation is −0.255 (95% CI = −0.251 to −0.259). This indicates that cheek teeth shape is less variable in clades of species that moved longer distances.

4. Discussion

Testing habitat tracking is difficult because palaeohabitats should be recognized, and the presence of fossil species living there determined. Consequently, tests of the habitat-tracking hypothesis are mostly anecdotal. Here, we equated habitat tracking with the spatial displacement through time of the territory most densely dwelt by a species. Whether or not this is a good approximation, we still have a measure of dispersal that can be used for testing two corollaries of the habitat-tracking hypothesis: (i) species movement enhanced the probability of survival, and (ii) species that shifted their range were morphologically static. Hypothesis (i) is supported by the positive relationship between the average distance between CFs and their stratigraphic duration, a relationship that holds even when phylogeny is taken into account. On the one hand, this test is very conservative. If there is no environmental change, and therefore no subsequent habitat shifts, then species are not expected to move. As a consequence, the relationship between species' dispersal distance and duration can be weak. On the other hand, species living longer might be expected to cover longer distances by chance. Even if the latter was true, however, we found that the maximum distance covered over a single interval disproportionately occurs in the one before extinction. Our interpretation is that when ecological conditions worsen, extinction becomes more probable and species react by moving over longer distances. This may or may not prevent extinction in the short term, but certainly prolongs survival.

Our second hypothesis similarly received strong support. Over 100 randomly resolved trees, the correlation between morphological disparity and the total distance covered is always negative, 65 times significantly so at α = 0.05, and 99 times at α = 0.1. Teeth are highly indicative of mammals' diets and habitat preferences [13–15], at least in ungulates. As such, it is conceivable that a morphologically stable tooth design indicates that species stayed in the same habitat [13–15]. This is expected if species moved to seek after their preferred habitats.

Habitat tracking is a major explanation for morphological stasis, which is one of the pillars of the theory of punctuated equilibria [9,10]. The core idea is that species remain static by actively seeking the same ecological conditions, therefore requiring little adaptive change in the face of changing environments [9,10]. We found consistent evidence in support of both these contentions.