Discussion
Here, using a data collation and meta-analytical approach, we
demonstrate that network type is the most important driver of ant-plant
network interaction specificity. Unstandardised H2’
specialisation results recapitulate previous studies, showing reduced
specialisation with elevation for myrmecophytes, higher specialisation
of myrmecophytes than other network types at lower elevation, and
decreased specificity in response to disturbance. However, all effects
were lost when H2’ was standardised by comparison with
null network models. This may have been due to confounding effects of
matrix size and connectance on uncorrected H2’. Both
ants and plants were significantly more phylogenetically specialised in
myrmecophytic networks than in other network types, although there was
no significant effects of elevation, latitude, or disturbance. This is
the most comprehensive analysis of global scale ant-plant network
structure to date, and the first to employ null models to standardise
network metrics and to explore phylogenetic specialisation.
More physically intimate network that involve greater commitment to
exchange of goods and services (in the case of mutualisms), are
predicted to evolve to become more specific (González-Teuber & Heil,
2015). We found this expected effect when assessing specificity
independent of phylogeny and uncorrected by null modelling
(H2’), with myrmecophytic networks at lower elevations
showing higher specificity than for other network types. Unexpectedly,
this effect was not present when H2’ standardised effect
sizes were used as the response variable. This indicates that the
results for unstandardised H2’ may have been driven by
other aspects of network structure, and not by specificity per
se . post-hoc analyses show that both total network species richness,
and weighted connectance are strongly negatively correlated with
H2’ (see supplementary, appendix 8), but are only weakly
related to H2’ z-scores (appendix 8). This might be
because networks with higher connectance are those in which links are
found evenly across species in the network, and hence have lower
specificity. However, the differences in results when standardising
H2’ suggest that, at least in terms of network
specificity, this evenness of links is to be expected on the basis of
network structure alone. Hence it is vital to compare observed network
metrics to distributions expected under random network assembly to avoid
biases due to confounding aspects of network structure. This approach
has not always been implemented in previous studies.
However, in terms of network phylogenetic specialisation, using the dsi*
metric, which is already standardised against null models, we found that
myrmecophytic networks were more specialised than other network types.
This was the case both in terms of ant specialisation on plants and
plant specialisation on ants. Each ant species interacted with a more
phylogenetically clustered group of plant species than would be expected
at random, and vice versa. This potentially indicates long-term
coevolution between myrmecophytic plants and their ant partners,
characterised by exchange of multiple goods and services. Plants can
provide specialised morphological structures (e.g., domatia) that
promote partner choice and thus allow direction of benefits to more
beneficial mutualistic partners (Heil & McKey 2003). Myrmecophytic
plants also sometimes provide FB and EFN for their ant partners. Ants
can provide herbivore protection, competitor trimming, nutrients (Mayer
at al., 2013), and CO2 for photosynthesis (Treseder et al., 1995). Such
complex behaviours and morphologies are likely to be phylogenetically
conserved, resulting in the observed high levels of network phylogenetic
specificity in myrmecophytic networks. This is consistent with previous
(non-phylogenetic) work showing that myrmecophytic networks are
characterized by strong compartmentalization (Fonseca & Ganade, 1996).
The other network types (myrmecophilic, myrmecochorous and foraging)
showed lower levels of phylogenetic specialisation than myrmecophilic
networks. This is likely due to the lack of evolution of specialized
morphological structures relating to the ant-plant interaction. Although
EFNs predominantly attract ants for protection, they can also attract
other defensive arthropods including parasitoids, wasps, spiders, mites,
bugs, and predatory beetles (Heil, 2015). It is also possible that
plants produce EFNs to reduce ant consumption of flower nectar and so to
maximise visitations of other pollinators on flowers (Wagner & Kay,
2002). However, myrmecophilic networks showed higher specialisation than
foraging networks in terms of ant specialisation on plants. This may be
because ants can benefit greatly from EFNs, which contain
monosaccharides and disaccharides, and amino acids that are an important
energy source (Marazzi et al., 2013) and even alter the predatory
behavior of some ants (Wilder & Eubanks, 2010). Indeed, some plants can
even coerce their EFN-feeding ant partners through “addiction” based
on enzyme inhibition, preventing the ants from feeding on other food
sources (Heil et al., 2014; Houadria et al., 2023). Myrmecochorous ants
tended to be generalists, being attracted to the non-specific food
offered by the plants in the elaiosome (Levine et al., 2019). Although
Anjos et al. (2018) showed that ants attracted to elaiosomes (a small
lipid-rich structure used by ants as a food source) are more specialized
than ants attracted to fruit pulp, our data combined both network types
to increase statistical power, and so we were unable to explore this.
Lastly, foraging networks exhibit a lower specialisation, presumably in
part because this was the only network type that did not necessarily
involve a mutualism between the partners. Species involved in these
networks are highly adaptable and tend to exploit a wide range of
resources within their environment.
Although unstandardised specialisation (H2’) showed
similar relationships with elevation and habitat disturbance to previous
studies, these relationships were not present for analysis of
standardised H2’ or for phylogenetic network
specialisation (dsi*). The interaction between network type and
elevation was due specifically to differences between responses of
myrmecophytic networks and myrmecochoric and foraging networks. The
former showed a strong reduction in specificity (H2’)
with elevation, while the latter two showed uniform low specialisation
across all elevations. This pattern is consistent with the lack of
herbivores at higher elevations and hence the reduced need for plant
protection by ants (Moraes & Vasconcelos, 2009). For example,Myristica subalulata , a myrmecophytic plant that is abundant
across a range of elevations, benefits less from myrmecophytic networks
at higher elevations, and shows lower specificity towards its ant
partner (Plowman et al., 2017). Previous work has showed that more
intimate networks, such as myrmecophytic networks, can be more
susceptible to disturbance (Emer et al., 2013; Fayle et al 2017), while
less intimate networks might not be significantly affected by
disturbance in island regions (Klimes 2017) but may experience greater
impact in mainland or continental contexts (Corro et al., 2019).
However, we found no difference in response to disturbance between
network types (no significant interaction between the predictors),
although we did find an overall decreased specialisation
(H2’) in response to anthropogenic habitat disturbance.
Our failure to replicate any of these results with either standardised
H2’ or with measures of phylogenetic network structure
(dsi*) raises the possibility that results from previous studies are
artefacts, again driven by variation in other network properties. For
example, mean network species richness in undisturbed networks was 61.7,
compared to that in disturbed networks of 33.9, and network species
richness is negatively correlated with H2’
(supplementary, appendix 8). However, combining networks that have been
collected using different methods, and with differing sampling efforts
is likely to introduce considerable noise into response variables, even
when these are standardised against null models and so we feel that our
results do not necessarily preclude the existence of such patterns.
Taken together, our results show that ant-plant network specificity is
not strongly affected by latitude, elevation or anthropogenic habitat
disturbance, but that rather the mode of interaction between the
partners is most important. Mutualistic networks involving myrmecophytic
plants are highly phylogenetically specialised, due to their long term
coevolutionary interaction. Although our meta-analysis recapitulates
previous results in terms of relationships between unstandardised
H2’ and elevation and disturbance, these results are not
present when H2’ is standardised against null models, or
where phylogenetic specialised is assessed. This demonstrates the
importance of standardising metrics of network structure against null
expectations. Overall, we show that strength and intimacy of mutualistic
interactions drives patterns of network specialisation at global scales,
even across gradients of elevation, latitude and anthropogenic land-use
change, all of which have minimal impacts on network structure.