Evolution of Crassulacean Acid Metabolism (CAM) as an escape from ecological niche conservatism in Malagasy Bulbophyllum (Orchidaceae)

Despite growing evidence that niche shifts are more common in flowering plants than previously thought, still little is known about the key physiological (e.g. photosynthesis) traits underlying such niche shifts. To address this question, we here combine a comprehensively sampled phylogeny for mostly epiphytic Malagasy Bulbophyllum orchids (c. 210 spp.) with climatic niche and carbon isotope-derived photosynthesis data to infer the groups’ spatial-temporal history and to test the role of Crassulacean Acid Metabolism (CAM), a highly water-use efficient type of photosynthesis, in facilitating niche shifts and diversification. We find that most extant species still retain niche (Central Highland) and photosynthesis (C3) states as likely present in the single mid-Miocene (c. 12.70 Ma) ancestor colonizing Madagascar. However, we also infer a major transition to CAM, linked to a late Miocene (c. 7.36 Ma) invasion of species from the sub-humid highland niche first into the island’s humid eastern coastal, and then into the seasonally dry ‘Northwest Sambirano’ rainforests, yet without significant effect on diversification rates. These findings support the rarely recognized hypothesis that CAM in tropical epiphytes may be selectively advantageous even in high rainfall habitats, rather than presenting a mere adaptation to dry environments or epiphytism per se. Overall, our study qualifies CAM as an evolutionary ‘gateway’ trait that considerably widened the spatial-ecological amplitude of Madagascar’s most species-rich orchid genus.


Introduction 4 0
Analysing the extent to which species or clades are able to retain or shift their environmental 4 1 niche space over time is key to understanding large-scale patterns of lineage diversification 0 4 comparing a full (six-parameter) model with unconstrained parameters (i.e. q AB , q AC , q BA , q BC , 2 0 5 q CA , q CB ) to six models of zero uni-directional (i.e. asymmetrical) transitions for all pairwise 2 0 6 combinations of states (i.e. q AB = 0; q AC = 0; … q CB = 0) (see Table S3). Model selection 2 0 7 followed 2 logarithmic Bayes Factors (BFs), with values > 6 (or 10) considered as strong (or  Nocturnal fixation of atmospheric CO 2 into malic acid is characteristic of CAM 2 1 2 photosynthesis and results into a higher accumulation of 13 C in plant carbon when compared 2 1 3 to C 3 photosynthesis (O'Leary 1988;Bone et al. 2015). Thus, carbon isotope ratios ( 13 C/ 12 C) 2 1 4 can be used as proxy for photosynthetic pathway (Osmond et al. 1973). Accordingly, we which were included in the phylogeny of the present study (see also were collected and silica-dried in the field (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) or obtained from herbarium 2 1 9 specimens (WU, SZU, MO, P). About 2 mg per sample were subjected to carbon stable values less negative than -20‰ indicate CO 2 fixation predominantly via CAM, values more 2 2 8 negative than -20‰ do not preclude the possibility of low-level CAM, but indicate that this 2 2 9 did not make a major contribution to carbon gain. Ancestral character states for C 3 vs. CAM 2 3 0 were reconstructed in BAYESTRAITS, using the same methods and settings as described above was subsumed together with B, resulting in a niche coding C vs. not C (or C vs. AB). Model 2 3 5 selection followed BFs (see above). For quantifying rates of niche evolution, we first subjected dataset 2 (3036 locality-climate 2 3 9 points) to PCA (as described above) and calculated mean scores per species based on the first 2 4 0 two components (PC1, PC2), respectively. We then used BAMM v.2.5.0 (Rabosky 2014) to 2 4 1 infer rates of niche evolution along the MCC chronogram (with outgroups pruned) by we used BAMM to estimate rates of net lineage diversification [r = speciation (λ) -extinction 2 4 4 (μ)], taking incomplete taxon sampling into account (sampling frequency = 0.852). One visualization of rates through time were done in BAMMTOOLS after a 10% burn-in (see above).

4 9
In addition, we fitted ten birth-death (BD) models to the MCC chronogram with either 2 5 0 time-dependent (six models), palaeo-temperature-dependent (two) or diversity-dependent 2 5 1 (DD) diversification rates (two; see Table S5), using, respectively, the fit_bd and fit_env analyses (BAMM; RPANDA/DDD) were also run separately on subclade C and the pruned tree   respectively, each based on 1000 permutations (two tailed).   respectively (data not shown). However, we favoured K = 3 as the optimal (i.e. more All pairwise comparisons between clusters revealed significant differences for each of the two 2 9 8 axes tested (Mann-Whitney test, all P < 0.001).

9 9
The present-day ENMs generated for clusters A, B, and C (hereafter 'niches') had high humid montane forest and sclerophyllous forest/scrubland, receiving relatively low annual 3 1 0 rainfall and low annual (but variable seasonal) temperature. Highlands (C) with high probability (PP = 81.1), and the same was true for the crown nodes Eastern Lowlands (B) or, very rarely, from C to the Sambirano (A). Clade A, however, transitions occurred (towards the tips) from B to A or C (or very rarely from A to B), resulting 3 2 4 in a subclade with relatively high niche diversity (Fig. 1b). Nonetheless, across the entire 3 2 5 phylogeny, ancestral niche conservatism (C) prevailed, with shifts occurring mainly from C to 3 2 6 B and from B to A (Fig. 1b). In support of this, BAYESTRAITS strongly rejected models 3 2 7 opposing these two latter transitions (logBF q CB = -6.9; q BA = -10.4) but none with zero (uni-3 2 8 directional) transitions (i.e. q BC , q CA , q AB or q AC = 0; logBF = -1.0-0.9; see Table S4).  (Table S4), resulting in a bimodal frequency distribution (Fig. 1b). Based on our (and fixed within) subclade C (29/31, c. 93.5%; Fig. 1b; Table S4). Accordingly, we  For quantifying rates of niche evolution in BAMM, we used the mean PC1 and PC2 scores of 3 4 7 each species based on the 3036 locality-climate dataset 2 (see Table S6 for component for PC1 relative to PC2 (Fig. 2a, b). Notably, for the full tree, niche evolution along PC2 only Ma), and then steadily increased toward the present. By contrast, in subclade C, niche 3 5 4 evolution for PC1 and PC2 was temporally congruent and steadily increased toward the 3 5 5 present at higher rates than in the remaining lineage (Fig. 2c).

5 6
Considering lineage diversification, we generally inferred relatively constant, yet 3 5 7 slightly declining rates of r throughout the history of Malagasy Bulbophyllum, regardless of 3 5 8 the dataset used (i.e. full tree with or without subclade C, or the latter separately; Fig. 2a-c).

6 1
According to RPANDA/DDD, a time-dependent, pure birth model of diversification, with 3 6 2 constant speciation and no extinction, provided the best fit for each of these three datasets 3 6 3 (Table S5). We note that models of environmental-dependent (palaeo-temperature) and full tree with or without subclade C included (Table S5). In both instances, however, the 3 6 6 estimated carrying capacity (K) was almost two to three times the current species richness (K 3 6 7 = 555/210 and 400/171 spp., respectively). Overall, these results indicate mostly constant 3 6 8 speciation with zero extinction throughout the group's history, even though minor effects of 3 6 9 temperature and/or diversity on diversification cannot be fully excluded.  Based on a multi-locus plastid/nuclear phylogeny, we dated the crown age of Malagasy 3 7 9 Bulbophyllum to the mid-Miocene, c. 12.70 Ma (95% HPD: 10.39-15.27 Ma) (Fig. 1). This  Major transition from C3 to CAM, associated with invasion of high rainfall forest 3 9 8 Our results clearly identify C 3 as the ancestral and predominant mode of photosynthesis in  Fig. 1b, Table S4).

0 2
Thus, the majority of species still occupying the ancestral 'Central Highlands' niche (C) 4 0 3 exhibit C 3 photosynthesis, while CAM is mostly associated with relatively rare niche shifts to  (Fig. 1b, Table S3).

1 7
Instead, these data suggest that CAM initially served as a 'gateway' trait (sensu Donoghue & conditions in the mid-to-late Miocene is frequently cited as potential factor promoting the 4 2 8 evolution of CAM (e.g. Bone et al. 2015;Hu 2018;Li et al. 2019). In fact, this timing broadly 4 2 9 matches the late Miocene diversification of subclade C and its corresponding shifts towards 4 3 0 CAM and into the Eastern Lowlands (see above). However, this timing also matches a period 4 3 1 when the Indian Summer monsoon, whose seasonal precipitation influx is thought to have Coast of Madagascar, respectively (Cornet 1974;Buerki et al. 2013). In turn, these climatic