Differential responses of amphibians and reptiles to land-use change in the biodiversity hotspot of north-eastern Madagascar

Large expanses of tropical rainforest have been converted into agricultural landscapes cultivated by smallholder farmers. This is also the case in north-eastern Madagascar; a region that retains significant proportions of forest cover despite slash-and-burn hill rice cultivation and vanilla agroforestry expansion. The region is also a global hotspot for herpetofauna diversity, but how amphibians and reptiles are affected by land-use change remains largely unknown. Using a space-for-time study design, we compared species diversity and community composition across seven prevalent land uses: unburned (old-growth forest, forest fragment, and forest-derived vanilla agroforest) and burned (fallow-derived vanilla agroforest, woody fallow, and herbaceous fallow) land-use types, and rice paddy. We conducted six comprehensive, time-standardized searches across at least ten replicates of each land-use type and applied genetic barcoding to confirm species identification. We documented an exceptional diversity of herpetofauna (119 species; 91% endemic). Plot-level amphibian species richness was significantly higher in old-growth forest than in all other land-use types. Plot-level reptile species richness was significantly higher in unburned land-use types compared to burned land-use types. For both amphibians and reptiles, the less-disturbed land-use types showed more uneven communities and the species composition in old-growth forest differed significantly from all other land-use types. Amphibians had higher forest dependency (38% of species occurred exclusively in old-growth forest) than reptiles (26%). Our analyses thus revealed that the two groups respond differently to land-use change: we found less pronounced losses of reptile species richness especially in unburned agricultural habitats, suggesting that reptiles are less susceptible to land-use change than amphibians, possibly due to their ability to cope with hotter and drier microclimates. Overall, old-growth forest harboured a unique diversity, but some species also thrived in vanilla agroforestry systems, especially if these were forest-derived. This highlights the importance of conserving old-growth forests and non-burned land-use types within agricultural landscapes.

villages and represent remnants of the continuous forest cover that existed in the region prior 173 to the large-scale deforestation that began in the early 20 th century (Gade 1996). The ten 174 fragments have not been burned in historic times. The forest fragments are all used for the 175 extraction of timber and non-timber forest products. Herbaceous fallows occur after shifting 176 slash-and-burn hill rice cultivation (locally referred to as tavy; Styger et al. 2007)  Park, we also organized two sampling campaigns during the driest period (late  early September 2018 and December 2018) and one during the wettest period (February 199 2019). During each campaign, we visited each plot once during the day (08:00 -17:00) and 200 once at night (18:30-23:00). Overall, we did six visits per plot, three times during the day and 201 three times at night. 202 We collected data on amphibian and reptile communities during time-standardized searches 203 (Kadlec et al. 2012). During the survey, we systematically searched each plot in a zig-zag 204 pattern (Kadlec et al. 2012 for follow-up DNA analysis and/or collected the specimen. To denote those individuals, we 215 used 'cf.' (confer or compare with) and 'sp. aff.' (affinis) in conjuncture with the genus name 216 when we found slightly different morphological characters compared to the literature that we 217 used or compared to a given identified species. We used sp. (sp1 or sp2) in conjuncture with 218 the genus name when we could not find any given identified species to compare with the 219 encountered individual. The sp. denotation was thus used to differentiate several unknown 220 species belonging to the same genus and we consider them as morphospecies but used 221 DNA barcodes to confirm species identification (see below). Following the field identification, 222 we resumed the searching time; thus abundance and diversity of amphibians and reptiles 223 were independent of the time spent on the plot. Throughout this manuscript, we refer to each 224 encountered individual as an 'encounter' rather than an individual as we cannot exclude the 225 possibility of having encountered the same individual at more than one sampling event. 226 227 We collected muscle or toe clips as tissue samples of individuals in case of non-reliable 228 morphological characteristics based identification. We preserved tissue samples in 229

Species identification with DNA barcodes
Eppendorf tubes with 90% ethanol, stored and analysed at the Evolutionary Biology 230 laboratory of Prof. Miguel Vences at the University of Braunschweig, Germany. We 231 preserved specimens in 70% ethanol for amphibians and 70% and deposited them in the 232 collection of the Regional University Centre of the SAVA region (CURSA), Antalaha, 233 Madagascar. We extracted genomic DNA following the standard single-tube salt extraction 234 protocol (Bruford et al. 1992). To do so, we cut small pieces from the collected tissue and 235 proceeded to protein digestion following the protocol of Cacciali  reptiles. Because most amphibian and reptile genera are completely endemic to 262 Madagascar, we could attribute endemicity also to morphospecies. 263

264
During the data analysis, we conducted all statistical using R version 3.5.1 (R Core Team Bonferroni correction. We then used the same approach to compare observed species 272 between unburned plots (old-growth forest, forest fragment, forest-derived vanilla) and 273 burned plots (herbaceous fallow, woody fallow, fallow-derived vanilla). 274 We used the encounter data to compute sample-size-based extrapolation curves (Hsieh et 275 al. 2016) with the 'iNEXT' package to assess the diversity (species richness, Shannon, and 276 Simpson diversity) across land-use types using the Hill number framework (Chao et al. Simpson diversity indices (Roswell et al. 2021). 284 To display the total species diversity in each land-use type, we sub-sampled 10 plots of the 285 20 fallow-derived vanilla agroforests. To do so, we randomly selected one fallow-derived 286 vanilla agroforest from each village. As one of the villages lacks fallow-derived vanilla 287 agroforests (village Andramanolotra, see Fig.1), this resulted in 9 agroforests. We proceeded 288 to select one additional plot from all remaining fallow-derived vanilla agroforests, enabling a 289 fair comparison of total species diversity across 10 plots of each land-use type.

Data analysis and visualization: species composition 295
To evaluate the differences in species community composition between land-use types, 296 between burned and unburned plots, and between villages, we used the metaMDS function

Encounters and observed species richness 313
In total, we made 6215 encounters and observed 119 species of amphibians and reptiles. 314 The 3694 amphibian encounters belong to 58 species, 15 genera, and four families (see SI 315 1). The most species-rich genera of amphibians are Boophis (11 species), Stumpffia (10 316 species) and Gephyromantis (eight species). We found all but one species to be endemic according to AmphibiaWeb (2020). Among the observed amphibian species, 22 (37%) could 318 not be identified to species level in the field. The 12 of them were recognized as candidates 319 for new species based on genetic barcoding, another eight of them were counted as 320 morphospecies based on criteria mentioned in the methods section and two were conferred 321 to identified species. Among the encountered amphibian species, seven are listed in the 322 'threatened' category (IUCN 2019). Among the threatened species, we recorded six 323 vulnerable and one endangered species. 324 The 2521 reptile encounters represent 61 species, 28 genera, and five families (see SI 1). 325 The most species-rich reptile genera were Phelsuma (11 species) and Uroplatus (six 326 species). We found 83% of reptile species to be endemic (Uetz et al. 2021). Amongst Old-growth forests plots had significantly higher observed amphibian species richness while 337 rice paddies plots had significantly lower compared to other land-use types ( Fig. 2A). The 338 other land-use types had similar amphibian species richness with no significant differences 339 (see SI 2). Plot level amphibian encounters also differed significantly between land-use types 340 (F 6, 72 =13.01, p-value < 0.001; Figure  Simpson diversity, old-growth forest varied significantly compared to all other land-use types. 362 Shannon and Simpson diversity also did not differ significantly between forest fragment and 363 forest-derived vanilla agroforest, respectively, compared to burned land uses, except to rice 364 paddy (Fig. 3, Table 1). 365 Encounter-based accumulation curves for reptiles revealed the highest species richness and 366 Shannon diversity in old-growth forest and the lowest in herbaceous fallow. Observed 367 species richness curves flattened off in all land-use types except woody fallow and fallow-368 derived vanilla agroforest. We also found an overlap of the 95% confidence intervals of 369 extrapolated species richness as well as Shannon and Simpson diversity within unburned 370 and burned land-use types except for Shannon and Simpson diversity in burned land-use 371 types (Fig. 3, Table 1). 372 The species diversity dropped more strongly across hill number orders (q=0 to q=2) in 373 amphibians than in reptiles after old-growth forest transformation (Fig. 3, SI 3 & SI 5), 374 highlighting that amphibian communities were more uneven than reptile communities (SI 5). 375 Furthermore, the comparison shows that old-growth forest communities of both amphibians 376 and reptiles were most uneven (SI 5).  comparisons showed that amphibian communities in old-growth forest and rice paddy were 382 significantly different from those of other land-use types. Forest fragments differed 383 significantly from fallow-derived vanilla, woody fallow and herbaceous fallow, but no 384 significant differences were observed between forest fragment and forest-derived vanilla. We 385 found no significant differences between forest-derived vanilla, fallow-derived vanilla, woody 386 fallow, and herbaceous fallow for amphibian communities (Fig. 5A, see SI 4). We compared 387 also the amphibian communities between unburned and burned land-use types and found 388  SI 21). 395 Reptile species composition showed significant differences among land-use types 396  Finally, the analyses of forest dependency revealed that 22 amphibian species (38%) were 409 found exclusively in old-growth forest, 11 amphibian species (19%) exclusively in forest 410 fragment, three amphibian species (5%) exclusively in forest-derived vanilla agroforest, and 411 four amphibian species (7%) exclusively in fallow-derived vanilla agroforest (Fig. 4A). 412 We observed 16 reptile species (26%) exclusively in old-growth forest, three reptile species 413 (5%) exclusively in forest fragment, four reptile species (7%) exclusively in forest-derived 414 vanilla agroforest, and one reptile species (2%) exclusively in fallow-derived vanilla 415 agroforest (Fig. 4B). No species were exclusively found in woody fallow, herbaceous fallow 416 and rice paddy for both groups. 417

418
Based on a rigorous study design, extensive field sampling, and DNA-based species 419 identification, we provide a comprehensive assessment of the response of amphibian and 420 reptile species diversity to land-use change in the biodiversity hotspot of north-eastern 421 Madagascar. Overall, we document a highly diverse herpetofauna with 58 amphibian and 61 422 reptile species and a high proportion of endemic species, 98% for amphibians and 83% for 423 reptiles.
At a plot-level, species richness for both amphibians and reptiles was very high, with up to 12 425 and 14 species, respectively. For both groups, old-growth forest was significantly different 426 from all other land-use types in terms of total species richness and community composition. 427 Rice paddy for amphibian and herbaceous fallow for reptile harboured the lowest species 428 richness. Reptile species diversity varied significantly between forest-derived and fallow-429 derived vanilla agroforests whereas amphibian species showed no difference. For woody 430 fallow, herbaceous fallow and rice paddy, no unique species occurred in any of these land-431 use types. Importantly, we found that amphibians and reptiles responded differently to land-432 use history. After any kind of old-growth forest conversion, whether through burning or not, 433 amphibian species communities were decimated and showed a high shift in community 434 structure compared to old-growth forest. In reptile communities, the forest alteration through 435 slash-and-burn showed a strong species loss as well, but changes were less pronounced if 436 old-growth forest was transformed to forest fragment or forest-derived vanilla agroforest 437 since these transformations refrain from using fire. 438

Outstanding diversity of amphibians and reptiles in north-eastern Madagascar 439
The diversity of amphibians and reptiles documented here was exceptional, both within

Response of amphibian diversity to land-use change 456
We found a strong negative response of amphibian species richness to any anthropogenic 457 land uses. Old-growth forest had a significantly higher observed amphibian species richness 458 with 38% species exclusively occurring there. Besides, the amphibian species composition is 459 unique compared to all other land-use types, driven by a high share belonging to the 460 Microhylidae family and Spinomantis, Guibemantis, and Mantella genera (Fig. 4A) forest fragments cannot substitute large and continuous old-growth forest (Vallan 2000). 468 The other land-use types (fallows that form part of the slush-and-burn cycle, vanilla 469 agroforests, and rice paddies) are of minor importance for amphibian conservation, given the 470 low species diversity that consists mainly of common and widely ranged species. 471 Nonetheless, amphibians could play an important functional role in these habitats: 472 abundance of amphibians is high throughout, reflected by high number of encounters, 473 particularly in rice paddies (see Fig. 3B and SI 9). As such, they may be an important food 474 source for other taxa or could provide pest control services in crops (Hocking et al. 2014). 475 The strong negative response of amphibians to deforestation could be explained by the fact 476 that numerous species rely on moist environments to avoid dehydration (Watling and Braga

Response of reptile diversity to land-use change 484
We found a strong effect of land-use history on reptile diversity. Unburned land-use types 485 had a significantly higher average species richness, more uneven communities, more unique 486 species, and a distinct species composition compared to burned land-use types and rice 487 paddy. The distinct species composition in unburned land-use types was caused by an 488 increased abundance compared to other land-use types of species belonging to the 489 Chamaeleonidae family and Uroplatus genus (Fig. 4B). This is, to our knowledge, the first impermeable skin covered with scales. This structure prevents excessive evaporative water 499 loss and adapts reptiles to various microclimates (Watling and Braga 2015). Furthermore, 500 egg development is not heavily impacted by temperature rise, which reduces the size of 501 hatchlings and accelerates incubation (Packard et al. 1982). Reptile lifestyle and traits may 502 thus enable reptiles to adapt to more open, hotter, and drier environments (Morin 2005), 503 making them less vulnerable to land-use change. 504

Land-use history of vanilla agroforests affects reptiles but not amphibians 505
In our study, we have separated forest-derived vanilla from fallow-derived vanilla agroforests, 506 thereby explicitly accounting for land-use history (Martin et al. 2020). For amphibians, we 507 found no differences between the two agroforestry systems across metrics. Reptile 508 communities, on the other hand, were significantly more diverse in forest-derived vanilla 509 agroforests on plot level, more species-rich overall, more uneven, and compositionally We further hypothesize, that the strong importance of forest-derived vanilla agroforests for 520 reptiles may be in part driven by leaf-litter depth, which is known to positively influence reptile 521 diversity and abundance (Urbina-Cardona et al. 2006). According to the same study, these 522 microenvironmental variables also showed positive effects on amphibians and thus cannot 523 explain why there are no differences between the two kinds of agroforests for amphibians. 524 Elsewhere, the loss of canopy cover had negative impacts on amphibian and reptile species 525 richness (Scott et al. 2006). Since canopy cover differed significantly between forest-derived 526 vanilla agroforest (higher) and fallow-derived vanilla agroforest (lower) in the same study site 527

Conservation implications 531
The strong negative response of amphibians to old-growth forest modification and the high 532 old-growth forest dependency of amphibians and reptiles calls for the effective protection of the last remaining old-growth forests. Additionally, conserving forest fragments within the 534 agricultural landscape will be important to many reptile and amphibian species that are 535 absent from other land-use types. This is particularly important given that numerous species Our findings from the north-eastern Madagascar also have broader relevance to 546 herpetofauna conservation in other tropical biodiversity hotspots. The differential response of 547 amphibians and reptiles to land-use change shows that conservationists should not treat 548 herpetofauna as a homogenous group when devising conservation programs. Furthermore, 549 the distinct differences in reptile communities between plots of contrasting land-use history 550 shows that past land-use should also not be overlooked, suggesting that the maintenance of 551 remnant forest fragments and forest-derived agroforests is important. Lastly, we highlight the 552 importance of unburned compared to burned land-use types in the agricultural matrix for the 553 conservation of reptiles. 554

555
We are grateful to all chef de fokontany, landowners, and Madagascar National Parks for 556 granting us access to sites and information. We thank Prof. Miguel Vences and his 557 laboratory staff for invaluable support with the DNA barcoding, Saskia Dröge for preparing 558 panel C of Fig. 1