Namib Desert primary productivity is driven by cryptic microbial community N-fixation

Carbon exchange in drylands is typically low, but during significant rainfall events (wet anomalies) drylands act as a C sink. During these anomalies the limitation on C uptake switches from water to nitrogen. In the Namib Desert of southern Africa, the N inventory in soil organic matter available for mineralisation is insufficient to support the observed increase in primary productivity. The C4 grasses that flourish after rainfall events are not capable of N fixation, and so there is no clear mechanism for adequate N fixation in dryland ecosystems to support rapid C uptake. Here we demonstrate that N fixation by photoautotrophic hypolithic communities forms the basis for the N budget for plant productivity events in the Namib Desert. Stable N isotope (δ15N) values of Namib Desert hypolithic biomass, and surface and subsurface soils were measured over 3 years across dune and gravel plain biotopes. Hypoliths showed significantly higher biomass and lower δ15N values than soil organic matter. The δ15N values of hypoliths approach the theoretical values for nitrogen fixation. Our results are strongly indicative that hypolithic communities are the foundation of productivity after rain events in the Namib Desert and are likely to play similar roles in other arid environments.

Desert ecosystems cover a substantial portion of the Earth terrestrial surface 1 and are characterized by very low productivity which is limited by water and nitrogen (N) availability [2][3][4] . During ubiquitous dry periods, deposited atmospheric or litter-bound N accumulates in these ecosystems and N-mineralization is stimulated only by spatially and temporally stochastic precipitation events 5 . During wet periods, drylands productivity increases exponentially and the regions act as transient carbon sinks 6 . For example, the 2011 global carbon land sink anomaly is attributed in large part to transient desert greening in Australia 6 . Biomass production in desert soils that are nutrient-(particularly N-) deficient, should be limited even after particularly intense wet events 7,8 . This perception is exacerbated by the fact that denitrification, the main process for N-loss in desert terrestrial ecosystems, is also controlled by water availability 9 . It has been estimated that over 99% of the nitrogen fixed by cyanobacterial-dominated biocrusts in arid ecosystems (~8 kg.ha −1 .yr −1 ) 10 is lost through this process 11 . Furthermore, desert N fixation has been estimated to range between 4.8 and 10.8 kg.ha −1 .yr −1 3 , which is similar to plant annual uptake in arid lands (12 kg.ha −1 .yr −1 ) 12 , and at the global scale, atmospheric N land deposition (in the form of nitrate [NO 3 ] or ammonium [NH 4 ]) is negligible when compared to atmospheric biological N 2 fixation [BNF] 13 . Consequently, N seems to be lost through denitrification and neither mineralized nor fixed at a sufficient rate to support the rapid and dense vegetation growth observed after pulsed precipitation events in drylands 6 .
In desert ecosystems, microorganisms are believed to be the key drivers of ecosystem processes and services 14,15 . In particular, cyanobacterial-dominated microbial communities found in cryptic niches (biological soil crusts, endo/hypo/chasmoliths) have the potential to actively participate in desert soil C and N budgets as dominant primary producers 10,[14][15][16][17][18][19] . In Antarctic desert soils, hypolithic N-fixation potential was implicated using acetylene reduction assays (ARA) and by the detection of bacterial nifH genes 17 . In the Mojave Desert, phototrophic microorganism-dominated hypolithic community N-and C-fixation was shown using stable isotope ratio of Nitrogen (δ 15 N), ARA and photosynthetically active radiation (PAR), respectively 16 . A recent shotgun metagenome study also revealed that Namib Desert hypolithic communities could mediate the full N-cycle (apart from ANAMMOX) and possessed substantial capacity for C-fixation as multiple copies of photosystem I and II cyanobacterial genes were detected 18 . A GeoChip microarray analyses of Antarctic Dry Valley hypoliths also recently confirmed that pathways comprising a complete microbially-mediated N-cycle were detected; i.e., N-fixation, (de)nitrification, ammonification, assimilatory/dissimilatory nitrogen reduction (A/DNR), and ANAMMOX 19 .
In deserts where microbial colonized rocks (hypo-, endo-and/or chasmo-liths) are highly distributed, their cryptic communities may represent a likely nutrient cycling hub providing desert soils with sufficient mineralized N to support plant growth. However, this hypothesis has never been experimentally confirmed in the field. In this study, we have used natural stable isotope ratios of Nitrogen (δ 15 N) and Carbon (δ 13 C) to evaluate if desert hypoliths constitute atmospheric N fixation hubs 20,21 .

Results
Soil and hypolithic samples were collected in the gravel plains and the dune fields of the central Namib Desert over a period of 3 years ( Fig. 1; Supplementary Table 1). The full analytical dataset (δ 13 C, δ 15 N, %C, %N and C/N) of the 152 samples (hypoliths: n = 52; surface soils: n = 53; subsurface soils: n = 47) is provided in Supplementary  Table 1. Of the thirteen SASSCAL weather stations spanning the sampling sites at the time of sampling, only two (Ganab once and Dieprivier thrice) recorded precipitation events over 20 mm a month before sampling and the sampling month (Supporting Table 2). This amount has been identified as the minimal necessary for C4 grasses, which cannot fix atmospheric N, to grow in the Namib Desert gravel plains 22 . In March and April 2014, 2015 and 2016 only 3 of the 6, 4 of the 26 and 3 of the 26 monthly cumulative precipitations records, respectively, represented more than 5 mm and precipitation was highly localized both in time and space (Supporting Table 2). Consequently, this strongly indicates that the C and N chemistries measured in this study were globally not influenced by atmospheric N deposition nor plant decay.
The edaphic and hypolithic C and N chemistries were found to be independent of the year of sampling (Kruskal-Wallis H, p > 0.05; Table 1) while they were globally significantly different in the different environments studied (Kruskal-Wallis H and Dunn's pairwise tests, p > 0.05; Table 1; Fig. 2A). Pairwise comparison showed that hypoliths presented significantly different N and C chemistries from the soils but that surface and subsurface soils did not (Kruskal-Wallis H and Dunn's pairwise tests, p < 0.05; Table 1, Fig. 2). Hypoliths presented significantly higher %N, %C and C/N ratios and lower δ 13 C values than soils (p < 0.05; Table 1 Table 3, Supplementary Table 1). Hypoliths, surface and surface soils from the rain and fog samples from the dunes (Table 3) show the same trends in the gravel plains, with higher δ 13 C, %C and %N in rain-zones (Kruskal-Wallis H, p < 0.05; Table 3, Supplementary  Table 1). A significant positive linear relationship between the 'distance to coast' , which is a proxy for MAP in the central Namib Desert (see Supplementary Table 1) 23,24 , and the %N, %C and δ 13 C values of the hypoliths, surface and subsurface soils was also identified (Fig. 4). This further demonstrated that water availability is a crucial determinant in desert C and N cycling.

Discussion
Hypolithons are microbial communities colonizing the ventral surfaces of translucent rocks, mainly quartz, and are commonly found in hot and cold desert pavement environments (Fig. 1B,C) 1, 26,27 . The lithic substrate provides the underlying communities with a stable substrate with sufficient transmitted incident light to support 'cryptic photosynthesis' , while protecting them from extreme environmental conditions (e.g. UV radiation and desiccation) 27,28 and buffering (rapid) changes in microenvironmental parameters 27 . In the gravel plains of the Namib Desert, hypolithic communities have been shown to selectively recruit their constituent microbial species from the surrounding soils, to be dominated by primary producers, i.e., cyanobacteria (notably Chroococcidiopsis sp.), and to support a wide range of heterotrophic taxa 29,30 . Cyanobacteria, and particularly a cryptic cyanobacterial Operational Taxonomic Unit (OTU) assigned to the genus GpI, were also found to drive hypolithic food webs based on co-occurrence network analyses 30 .
Our chemical analyses show that Namib Desert hypolithic biomass and soils are highly oligotrophic habitats. However, hypoliths were significantly less nutrient-limited than desert soils (Fig. 2B,C; Table 1), which is consistent with previous genetic evidence for high C-and N-fixing capacities 18,19 . Nutrient stratification in surface and subsurface soils was not observed (Fig. 2B,C; Table 1) 14 which confirmed that nutrient cycling is very limited in Namib Desert soils 31 . Edaphic and hypolithic C and N chemistries were also found to be independent of the year of sampling (Table 1), most probably reflecting their local and long-term "hydro-histories" 32,33 . Moisture source (i.e., fog vs rain) has previously been found to impact hypolithic community structures 30 in the central Namib Desert. Similarly, hypolithic colonization has been shown to be highly correlated to water availability in the central Namib Desert 34 and at the global scale 27 . Edaphic bacterial community structures and functions have also recently been found to be influenced by moisture source 35 . This is consistent with our finding that %N and %C values were significantly higher in the rainfall zone (Fig. 4 statistically significant increase in %C and %N along a west-east longitudinal transect was observed (Fig. 4). Since the mean annual precipitation (MAP) is higher in the rain zone than in the coastal fog zone and increases from the coast inland 23,24 , the significantly higher biomass observed in the rain zone samples (%N and %C, p < 0.005, Tables 2 and 3, Supplementary Table 1) almost certainly reflected increased edaphic and hypolithic microbial activities 35,36 . Both Namib Desert soils and hypoliths have δ 13 C and δ 15 N values typical of arid environments 16,33,37,38 ; both being significantly lower in the hypoliths (Fig. 3). With averaged δ 13 C values of −23.66 (±2.72) ‰ (Fig. 2E), Namib Desert hypoliths show a photosynthetic signature typical of C3-plants 39,40 . Surface and subsurface soils presented δ 13 C signatures located between those of the Namib Desert C3 plants/hypoliths and C4 plants (Fig. 3), indicating most probably that Namib Desert soils' C originate both by hypolithic and C3/C4 plant fixation and decaying of plant material. δ 15 N values close to 0‰ are characteristic of N acquisition via microbial fixation, values between +2 to +5‰ are typical of atmospheric N-fixation by fungal mycorrhizae or a mix of N-fixation and mineralized N from soils, and values >6‰ are the result of uptake of mineralized N 20,37 . Averaged δ 15 N values   for Namib Desert hypolith samples were 3.38 (±2.65) ‰ (Fig. 2F), which suggests that they derive their N via microbial (bacterial and/or fungal) fixation of atmospheric-N (as previously observed for Mojave Desert hypoliths which presented δ 15 N values of 0.6 (±0.18) ‰ 16 ), subsequently mineralized by ammonifying bacteria. This is supported by the detection of Cyanobacterial and Proteobacterial nifH sequences and contigs from Nitrosomonas sp., Nitrobacter sp. and Nitrospira sp. in a Namib Desert hypolith metagenome 18 . δ 15 N values for hypoliths were significantly lower than those of Namib Desert soils and plants (Fig. 3) 41 . Since discrimination of δ 15 N evolves positively in a system 42 , these results further demonstrate that (hot) deserts hypolithic communities are a N-fixation hub that positively contribute to N-fertilization of dryland soils. This result is fundamental in better understanding desert ecosystem functioning. While desert plants acquire C for growth through autotrophic photosynthetic activity, they require bioavailable N from soils for both growth and/ or increases in photosynthetic capacity 43 .
Desert soils are indeed globally N-limited 7,8 , but during wet anomalies, desert macrophytic plants (mainly C4 grasses) that cannot fix atmospheric N manage to acquire sufficient N to support rapid and substantial growth 6,43 . This has also been observed in the N-depleted gravel plain soils of the Namib Desert where the C4 short-lived perennial grass Stipagrostis ciliata grows extremely fast and covers their surface after precipitation events over 20 millimetres 22,44 . These plains are largely devoid of N-fixing plants such as Vachellia spp. that might increase the edaphic N supply. Our results therefore suggest that the bioavailable N necessary for S. ciliata to demonstrate such rapid growth in the Namib Desert after a rain event in the Namib Desert gravel plains is largely provided by the N-fixing capacity of hypolithic microbial communities. On the basis of δ 15 N values, it was suggested that plants acquire bioavailable N in the Atacama Desert from intermediate N-fixing hubs, such as lichens 45 and/or hypoliths 38 . They demonstrated that the plants did not directly obtain their N from fog precipitation; even in zones where fog constitutes the sole water source.

Conclusions
In unmanaged terrestrial systems, Biological Nitrogen Fixation is the primary process by which N enters the system 3,13,46 . BNF has also been found to be regulated by climate and positively correlated with moisture availability 11,47 . Consequently, we argue that hypolithic microbial communities, which are positioned at the bottom of the Namib Desert primary production web (Fig. 3), fix N 2 and produce sufficient bioavailable N (in the form of ammonium or nitrate 20 ) during wet anomalies to support their own growth as well as the growth of higher plants. The fact that 98% of the quartz rocks over 5 cm were colonized in a similar transect in the Namib Desert gravel plains, and that this coverage is independent from moisture source (fog vs rain, which conversely significantly impacted the Namib Desert hypolithic C and N chemistries; Fig. 4, Tables 2 and 3, Supplementary Table 1), strongly supports this view 34 . We therefore suggest that when hypoliths are abundant, as in the Namib Desert gravel plains 34 , they are a critical foundation of hot desert productivity via their capacity for N-fixation. In situ monitoring of S. ciliata growth in the Namib Desert (e.g., number of plants per m 2 ) in relation with the abundance of colonized hypoliths after controlled or natural 20 mm rain events in parallel with stable isotopic tracer studies 48 sh/could be performed to further demonstrate this point. Our results also particularly indicate that depending on the percent colonization and the distribution of colonized stones, desert plant response to rain could vary. For example, in deserts where quartz stone colonization vary, such as in the Atacama (from 27.6% to 0% in the semiarid and hyperarid regions respectively) 49 or the Taklimakan and Qaidam Basing deserts (from 12.6% to 0% along an aridity gradient) 50 , plant growth should be patchy after an intense rain event, while in deserts presenting 'constant' hypolithic colonization, such as the Namib Desert (~98% in every undisturbed gravel plain sites visited along the same aridity and fog/rain gradient we studied here) 34 , plant growth should be uniform.
The process rates of hypoliths, and also other cyanobacterial-dominated cryptic microbial communities (e.g., endolithic communities 14 ), during dry spells and wet anomalies, as well as plant growth after substantial rain events, should be characterized quantitatively in different deserts worldwide in order to be included in future climate models 51,52 . This is fundamental as hypoliths can cover up to 50% of dryland surfaces 1 , and thus arid lands may contribute more to global C-and N-cycling than previously estimated 10,53,54 .
Finally, Warren-Rhodes and colleagues 34 have observed that the colonization of quartz stones in the Namib Desert gravel plains were significantly lower in disturbed (i.e., only from 50% to 70% in highly disturbed and moderately disturbed sites, respectively) than in undisturbed (~98%) sites. This suggests that hypolithic communities need long-term soil stability to develop and shows that they are susceptible to environmental changes. Our study, therefore, further indicates that ecological restoration in the Namib Desert gravel plains, after mining for example 55 , may strongly depend on translucent rock re-colonization and thus be a rather long process.

Methods
Sample collection. Namib Desert hypoliths (n = 52) and surface (0-2 cm; n = 53) and subsurface (15-20 cm; n = 47) soil samples were collected aseptically in the dunes and gravel plains of the central Namib Desert in April 2014, 2015 and 2016 ( Fig. 1; Supplementary Table 1). At each sampling site, hypolithic biomass was scraped off the undersides of 3 to 5 translucent rocks collected within a 5 m radius, and pooled. At each site, this has led to the recovery of ~5 g of hypolithic biomass. Surface soil samples were collected from a hypolith-free area within the same 5 m radius and the subsurface soil samples immediately below them. All samples were kept at room temperature in sterile 15 mL falcon tubes prior to their analysis at the Stable Isotope Laboratory of the Mammal Research Institute, University of Pretoria.
Climatic data. The Namib Desert is characterized by an east/west longitudinal rainfall gradient which increases from the coast inland and by regular coastal fog events which can penetrate ~70 km inland 23 We also provide monthly precipitation records of 13 central Namib Desert weather stations from the Southern African Science Service Centre for Climate Change and Adaptive Land Management (SASSCAL) network (http:// www.sasscalweathernet.org/) 56 Table 2). The SASSCAL weather stations were chosen based on their proximity to the sampling sites. Twelve SASSCAL weather stations are located in the central Namib Desert gravel plains and one far west in the central Namib Desert dune fields. C and N analyses. Untreated aliquots were used for Nitrogen isotope measurements, while aliquots treated with a 1% HCl (v/v) solution (to remove all inorganic carbonates) were used for Carbon isotope value measurement. The samples were repeatedly washed with distilled water to neutral pH and dried at 70 °C. Aliquots of approximately 80 to 100 mg were weighed into tin capsules pre-cleaned in toluene and analysed using a Flash EA 1112 Series elemental analyzer coupled to a Delta V Plus stable light isotope ratio mass spectrometer via a ConFlo IV system (all equipment from Thermo Fischer, Bremen, Germany).
A laboratory running standard (Merck Gel: δ 13 C = −20.57‰, δ 15 N = 6.8‰, C% = 43.83, N% = 14.64) and a blank sample were run after every 5 unknown samples. All results were referenced to Vienna Pee-Dee Belemnite for carbon isotope values and to air for nitrogen isotope values. Results are expressed in delta (δ) notation using a per mille scale using the standard equation: where X represents 15 N or 13 C and R represents the 15 N/ 14 N or the 13 C/ 12 C ratio, respectively. Analytical precision was <0.09‰ for δ 13 C and <0.08‰ for δ 15 N.

Statistical analyses.
Statistical analyses were performed using the PAST v3.14 software package.
Principal component analysis (PCA) was performed on normalised datasets and based on Euclidean distances. Kruskal-Wallis H tests with pairwise Dunn's post hoc test were used to identify significant differences between samples from different environments (hypoliths vs surface soils vs subsurface soils), origin (Dune vs Gravel plain), years of sampling (2014 vs 2015 vs 2016) or moisture sources (fog vs rain) (Tables 1, 2 and 3). A total of 9999 random permutations were performed and p values were Bonferroni-corrected. Ordinary Least Square (OLS) was used to evaluate linear relationships between 'distance-to-coast' and hypolithic and edaphic δ 13 C, δ 15 N, %N, %C or C/N ratios. Hypolithic %N, %C and C/N ratios and subsurface %N were log-transformed to achieve near-normal distribution. Soil surface δ 13 C, δ 15 N, C/N, log(%N) and log(%C) and subsurface %C, and C/N were normally distributed.