Ecology and spatial pattern of cyanobacterial community island patches in the Atacama Desert, Chile



[1] Plant landscape ecology studies have been carried out for decades and are fundamental to biological research. In contrast, few corollary spatial landscape studies exist for microorganisms, particularly in extreme environments. To address this gap, we mapped the abundance and spatial distribution of photoautotrophs colonizing translucent rocks in several sites in the Atacama Desert, including the hyperarid core. Cyanobacterial communities at all sites are predominantly (≥75%) ‘perilithic’ (confined to the periphery of rocks) and occur in non-random spatial patterns (“island patches”) at multiple scales. Cyanobacterial patches typically contain 1-5 colonized rocks but in some cases support much higher numbers. A high resolution mapping of a single 18-m2 rock cluster at the Aguas Calientes study site (25°S, 69°W) revealed colonization of 5.2% (49 of 948 quartz rocks) and showed colonized rocks to be much larger (∼2X) than the available mean rock size. Ripley's K and point pattern analyses show that quartz rocks are not “selected” or occupied by cyanobacteria randomly, but that non-random processes distinct from those creating the background rock pattern must be invoked to explain microbial patchiness in the Atacama Desert. These processes include physical controls (rock size/orientation, microtopography) that reflect resource (water) limitations, and biological dispersal via rainfall, fog and wind.

1. Introduction

[2] Deserts and arid lands cover approximately one-third of the Earth's land surface, making them important ecosystems worldwide. Within these environments, desert pavements (gravels that mantle surface soils) extend for several meters to thousands of kilometers and are a predominant habitat for microbial life. In particular, “lithic” (lithobiontic, lithophytic) cyanobacterial communities (LCC), comprised of co-occurring cyanobacteria and heterotrophic bacteria, colonize the cracks (chasmoliths), pore spaces (endoliths) and undersides (hypoliths) of translucent rocks (e.g., quartz, gypsum, calcite) in the pavement to gain protection from the harsh external environment [Friedmann et al., 1967; Friedmann and Galun, 1974; Golubic et al., 1981]. In the world's driest deserts, these communities may harbor the few primary producers within the larger landscape [Friedmann, 1980; Allen, 1997; Warren-Rhodes et al., 2006; Wierzchos et al., 2006].

[3] A wealth of studies have characterized LCC ecology, including community composition, metabolic activity, photosynthesis, and microclimate [Friedmann et al., 1967, 1993; Nienow et al., 1988a, 1988b, 2003; Cockell and Stokes, 2004; Omelon et al., 2006]. Few studies, however, have explicitly examined the spatial component of LCC ecology—particularly at extents greater than a few mm [Broady, 1981; Nienow et al., 1988a, 1988b; Cockell and Stokes, 2004; Green and Bohannan, 2006; Omelon et al., 2006]. For higher plants, such studies have yielded fundamental insights into concepts and models such as self-organization, patch dynamics, hierarchical scale, biogeography, positive biological feedback and ecohydrology [Peterson, 2000; Urban et al., 2002; Barrett et al., 2004; Rietkerk et al., 2004; Belnap et al., 2005; Ludwig et al., 2005].

[4] Pioneering work in adapting plant ecology to the study of cyanobacterial soil crusts has recently been undertaken in arid environments [Belnap et al., 2005; Bowker et al., 2005, 2006]. This paper seeks to extend such work to LCC in hyperarid deserts. To that end, we use available data from earlier LCC ecological studies in the Atacama Desert, Chile (where spatial concerns were not an explicit focus) and combine it with a recent spatial mapping study to (1) determine LCC spatial pattern across multiple scales, and (2) discuss possible processes underlying the observed spatial patterns.

2. Field Locales

[5] The Atacama Desert is a unique location for microbial study due to extremely low water availability, particularly within its hyperarid core (24–25.5°S; mean annual precipitation, MAP, ∼0–5 mm). South of the core, rainfall increases with latitude, and a predominant shift to less arid conditions occurs at ∼26–27°S [Latorre et al., 2002; Warren-Rhodes et al., 2006]. Concomitant with this shift is a transition to more abundant and diverse microbial, floral and faunal populations [Villagrán et al., 1983; Latorre et al., 2002, 2003; Navarro-González et al., 2003]. In addition to the positive latitudinal rainfall gradient, precipitation and fog decrease longitudinally (minimum at ∼69–70°W) from the Pacific coast inland to the Atacama's Central Depression, where westerly rainfall and fog and easterly Andean snowfall rarely penetrate.

[6] Radiation and marine fogs frequently occur in the Atacama and are important to both coastal and inland ecosystems [Miller, 1976; Larrain et al., 2002; McKay et al., 2003]. Inland fog frequency is largely dependent upon distance from, and topographic connection to, the Pacific coast [Cereceda et al., 2002; Ewing et al., 2006]. In particular, sites east of gaps in the coastal range support comparatively rich plant and/or lichen communities, whereas sites east of large peaks (≥2100 m) are extremely dry and microbially depauperate. This barrier, or “fog shadow,” effect, coupled with exceedingly low inland precipitation in the hyperarid core, forms the boundary for a region possessing “Mars-like” soils. Two of our long-term study sites (Yungay and Aguas Calientes) fall within this region, while a third (Altamira) is located at the southern transition to relatively wetter hyperarid conditions.

[7] Owing to its hyperaridity, the Atacama has received considerable focus for geochemical and soils research [Rech et al., 2003; Quinn et al., 2005; Ewing et al., 2006], microbial study [Navarro-González et al., 2003; Wierzchos et al., 2006; Lester et al., 2007], and astrobiological and robotic exploration [Cabrol et al., 2007]. Our previous Atacama investigations of LCC revealed variations in abundance, diversity and organic carbon residence times at the aforementioned sites and provided preliminary details of LCC spatial patterns [Warren-Rhodes et al., 2006]. An overview of the methods and key results from these initial studies is provided below and these baseline data are included in the paper as needed. Additionally, we incorporate recent data from a more detailed high-resolution spatial study at Aguas Calientes (AC). AC was chosen because it is the only site in the hyperarid core where we have located sufficient numbers of LCC-colonized rocks to perform spatial analyses. Second, the higher density of quartz rocks at AC facilitates an examination of the possible effects of underlying quartz rock distributions and overall habitat availability on LCC spatial patterns and allows spatial mapping to occur in a small area characterized by a homogenous macroscale climate.

3. Methods

3.1. LCC Ecological Parameters: Previous Studies (2002–2003)

[8] At all sites, LCC abundance was measured as percent colonization (number of colonized quartz rocks per total quartz rocks × 100) [Warren-Rhodes et al., 2006]. To ensure that a representative site-level measure of LCC abundance was obtained, decisions regarding sampling unit shape, size and number of replicate units were determined through pilot studies [Andrew and Mapstone, 1987]. For this parameter, a site was defined as an ∼1 km2 area within which transects were completed. (For data other than abundance, such as climate, the term site may also refer to a larger geographical area up to ∼30 km2.) Two criteria for site-level abundance sampling were set: (1) a minimum of ten 1 m-by-50 m transects (as determined from pilot studies) and (2) a minimum of 1,000 quartz rocks.

[9] Study areas (≤0.5 km2) within Yungay and Altamira were chosen based on the presence of quartz rock. Within these areas, transects (50-m2) were placed and distributed randomly by distance and direction. At AC, this method was modified to accommodate limited quartz rock availability, which was concentrated in dense clusters. The study area within AC was selected randomly, and from that we examined the nearest cluster large enough to accommodate a 50-m2 transect, with two transects completed. At all sites, quartz rocks within each transect were counted and sized to the nearest 0.5 cm (maximum length), and LCC microbial parameters were assessed by visual inspection. The spatial location of each colonized rock was measured to the nearest cm (x-direction) along the 50-m transect, and these data were used to test for LCC spatial pattern (at m to tens of m extents).

3.2. LCC Spatial Pattern: 2006 Study at AC

[10] In 2006, a detailed study of LCC spatial pattern was completed at AC with the objective of mapping spatial pattern at a higher resolution than previous surveys. The study was designed to test the following null hypotheses: (1) the spatial distribution of LCC is random, and (2) LCC spatial pattern is explained by the underlying background pattern of quartz rock.

[11] An area ∼1 km west of the AC site was chosen for the 2006 spatial study, within which a 6 m-by-3 m quartz rock cluster was selected on the criterion that it contains ≥3 colonized rocks. (Hereafter, the original AC 2003 study site is referred to as AC and the 2006 site as AC2). Because the 2003 study showed that quartz clusters were few in number and only rarely colonized (<0.5% of available habitat), only a single cluster at AC2 was disturbed in order to preserve critical cyanobacterial habitat. A 1-m2 quadrat grid (1 cm-by-1 cm smallest mesh grid size) was used to examine all quartz rocks within the 18-m2 cluster and map LCC location and rock distribution. Within each 1-m2 quadrat (n = 18), all quartz rocks ≥5 cm were counted, sized (nearest 0.5 cm, maximum length, rock thickness) and recorded spatially, along with LCC type and position of organisms on each colonized rock. Slope was estimated using a spirit level in 6 places along the edges of quadrat squares: two measurements were taken at each point (along the vertical and horizontal grid directions). The deviation of ground surface from 0° slope was estimated by measuring the vertical difference between spirit level at 0° and the downslope ground surface at the end of the level (25-cm length). Errors in the vertical measurement are ∼±1 mm (error in slope ±0.004°). Measured slopes were used to estimate relative elevation along the length of the quadrat grid, with an estimated error of ±1.25 cm. The resulting topographic map is consistent with field observations.

3.3. Climate Monitoring

[12] In-situ environmental data at all sites has been collected since Oct. 2001, including air temperature (TA) and relative humidity (RHA) (Onset Computer, Hobo PRO® H08-032-08, ∼1 m above ground) and rainfall (Onset RG2-M). Self-contained Hobo PRO® dataloggers to measure soil temperature (TS) and pore-space relative humidity (RHS) were also placed 2–5 cm under or adjacent to quartz rocks (depending upon LCC location). These “nano-climate” sensors are important for monitoring soil and rock conditions “at the functional scale at which the organisms are likely to respond to their environment” [Andrew and Mapstone, 1987; Friedmann et al., 1987]. Liquid water in soil (LWS) available to hypoliths is estimated from Hobo® data by assuming liquid water is present at RHS ≥ 95% (corroborated by rainfall and soil conductivity data).

[13] As a proxy for liquid water on rock surfaces (LWR), Campbell (237-L) moisture-sensing grids (positive readings at grid voltage, V ≥ 0.005) and/or Spectrum leaf wetness grids (positive readings V ≤ 2.095) were employed. Both sensors consist of a flat plastic grid that measures a change in electrical resistance between interlaced gold-plated fingers with the presence of liquid water [Armstrong et al., 1993]. Originally designed to record moisture on leaf surfaces, the grid data are used to estimate the maximum total hours of liquid water available to surface and/or near-surface LCC from fog (or dew) condensation. These grids, along with Hobo® loggers, are rugged, practical ways to monitor rock and soil moisture in hyperarid environments over long periods [McKay et al., 2003; Warren-Rhodes et al., 2007a].

[14] A more direct measure of liquid water available to LCC is obtained from conductivity sensors (voltage drop across 2 bare wires), with wires epoxied in proximity to cyanobacterial communities [McKay et al., 2003]. At several sites, conductivity sensors were installed on rock tops, sides and bottoms at several depths (e.g., surface, 1-2 mm, 5 mm). These were used for comparison with Hobo® RH loggers and moisture grids. Measurements were typically taken every minute, with 30-min maximum and minimum output averages. Historical precipitation data were obtained from the Dirección General de Aguas II Región and Dirección Metereológica de Chile.

3.4. Statistical Analyses

[15] The index of dispersion (ID) was used to test LCC site-level spatial randomness at Altamira, with random quadrat combinations chosen to ensure the validity of the chi-square test (p < 0.05) [Diggle, 2003]. (Altamira was the only site with a sufficiently large colonized rock sample size to conduct site-level spatial analyses.) Spatial patterns based on the ID are defined as non-random (ID = 1), spatially aggregated (ID > 1) or uniform (ID < 1). For the AC2 spatial study, statistical summaries were calculated both on individual and combined quadrats.

[16] A Ripley's K approach [Ripley, 1981; Lancaster and Downes, 2004] was used to test the following null hypotheses: (1) the spatial distribution of LCC is random, and (2) LCC spatial pattern is explained by the underlying background pattern of quartz rock. The K(t) function, where t is distance, is the cumulative distribution function of all point-to-point distances. For ease of interpretation, a transform of this function, L(t) = equation image, is used that linearizes K(t). If a pattern has complete spatial randomness (CSR) L(t) = t; L(t) > t represents clumped or aggregated patterns where inter-point distances tend to be smaller than characteristic of a random pattern; and L(t) < t represents over-dispersed patterns where inter-point distances tend to be larger than those characteristic of a random pattern.

[17] In a preliminary step, we compared the observed L(t) of all rocks with the L(t) of 1000 simulations of a CSR process with the same number of rocks (948) within the 18-m2 area. To test hypothesis 1 (spatial pattern of all colonized rocks is random), we compared the observed L(t) of the 49 colonized rocks with the L(t) of 1000 simulations of a CSR process with 49 rocks. We also looked at the cumulative distribution function of nearest-neighbor distances (the G function) between colonized rocks [Ripley, 1988] and compared it to that from a CSR process. To examine hypothesis 2 (colonized rock pattern is not significantly different from the underlying rock pattern), we used point pattern analysis based on the L-function to test whether the spatial distribution of the colonized rocks (“marks”) is independent of the underlying quartz rock (“resource point”) pattern [Lancaster and Downes, 2004]. For this analysis, the pattern of colonized rocks was compared with a neutral model other than CSR, since LCC are always co-located with the background rocks. For this model, we considered rock locations to be fixed and randomly allocated 49 colonized marks to these locations (RAN), simulating this process 1000 times. This allowed us to statistically compare the observed colonization pattern, L(t)OBS, with the L(t) of the neutral model, L(t)RAN. The null hypothesis, that colonized rocks are a random selection of those available, is indicated by L(t)OBSL(t)RAN = 0 with 99% confidence envelope around the difference simulated by Monte Carlo permutations, as described above. If L(t)OBSL(t)RAN > 0 (i.e., the mean difference and its confidence envelope lie above the null hypothesis of no difference), the colonized rocks are more clumped than the underlying pattern. Because there were no rocks outside the plot boundaries, edge corrections for L(t) and G(t) were not appropriate [Lancaster and Downes, 2004] and were not used. Statistical calculations and simulations were done using the “spatial” library [Venables and Ripley, 2004] and the “spatstat” library [Baddeley and Turner, 2005] in R.

4. Results

4.1. Site Climate and LCC Abundance (Overview)

[18] LCC abundance in the Atacama's hyperarid core ranges from 0.08 to 0.33% (Yungay and AC, respectively), whereas it reaches ≥6.7% in wetter sites south of the transition zone, such as Altamira. LCC organic carbon residence times also differ between sites, with turnover times at Yungay and AC from 3200 to 650 yr, respectively—reflecting extremely low levels of biological activity over geologic time scales. These results contrast with the 180-yr steady state residence time for hypolithic soils at Altamira [Warren-Rhodes et al., 2006] and underscore the rarity and long-lived nature of certain photosynthetic communities in the Atacama core.

[19] Site-level variations in climate, primarily rainfall, explain the observed differences in abundance and organic carbon residence times [Warren-Rhodes et al., 2006]. A detailed view of historical and recent in-situ climate data at one of the Atacama study sites, AC, is shown in Tables 1 and 2. (For a balanced comparison of the Atacama study sites, the reader is referred to Warren-Rhodes et al. [2006].) Long-term MAP at AC is ∼5 mm (versus ∼2 and ∼10 mm at Yungay and Altamira, respectively), with little to no rainfall occurring during most years. In wet years, Hobo® data (RHS ≥ 95%) indicate rainfall provides ∼400 hrs y−1 LWS at AC. A single moderate rainfall event (∼5 mm) can provide liquid water to the LCC environment (2–5 cm) for ∼6 days, with a return to pre-rain baseline conditions after ∼1 month.

Table 1. Environmental Parameters at the Aguas Calientes Study Sitea
  • a

    MAP = mean annual precipitation; MAT = mean annual temperature; For all parameters, T = temperature (°C), A = air, and S = soil. All figures Jan. 2001-Dec. 2004, unless otherwise noted below. MAP data: Aguas Verdes station (Departmento de Hidrología Subdepto. de Meteorología y Nieve D.G.A., Antofagasta), 1988–2003. LWR: liquid water from fog or dew (excludes rainfall): Nov. 2001–Oct. 2002. This figure reflects a maximum estimate for total for 1 year. LWS: liquid water at 2–5 cm from rainfall: 423 (July 2002) and zero for 2003, 2004 and Jan. 2005; maximum estimate for total for 1 “wet” year. Metabolic hours: liquid water from rainfall and fog or dew during daylight hours and TA > −10°C; Nov. 2001–Oct. 2003.

MAP (mm)4.7
mean annual RHA (%)16.8
MATA (summer)20.9
MATA (winter)14.3
maximum annual TA37.7
minimum annual TA−5.8
Soil surface and below
LWR (surface grid sensor >0.005V), no. hrs yr−1112–181
   LWR hrs when TA < 02.5
   LWR mean summer TA21.8
   LWR mean winter TA5.3
LWS (soil RHS at 2–5 cm ≥ 95%), no. hrs yr−1423
   LWS hrs when TS < 016
   mean TS12.0
total metabolic hrs yr−1204
mean annual RHS (2–5cm)14.3
mean annual RHS (10cm)16.0
MATS (2–5 cm)18.6
MATS (10 cm)19.4
MATS (2–5 cm, summer)22.7
MATS (10 cm, summer)23.4
MATS (2–5 cm, winter)13.9
MATS (10 cm, winter)14.7
maximum TS (2–5 cm)42.5
minimum TS (2–5 cm)−5.8
Table 2. Historical Rainfall Data (mm) for the Aguas Verdes DGA Meteorological Station (S25°24′00, W70°00′00, ∼25 km South of Aguas Calientes Site)a
  • a

    For years with data not collected (nc) for ≥ 3 months, the mean was not calculated (na = not available).


[20] Of the three sites, fog frequency at AC is lowest (Figure 1) owing to the barrier effect of the Pacific coastal range and AC's high altitude (1975 m). LWR data indicate fog may contribute a maximum of ≤200 hrs yr−1 liquid water to surface or near-surface communities at AC [Warren-Rhodes et al., 2006]. Data from conductivity, Hobo® and moisture sensing grids indicate that ∼15–30% of this surface moisture may reach near-surface communities.

Figure 1.

(a) Fog frequency as measured by a Spectrum leaf wetness grid at Aguas Calientes (S25°18′, W69°50′) in the Atacama Desert hyperarid core from February 2004 to December 2005. The sensor is dry at voltage, V = 2.095, liquid water droplets on the grid produce V ≤ 2.095, and thin water films are present at ∼V ≤ 0.1; (b) Fog frequency at Yungay in the Atacama's hyperarid core (S24°06′, W70°01′); (c) Fog frequency at Altamira, Atacama Desert, Chile. The high frequency of fog at Altamira is due to its location due east of a major gap in the coastal range (S25°45′, W70°11′). Rainfall events occurred April 24–25, 2005 at all three sites and also July 15 and August 31, 2005 at Altamira.

4.2. LCC Spatial Pattern: Results From Previous Studies

[21] Transect spatial analyses could not be undertaken for either Yungay or AC, since only a single transect per site contained colonized rocks. However, site-level index of dispersion analyses for Altamira showed that LCC spatial pattern is significantly aggregated (Figure 2), or “patchy,” at 1-m (ID = 1.4551, χ2 test p < 0.0001), 5-m (ID = 2.5171, χ2 test p < 0.0001), 10-m (ID = 3.1129, χ2 test p < 0.0001) and 50-m scales (ID = 3.8756, χ2 test p < 0.0001). These spatial patterns are similar to the aggregated spatial distributions (or “island patches”) shown elsewhere for both desert plants and soil crusts [Belnap et al., 2005; Ludwig et al., 2005]. At Altamira, LCC “island patches” were comprised of 1–6 colonized rocks separated by large (relative to the organisms, mean = 4.6 ± 0.6 m) interspace areas of uncolonized (by LCC) rocks and soil.

Figure 2.

The location (black square) of colonized rocks (i.e., a patch) at Altamira within 1-m2 contiguous numbered quadrats along each horizontal 50-m belt transect (A-O) and 25-m transect (P-T). Numbers in the black square indicate >1 colonized rock in the quadrat. Transects A-O were randomly placed in the field. For transects P-T, two 25-m transects (e.g., P1 and P2) were placed side-by-side.

4.3. AC Quartz Rock Cluster Distribution and LCC Abundance: Previous Studies

[22] Initial surveys (2002–2003) mapped and examined the ∼27 quartz rock clusters existing within the AC study area (Figure 3a). Rock density per cluster was 4–67 rocks/m2, mean cluster area was 100 ± 25 (SE) m2, and mean distance between clusters was 42 ± 5 (SE) m. Most clusters differed in altitude by only a few m, with the exception of four clusters located on the top and slopes of small hills (1974 ± 3 m). Of the 27 clusters, 7 contained LCC, with roughly 1–5 colonized rocks per cluster. Three colonized clusters were located on the hilltops and slopes, while four occurred in local small depressions (1960 m), suggesting that topography is an important factor for rock cluster habitability.

Figure 3.

(a) Gentle rolling terrain of small hills and depressions within the AC study site. Quartz rock cluster # 2 is visible in the left foreground; (b) Photograph of high-resolution quartz rock study area at Aguas Calientes (AC2). Single 1-m2 quadrat (#3, see Table 3) is visible in foreground. Black numbers in bold indicate quadrat number.

4.4. LCC Island-Patch Ecology: 2006 Study

4.4.1. Patch Abundance

[23] In 2006, a high-resolution study of LCC spatial pattern at AC2 was undertaken. Abundance and spatial pattern analyses were limited to a single 3 m-by-6 m quartz rock cluster (Figure 3b) owing to the significant age of AC communities and because prior studies had already disturbed and removed substantial biomass from rare and critical habitat within the site. Using the methods described above, 49 of 948 (5.2%) quartz rocks in the AC2 cluster were found to contain LCC, which represents an ∼15.5 times increase over AC site-level (2002–2003) abundance. Colonized rocks formed a small, 2-m2 patch within the 18-m2 rock cluster, with uncolonized rocks surrounding and interspersed within the colonized patch (Figure 4 and Table 3). Five (of 18) quadrats contained LCC, with mean quadrat colonization (colonized rocks in quadrat/total quartz rocks in quadrat × 100) of 11.8 ± 1.7% (SE).

Figure 4.

Map of quartz rock locations in the 6 m-by-3 m quadrat at AC2. Circles are proportional to rock size (which ranged from 5 to 32 cm and overlapped considerably). Grey circles represent rocks with cyanobacterial communities. Due south is in the positive y-direction.

Table 3. Ecological Parameters for Each of 18, 1-m2 Quadrats at Aguas Calientes
1. 1-m2Quadrat Location
2. Total # Quartz Rocks/Quadrat
3. Total # Rocks Colonized by LCC/Quadrat
4. Mean Rock Size (Mean Colonized Rock Size)/Quadrat
5. Total # Rocks > 11 cm

4.4.2. Rock Size and Density

[24] Colonized rocks ranged from 5–28 cm [mean = 13.3 ± 0.9 (SE) cm] and were much larger than the overall mean rock size of the cluster [7.1 ± 0.1 (SE) cm]. No relationship was found between LCC abundance and overall rock numbers (i.e., #rocks/quadrat as proxy for habitat availability, Table 3), but higher abundance was correlated with the availability of large rocks within each quadrat (R2 = 0.650). Indeed, the LCC patch was centered within the single quadrat having the highest number of large rocks in the cluster (Table 3). Higher LCC abundance was observed in individual quadrats with more closely spaced rocks, although the difference was marginal [colonized distance (cm) median = 3.0, mean = 3.5, max = 8.0; un-colonized distance (cm) median = 3.6, mean = 3.7, max = 8.6].

4.4.3. LCC Type and Depth of Colonized Zone

[25] An examination of LCC orientation showed 17% were hypolithic (beneath rocks), 60% perilithic (defined here as LCC strictly on rock sides, from “peri”, fr. Gk, around, in excess; surrounding), 9% chasmolithic (occupying rock cracks only), and 15% mixed. Among all sites in the Atacama, periliths are the predominant form [Warren-Rhodes et al., 2006]. In contrast to findings from the Mojave Desert [Schlesinger et al., 2003] that suggest hypoliths are restricted to rocks <2.5 cm thick [measured as maximum thickness of rock above colonized area], hypoliths at AC2 were observed on rocks up to 5 cm thick, while periliths colonized rocks up to 14.5 cm thick. Excluding chasmoliths, mean rock thickness to LCC colonization was 5.0 ± 0.5 cm—greatly exceeding the 1.5 cm described for the Mojave Desert. The depth of the LCC colonized zone below the soil/rock interface was consistent within an individual site [meanAC = 1.1 ± 0.04 (SE) cm; meanALTAMIRA = 0.25 ± 0.06 (SE) cm].

4.4.4. LCC Orientation and Soil Stabilization

[26] Nearly 70% of colonized rocks contained LCC on either east and/or southeast rock faces (Figure 5), although LCC were often found on multiple aspects. These orientations did not appear to conform to (1) prevailing wind direction, (2) optimal light and moisture conditions (e.g., early morning fog), or (3) likely wind dispersal patterns (e.g., Goossens, 2006, for fine sediments).

Figure 5.

Side of rock face colonized by LCC (denoted as x for each instance located on particular rock aspect). Prevailing winds are light and cool in the early morning from the east-southeast towards the Pacific coast, with a shift in wind direction (from the coast west-southwest) and an increase in wind speed in the afternoon. Cool winds and infrequent fog occur towards late evening, flowing from the coast west-southwest.

[27] In-situ visual observations showed that 60% of colonized rocks were embedded in highly “cemented”, or aggregated, soil surfaces, with few loosely situated rocks supporting LCC. This soil cohesion observed in the field was confirmed in the laboratory by light and scanning electron microscopy (SEM), which showed a thick gel-like matrix associated with colonized areas, in which cells are embedded. These extracellular polymeric substances (EPS) extend into the adhered soils around colonized areas but are absent from non-colonized surfaces of the same rocks [Pointing et al., 2007]. EPS has been postulated in cyanobacterial communities to ward against desiccation [Grilli Caiola et al., 1993; Allen, 1997; Belnap et al., 2005].

4.5. LCC Spatial Pattern

[28] Ripley's K analysis shows that both the underlying rocks (Figure 6a) and the colonized rocks (Figure 6b) within the 18-m2 study area are spatially clumped (L(t) > t) at all scales examined, rejecting hypothesis 1 of a random colonized rock pattern. Nearest neighbor distances between LCC are generally short and consistent (Figure 7a, median = 8.2 cm, max = 80 cm, inter-quartile range = 7.7 cm). G-statistic results (Figure 7b) reveal that nearest neighbor distances between colonized rocks are smaller than would be expected in a completely random pattern, indicating a clumped pattern. These findings are similar to the patchiness found at larger scales for transect data at Altamira and demonstrate that LCC spatial aggregation occurs at multiple hierarchical scales (<1 cm to 50 m). Further, as the point pattern results in Figure 6c clearly show, the LCC colonized rock pattern at AC2 is spatially aggregated relative to (i.e., distinct and more clumped than) the underlying quartz rock pattern—rejecting hypothesis 2 that LCC distribution is explained by the background rock pattern. Thus, the point pattern analyses demonstrate that quartz rocks are not “selected” or occupied by LCC randomly, but that non-random processes distinct from those creating the background rock pattern must be invoked to explain LCC patchy distribution.

Figure 6.

Point pattern analysis without edge correction of (a) all rocks and (b) rocks colonized by LCC at AC2. Solid line indicates the mean difference between the observed L-function and the L-function under CSR, dotted lines indicate 99% confidence envelope for this difference, and dashed line indicates null hypothesis of no difference. (c) Spatial pattern of rocks with LCC compared with the underlying distribution of all rocks. Lines as in (a) and (b) except the difference between the observed L-function and the L-function of a random selection of rocks is plotted. For the AC2 site, both the underlying rocks (a) and the colonized rocks (b) are spatially aggregated at all scales (i.e., difference in L-functions is positive), and the colonized rock pattern is significantly different (c, more clumped since L-difference is positive) than the underlying rock pattern at all scales.

Figure 7.

(a) Histogram of nearest neighbor distances between colonized rocks at the AC2 site. (b) Cumulative distribution function, G(t), of the observed nearest neighbor distances (solid line) compared with maximum and minimum cdfs from 1000 simulations of a CSR process (dashed lines). Because the solid line falls far outside the 99% confidence envelope described by the dashed lines for most distances, the observed pattern can be said to be significantly different than random.

5. Discussion

[29] This work shows that photosynthetic communities inhabiting desert pavement in the Atacama occur as discrete “island patches” [Belnap et al., 2005] whose non-random spatial distributions are independent and distinct from the background quartz rock pattern. Most LCC in the Atacama sites surveyed are perilithic and are located on particular rock aspects. Colonized rocks tend to be larger than the mean available rock size and well-cemented within the local soil matrix. This latter feature is a well-known characteristic of LCC and biological soil crusts in other deserts. In these communities, cyanobacteria exude EPS, which enhances soil particle aggregation and leads to soil stabilization in deserts [Bertocchi et al., 1990; Allen, 1997; Mazor et al., 1996; Philippis and Vincenzini, 1998]. For this role in soil aggregation, LCC, like desert crusts and plants, can be considered as ecosystem engineers [Jones et al., 1997; Rietkerk et al., 2004; Bowker et al., 2006].

5.1. Perilithic Phenomenon and Its Distinction From Hypolithic Growth

[30] Friedmann et al. [1967] and Friedmann and Galun [1974] characterized desert cyanobacterial habitats as hypolithic (colonizing lower surface of translucent rocks partially buried in soil), chasmolithic (colonizing rock fissures) and endolithic (colonizing rock fabric) environments. Based on our Atacama data, we find it useful to introduce a new term that further distinguishes hypolithic communities (i.e., those colonizing the bottom of rocks) from those occupying a distinct zone directly below the rock/soil surface interface—which we term periliths. In contrast to other deserts, such as the Negev and the Mojave, in our Atacama sites periliths comprise the majority of LCC.

[31] While not always mutually exclusive (i.e., hypolithic and perilithic colonization can occur on a single rock), the perilithic phenomenon is important to an understanding of LCC ecology under hyperarid conditions, since it sheds several new insights into the relationships between LCC and key environmental constraints such as light intensity and moisture availability. First, data from our hyperarid sites in the Atacama indicate that the average depth of the cyanobacterial colonized zone below the rock/soil interface, d (Figure 8), varies between sites but is consistent within a single site, such that d is independent of r + d [n = 32, R2 = 0.299]. At Altamira, the depth of the colonized zone (d,Figure 8) was 0.25 ± 0.06 cm (n = 44), whereas at AC2 it was 1.1 ± 0.04 cm (n = 35). We think this difference is not only attributable to light intensity but also reflects variations in moisture source and availability. At Altamira, fog is a predominant source of moisture that generally tends to wet only the shallow upper soil surface [Warren-Rhodes et al., 2006]. Because LCC would need to be as close to the soil surface as possible to exploit fog condensation, this may explain the relatively shallow mean colonization depth at Altamira. In contrast, rainfall tends to wet soils more deeply—and be retained longer at deeper versus shallower depths, where rapid drying occurs. Therefore, in hyperarid sites such as AC2, where rainfall (albeit slight) is the main moisture source, the mean colonization depth is larger.

Figure 8.

Schematic illustration of translucent rock colonized by LCC. Shown is a “perilithic” colonized zone (black bar) located at a vertical depth, d, from the soil/rock interface (shown as dotted line); L is the rock maximum length (longest axis); H is the maximum rock height or total thickness; r is the maximum thickness of the rock above the soil surface/rock interface; r + d is the maximum rock thickness above the colonized zone; and t is the thickness of the colonized zone. The rock “sail” or “canopy” is defined as the rock surface area (shaded in grey) protruding above the soil/rock interface.

[32] Second, we have shown that within hyperarid deserts LCC disproportionately inhabit large rocks; small rocks (<5 cm)—although frequently colonized in wetter deserts—remain virtually uncolonized [Warren-Rhodes et al., 2006, 2007b]. In such hyperarid regions, rainfall and fog are typically small events of short duration (Table 2) [McKay et al., 2003; Warren-Rhodes et al., 2007a]. Large rocks, which protrude well above the soil surface—the area of which we define as a rock's “sail” or “canopy” (Figure 8, shaded area)—serve as better moisture collectors. The surface area of the rock sail scales approximately as r2, where r is the maximum radius of the rock sail protruding above the soil surface (Figure 8). Fog or rainfall collected by the sail is distributed over the rock perimeter, which is proportional to r. Thus the ratio of collection area to perimeter increases linearly with increasing r. Larger r would theoretically collect and distribute more water to rock sides, and thus confer an advantage to LCC survival in extremely dry conditions. This benefit, along with other effects such as a moderating thermal regime that may also facilitate condensation and reduce overall water loss [Mehuys et al., 1975; Warren-Rhodes et al., 2007b], would explain the preferential colonization of large rocks we have measured in the Atacama and deserts in northwest China [Warren-Rhodes et al., 2007b]. In contrast, in relatively wet desert environments, where copious rain would enable soil surfaces to reach field capacity (e.g., Mojave site in Schlesinger et al., 2003, where 100% of rocks were colonized by hypoliths), the amount of water in soils adjacent to the rock surface will be independent of rock size.

[33] Our data for the Atacama and other hyperarid deserts indicate that rock size (maximum length, L, measured as longest horizontal axis, Figure 8), rock height/thickness (H, measured as maximum vertical distance from rock top to bottom, Figure 8), colonization type (hypolithic, perilithic, chasmolithic, and endolithic), depth of the colonized zone/band (r and d, measured as maximum vertical distance from top of rock to soil surface/rock interface and maximum vertical distance from soil/rock interface to top of colonized zone, respectively, Figure 8) and thickness of the colonized zone/band (t, Figure 8) are important variables potentially signaling the effects of environmental constraints on overall quartz rock habitability and LCC ecology. As such, each of these parameters should be defined and measured consistently for systematic comparison in future studies.

5.2. LCC Spatial Patterns at Multiple Scales and Possible Explanations

[34] LCC in the Atacama Desert are patchily distributed (cm to 50 m scales) in patterns distinct from the underlying rock pattern. These findings complement extensive studies in China also showing LCC aggregation over multiple scales [Warren-Rhodes et al., 2007b] and previous work for other organisms demonstrating non-random spatial distributions, including soil bacteria [Nunan et al., 2002; Lancaster, 2006; Hughes et al., 2006].

[35] The Atacama results suggest that non-random factors and processes influence LCC spatial pattern, with the simplest explanation for the observed patchiness being water availability. Soil moisture, and the physical and biological factors that determine its heterogeneous distribution, have repeatedly been shown in landscape ecology and ecohydrology studies to control plant and soil crust distributions in deserts and arid lands [Ludwig and Tongway, 1995; Ludwig et al., 1997; Fernandez-Illescas et al., 2001; Lookingbill and Urban, 2004; Belnap et al., 2005; Bowker et al., 2005, 2006]. Below, we examine these factors in light of our initial Atacama findings.

5.2.1. Physical Factors That Affect Rock Habitability and LCC Spatial Pattern

[36] Physical factors that reduce or enhance water availability in deserts place limits on habitability and, therefore, influence LCC spatial pattern. Based on the aforementioned literature, three classes of physical variables most strongly affect soil moisture patterns in deserts: (1) topography, (2) rock/soil properties and (3) climate (light, temperature, moisture source) [Nienow et al., 1988a, 1988b; Friedmann et al., 1993; Fernandez-Illescas et al., 2001; Barrett et al., 2004; Lookingbill and Urban, 2004; Belnap et al., 2005]. Initial Atacama data highlight the importance of topography to LCC ecology. At AC, macroscale (tens of m vertical relief) topography was linked to higher LCC abundance, with colonization of particular rock clusters largely occurring in topographic highs and lows. This finding suggests that areas conducive to fog deposition (hilltops and their slopes) and/or receiving greater runoff from rainfall (depressions) are most likely to support LCC, with these communities' spatial patterns reflecting this heterogeneity in water distribution. Slope aspect has been shown to affect plant cover and shrub-clump size in the matorral of central Chile [Badano et al., 2005], and plant communities in certain coastal fog and inland hyperarid areas in the Atacama Desert exhibit spatial patterns similar to those for LCC, being confined to washes and depressions that receive runoff and/or hillsides and summits that intercept fog [Larrain et al., 2002; Latorre et al., 2003; Ewing et al., 2006].

[37] Microtopography has also previously been shown to influence plant, lichen and desert soil crust spatial pattern [George et al., 2000; Bowker et al., 2005; Kuntz and Larson, 2006; Lalley and Viles, 2006; Yao et al., 2006], and there is evidence at AC2 for microtopographical effects on LCC spatial pattern. Within the AC2 cluster, 90% of colonized rocks were situated in a microtopographic high (cm-scale vertical relief) where moisture availability from fog is likely greatest (Figure 9). That LCC spatial patterns at AC may reflect non-random fog distribution at both macroscales and microscales was unexpected given this site's relatively low fog frequency.

Figure 9.

Contour map showing local slopes within 18-m2 study area (length = 600 cm, width = 300 cm; large numbers in bold indicate quadrat number). Contour interval is 1 cm, with highest elevation shown in cyan (0 to –1 cm relative elevation) to lowest (lower right) in black (−12.5 cm). Due south is in positive y-direction (see also Figure 3b). A majority of the colonized rocks (shown as stars) in the LCC patch are located in a microtopographical high.

[38] In addition to topography, rock parameters, such as rock size, thickness, spacing, and orientation, also appear to be important drivers of LCC spatial pattern. At AC2, habitability on individual rock microaspects was not equal, with LCC colonization often (but not exclusively) on east-southeast faces. Prevailing winds may dictate this orientation, but diurnal shifts in wind direction at AC make definitive linkages to LCC pattern problematic. Alternatively, LCC orientation may reflect favorable microclimates for photosynthesis during early morning, when light, temperature and moisture from fog are optimal. This latter explanation would agree with other studies showing that (1) desert soil crust organisms favor certain microaspects that provide less xeric microclimates [George et al., 2000; Bowker et al., 2005; Kuntz and Larson, 2006] and (2) endoliths colonize microaspects with favorable temperature, light and moisture conditions [Friedmann et al., 1987; McKay and Friedmann, 1987; Nienow et al., 1988a, 1988b; Friedmann et al., 1993].

[39] Another factor governing LCC spatial pattern is rock size, with LCC at AC overwhelmingly concentrated on large rocks. Indeed, the LCC patch at AC2 was centered within the single quadrat with the highest number of rocks >11 cm. These findings concur not only with previous data from other LCC studies in the Atacama and China [Warren-Rhodes et al., 2006, 2007b] but also for other organisms, such as lichens, whose distributions are influenced by rock size [Danin et al., 1982]. For LCC, the effect of rock size is likely, as described for certain aspects above, related to the greater relative efficiency of larger rocks to redistribute (blocking lateral surface flow), collect and retain water [Mehuys et al., 1975; Ludwig and Tongway, 1997; Belnap et al., 2005].

[40] Clearly, the underlying link between the above factors influencing LCC pattern is water availability, and more specifically, the physical factors that dictate its distribution and retention in extremely dry environments. In previous studies in the Atacama, we showed that mean annual rainfall explains site-level LCC abundance [Warren-Rhodes et al., 2006]. In this work, the importance to LCC colonization of landscape- (i.e., within site) and individual rock-scale physical factors that affect moisture availability is highlighted. To further elucidate this, detailed climate monitoring and modeling will be necessary. In Antarctica, results from such studies support our initial Atacama data and interpretations, in which rock physical properties and surface orientation produce nano-climate differences that control cyanobacterial community distribution, but with temperature rather than moisture as the limiting factor [McKay and Friedmann, 1987, Friedmann et al., 1987; Nienow et al., 1988a, 1988b; Friedmann et al., 1993; Broady, 1996].

5.2.2. LCC Dispersal Processes and Spatial Pattern

[41] Physical factors that determine habitability are central to an understanding of LCC abundance and distribution, but they do not represent the actual mechanisms responsible for LCC island-patch formation. For this, knowledge of dispersal is fundamental, but to date relatively sparse information on this critical aspect of LCC ecology exists. In the Atacama, LCC spatial data show a distinctly non-random pattern where the probability of colonization increases as the distance to an adjacent colonized rock decreases. This pattern is likely a consequence of, and thus provides insights into, the possible mechanics of LCC dispersal.

[42] Past work has generally considered LCC dispersal to occur by wind [Friedmann and Galun, 1974; Broady, 1996]. In Antarctica, for example, cyanobacteria are dispersed by strong katabatic winds from terrestrial mats to sediments on permanent ice cover [Gordon et al., 2000], and chasmolithic colonization patterns have been attributed to prevailing winds [Broady, 1981]. Similarly, at AC2, certain rock aspects are disproportionately exploited by LCC, and this pattern may be, at least in part, a function of predominant wind direction. While wind and other atmospheric forces (e.g., dust storms) traditionally have been argued to produce ubiquitous and random microbial dispersal [Fenchal, 2002; Green and Bohannan, 2006], recent models for aeolian dust deposition across cm- (e.g., quartz rocks) to hundred m-scale (e.g., hillslopes) landforms [Goossens, 2006] suggest that wind dispersal might also generate non-random LCC distributions (e.g., windward sides if bacteria are rafted to soil particles ≥10 μm). Predicting dispersal patterns with aerobiology models [Lighthart and Mohr, 1994; Goossens, 2006], however, remains challenging due to uncertainties in bacterial aerodynamic diameters [Lighthart, 1994], yet is a promising area of future research.

[43] Another potential dispersal mechanism contributing to non-random LCC spatial pattern is liquid water events. Water as a primary pathway for the dispersal of soil biota and connectivity of island patches has been postulated for LCC [Cockell and Stokes, 2004] and biological soil crusts [Belnap et al., 2005] and shown for the movement of cyanobacteria in desert soils [Pringault and Garcia-Pichel, 2004]. Short-range cell movements and spatial structure of biota in soils appears to be common [Grundmann, 2004], with the mm and μm spatial scales especially relevant [Nunan et al., 2003; Grundmann, 2004].

[44] For LCC, propagation via liquid water enables them to expand along rock surfaces. During liquid water events, EPS surrounding LCC swell rapidly (to several times the original volume) and produce forces that propel cells from the EPS matrix [Bell, 1993; Allen, 1997]. In this way, LCC are dispersed across rock surfaces and to soils at the rock/soil interface. Short-range (mm) LCC dispersal could occur during very light rainfall (few mm) and/or fog events, which may wet and infiltrate soils but do not likely generate runoff. Desiccated airborne LCC may also be transported during fog events, with their spatial patterns reflecting fog deposition patterns on intercepting rocks.

[45] Larger (5–40 mm), infrequent rainfall events may facilitate LCC movement from rock to nearby (cm) rock. Our nano-climate monitoring shows that such events saturate surface soils and deliver liquid water to the hypolithic environment for days to weeks. Large rainfall events likely generate runoff and probably explain the location of colonized clusters at AC in local depressions. Such events occurred recently at AC in May 2002 (5 mm), April 2005 (18 mm) and August 2006 (31 mm). The 2005 storm produced visible and significant runoff in the region (B. Gómez-Silva and J. Owen, pers. obsv.).

[46] The cumulative effect over geologic timescales (rocks at AC and Yungay have been colonized a minimum of 650–3200 years [Warren-Rhodes et al., 2006]) of infrequent rainfall events could produce the aggregated spatial pattern mapped at AC2, with LCC dispersing from nascent colonized rocks to inoculate new, nearby rocks with each event. The stochastic nature of rainfall at AC (Table 2) could produce both the tight, confined pattern of the main LCC patch (small events) as well as the outliers (large events) observed. Rainfall dispersal enables a simple explanation for the observed spatial pattern at AC2 and is reflected in the decreasing colonization probability with increasing distance from a colonized neighbor.

[47] From the discussion above, a number of possible dispersal and colonization scenarios may underlie the spatial patterns mapped: (1) initial random dispersal to a single rock, followed by subsequent colonization in non-random spatial niches—a hypothesis that might include any combination of the mechanisms described (wind, fog and rainfall); (2) non-random dispersal by air (wind or fog); and/or (3) non-random dispersal via water through the soil matrix (rainfall, fog). The first scenario is a more traditional view of eukaryotic and prokarytic dispersal [Fenchal, 2002; Finlay, 2002]. Recently, however, non-random patterns and explanations for bacterial distributions and diversity have been gaining ground [Papke et al., 2003; Whitaker et al., 2003; Green et al., 2004; Hughes et al., 2006]. The Atacama data raise more questions than can be answered by our current study. Laboratory and field studies that track and model LCC dispersal are a next step, along with detailed comparisons of climate, topography and geophysical properties across (1) individual rocks in a single patch; (2) patches within a site; (3) sites in a single desert; and (4) deserts with varying climate and geology.

6. Conclusions

[48] Characterizing the spatial distribution of organisms is of fundamental importance to ecology [Nunan et al., 2002, 2003; Grundmann, 2004; Lancaster and Downes, 2004; Wagner and Fortin, 2005; Lancaster, 2006; Green and Bohannan, 2006]. As is the case for higher plants, spatial patterns can illuminate dispersal and survival mechanisms in microbial extreme environments. Analyses in this paper reveal that photosynthetic communities inhabiting quartz rock pavement in the Atacama Desert are not distributed randomly throughout the landscape but occur as discrete island-patches at multiple hierarchical scales. In these patterns, we have likely detected the imprint and influence on LCC colonization of both non-random abiotic factors—i.e., those that control water availability, and thus habitability—and biotic processes. Aggregated spatial patterns raise intriguing possibilities regarding LCC dispersal, and their role as ecosystem engineers [Jones et al., 1997]. Exciting directions in future work are molecular biogeography and self-organization [e.g., Rietkerk et al., 2004]. Beyond theoretical ecology, understanding the drivers of bacterial spatial pattern in the Atacama Desert is a necessary step towards modeling life's predictability in extremely dry environments. Predictability, in turn, equates to greater efficiency in the search for life, either across the Earth's deserts or within Mars' expansive surface and subsurface realms.


[49] The authors thank Nicanor Plaza and Juan Rojas at the Instituto del Desierto, Universidad de Antofagasta, Antofagasta, Chile, for logistical assistance. Precipitation data were generously provided by the Dirección General de AguasII Región ( and the Dirección Metereológica de Chile ( We also gratefully acknowledge the insights and perilithic terminology suggested by K. L. Rhodes and constructive comments from two anonymous reviewers and Jill Lancaster.