OUP user menu

Cyanobacterial ecology across environmental gradients and spatial scales in China's hot and cold deserts

Kimberley A. Warren-Rhodes, Rhodes Kevin L., Linda Ng Boyle, Stephen B. Pointing, Yong Chen, Shuangjiang Liu, Peijin Zhuo, Christopher P. McKay
DOI: http://dx.doi.org/10.1111/j.1574-6941.2007.00351.x 470-482 First published online: 1 September 2007


Lithic photoautotrophic communities function as principal primary producers in the world's driest deserts, yet many aspects of their ecology remain unknown. This is particularly true for Asia, where some of the Earth's oldest and driest deserts occur. Using methods derived from plant landscape ecology, we measured the abundance and spatial distribution of cyanobacterial colonization on quartz stony pavement across environmental gradients of rainfall and temperature in the isolated Taklimakan and Qaidam Basin deserts of western China. Colonization within available habitat ranged from 0.37±0.16% to 12.6±1.8%, with cold dry desert sites exhibiting the lowest abundance. Variation between sites was most strongly correlated with moisture-related variables and was independent of substrate availability. Cyanobacterial communities were spatially aggregated at multiple scales in patterns distinct from the underlying rock pattern. Site-level differences in cyanobacterial spatial pattern (e.g. mean inter-patch distance) were linked with rainfall, whereas patchiness within sites was correlated with local geology (greater colonization frequency of large rocks) and biology (dispersal during rainfall). We suggest that cyanobacterial patchiness may also in part be self-organized – that is, an outcome of soil water-biological feedbacks. We propose that landscape ecology concepts and models linking desert vegetation, biological feedbacks and ecohydrological processes are applicable to microbial communities.

  • hypolithic
  • photoautotrophs
  • hyperarid desert
  • landscape ecology
  • patchiness
  • trigger-transfer-response-pulse framework


Stony desert pavements, composed of surface soils mantled by gravels and bedrock debris, are important microbial habitats in arid environments (Friedmann & Galun, 1974; Golubic et al., 1981). Translucent rocks support unique lithic bacterial communities that are often the principal primary producers in hyperarid deserts (Allen, 1997; Warren-Rhodes et al., 2006). These communities are dominated by an apparently ubiquitous cyanobacterium of the form-genus Chroococcidiopsis that is extremely resistant to ionizing radiation and desiccation (Billi et al., 2000), often in association with other filamentous cyanobacterial and heterotrophic taxa (Pointing et al., 2007). They are referred to as lithic (‘lithobiontic’, ‘lithophytic’, Table 1) cyanobacterial communities (LCC) to reflect the importance of the rock niche and photoautotrophic component. By inhabiting diaphanous rocks and minerals (e.g. quartz, granite, gypsum, halite, sandstone) these organisms gain protection from solar radiation and receive increased moisture relative to the bare soil (Vogel, 1955; Friedmann et al., 1967; Cockell & Stokes, 2004; Cockell et al., 2005).

View this table:

List of abbreviations

Lithic (lithophytic, lithobiontic) cyanobacterial communitiesLCC
Mean annual precipitationMAP
Mean annual air temperatureMATA
Air temperatureTA
Relative humidity (air)RHA
Soil temperatureTS
Relative humidity (soil)RHS
Soil liquid water (Hobo® logger reads RHS≥95%)LWS
Rock surface liquid water (Campbell grid reads Volts ≥0.005)LWR
China Meteorological AdministrationCMA
Scanning electron microscopySEM
Extracellular polymeric substancesEPS
Index of dispersionID

Many aspects of LCC ecology, including photosynthetic carbon fixation, nitrogen cycling, and biodiversity have been investigated in hot and cold arid and hyper-arid desert environments (Friedmann et al., 1967; Allen, 1997; Smith et al., 2000; Garcia-Pichel et al., 2001; Matthes et al., 2001; Schlesinger et al., 2003; Boison et al., 2004; Cockell & Stokes, 2006; Omelon et al., 2006; Warren-Rhodes et al., 2006). However, there is a limited understanding of their spatial distribution and the effects of moisture, temperature and substrate parameters (e.g. rock size) on their abundance and diversity. Moreover, consistent sampling and statistically rigorous design has largely been absent from past studies (Matthes et al., 2001), thus restricting comparisons between and an understanding of LCC and the physical environment, especially on larger scales. Schlesinger et al. (2003), for example, measured a single plot in the warm semi-arid Mojave Desert, which showed 100% of quartz rocks – a common desert pavement substrate – were colonized by hypoliths. This result was similar to that reported by Cockell & Stokes (2006) for opaque rocks in the cold, wet Arctic environment. In contrast, the warm arid Atacama supported 28% and the warm hyper-arid core of the Atacama <0.1% colonization (Warren-Rhodes et al., 2006). These studies provide an interesting glimpse into possible geographical patterns and highlight the need for robust statistical approaches and comparative ecological investigation. It is also tantalizing to hypothesize that proliferation of LCC in extremely arid conditions may involve biological feedback mechanisms to maximize use of scarce water resources, such as those demonstrated by higher plants in deserts (Rietkerk et al., 2004; Ludwig et al., 2005). While laboratory testing of some aspects of LCC ecology is possible, many ecological hypotheses in microbial systems require testing in ‘natural laboratories,’ where relatively few key variables exist and trophic interactions are simpler. The western deserts of China comprise swathes of contiguous quartz desert pavement spanning thousands of kilometers across wide climatic gradients, and as such provide an ideal region for investigating not only LCC ecology but also for examining issues of broader relevance to microbial ecology.

To more fully understand LCC ecology we carried out multiple ecological studies at four desert locations in western China. The sites all supported a predominantly quartz pavement substrate, thus minimizing variability, but varied in their level of hyperaridity by virtue of different long-term mean annual temperature and rainfall. Here we report both landscape-scale and small-scale abundance patterns and identify possible linkages between ecohydrological processes and LCC across multiple scales. We interpret our observations in terms of macro-ecological theory as applied to microbial distributions and biological feedbacks.

Materials and methods

Field locales and sampling

China's Northwest contains some of the oldest, driest, and most isolated deserts on Earth (Guo et al., 2002; Sun & Liu, 2006). Within the region, desert pavement occurs on the periphery of hyperarid basins and flanks most major mountain ranges (Fig. 1) (Walker, 1982). From preliminary surveys and precipitation data, we chose three locations (10–100km scale) for study (Fig. 1): (1) Tokesun, northern Turpan Depression; (2) Ruoqiang, southern Taklimakan Desert; and (3) Sorkuli (also Suoerkuli) Qaidam Basin, Qinghai-Tibetan Plateau (Table 2).


Map of China, with field locations highlighted.

View this table:

General site characteristics, China

Temperature category Moisture category Site/ParameterHotCold
Dry Tokesun, Turpan DepressionWet Ruoqiang, Taklimakan DesertDry Sorkuli 01, Qaidam BasinWet Sorkuli 03, Qaidam Basin
Elevation (m)119121728823440
MAP (mmyear−1)10.023.315.1N/A
MATA (°C)14.613.13.2−0.25
  • Mean annual precipitation (MAP) and mean annual temperature (MAT) are 13-year historical averages for Ruoqiang and Sorkuli from the China Meteorological Administration (CMA). Ruoqiang site data are from the Qiemo CMA station (∼30km away), and Sorkuli data are from the Lenghu CMA station (∼100km away). Tokesun site MAT data (13-year average) are from the Turpan CMA station (∼70km away), but Tokesun MAP data (10-year average) are from the local Tokesun meteorological bureau.

To determine the environmental influence on inter-site LCC abundance and diversity, sites (1–10km scale) within each location were chosen based on mean annual precipitation (MAP) and mean annual air temperature (MATA) to allow wet and dry, and hot and cold site comparisons (Table 2). Inter-site geological variation was minimized by choosing silicious desert pavement sites of similar soil and rock type, particularly quartz, which forms the bulk of LCC habitats in our study. Four main sampling sites were chosen within the three locations (Table 2).

Environmental variables

In situ environmental data, including air temperature (TA), relative humidity (RHA) (Onset Computer, Hobo PRO®) and rainfall (Onset Computer, RG2-M), were collected 8 August 2001 to 12 September 2002 (Warren-Rhodes et al., 2007). Soil temperature (TS) and pore-space relative humidity (RHS) dataloggers (Onset Computer, Hobo PRO®) were placed under rocks such that sensors were directly below visible hypolith communities (∼1–5cm below soil surface, depending on hypolith location) and 10cm (to monitor soil moisture retention). Soil liquid water (LWS) was considered present at RHS≥95% (Warren-Rhodes et al., 2007). The presence of rock surface liquid water (LWR) – a proxy for moisture available to chasmolithic communities – was measured with a moisture-sensing grid (Campbell Scientific, 237-L). Historical climate data are courtesy of the China Meteorological Administration (CMA).


Microscopic examination was carried out using light microscopy (BX50 compound microscope, Olympus, Tokyo, Japan) and scanning electron microscopy (SEM) (Stereoscan S440, Leica Cambridge, UK). Colonized quartz fragments were stored in a silica-gel desiccator for 48h prior to gold sputter-coating (SCD 005, BAL-TEC, Lichtenstein) and observation under low vacuum (to avoid artifacts generated during chemical fixation). Working depth for imaging was 20mm, and beam current 12.0kV.

Cultivation and phylogenetic analysis

A sterile scalpel was used to scrape a small amount of colonized quartz from each pebble into BG11 cyanobacterial growth medium and enrichment cultures were incubated at 25°C and 250μmolm−2s−1 photosynthetically active radiation (12-h light-dark cycle) with periodic subculture (Castenholz, 1988). In all samples a single cyanobacterial morphotype of the form-genus Chroococcidiopsis was the only taxon recovered and these were further characterized by 16S rRNA gene sequence. Sequencing was achieved using primers with Escherichia coli equivalent nucleotide positions of 27F-1492R as previously described (de la Torre et al., 2003). Amplicons were purified (GFX, Amersham, Bucks, UK) prior to automated sequencing using the BigDye Terminator Cycle Sequencing kit and ABI 3730 Genetic Analyzer (Applied Biosystems, CA). Approximate phylogenetic affiliations were determined by blast searches of the NCBI GenBank database (http://www.ncbi.nlm.nih.gov/). Multiple alignments were then created with reference to selected GenBank sequences using clustal× v.1.81 (Thompson et al., 1997). Maximum likelihood analysis using paup* 4.0b8 (Swofford, 2001) was used to illustrate the relationship of sequences to representative taxa. Bootstrap values for 1000 simulations were calculated and are shown for branch nodes supported by more than 50% of the trees. Sequences from this study have been deposited in the NCBI GenBank database under accession numbers DQ914863DQ914866.

Field measurements of colonization

Abundance was defined as percent colonization [hereafter, colonization=(# of colonized quartz rocks/total quartz rocks) × 100]. The term LCC or cyanobacteria hereafter refers to hypolithic (subsurfaces) or chasmolithic (rock cracks, crevices) forms, unless otherwise noted. Sampling design and field methods from plant and animal landscape ecology were used to measure LCC abundance and spatial pattern (Andrew & Mapstone, 1987; Krebs, 1989; Waite, 2000). A pilot study was conducted to optimize sampling unit shape, size and number of replicates (Krebs, 1989). As site remoteness precluded conducting pilot studies at all sites, Tokesun was chosen as a proxy. The results indicated that approximately six 1m × 50m transects were needed (Fig. 2) to obtain a representative site measure of abundance at the desired precision (0.25). However, sampling at other sites revealed subsequently that additional transects were needed to achieve the stated precision at all sites, such that 10 transects were used for all sites in the final study. The 10 transects of 50m2 were randomly distributed over a ∼0.5km2 area, and all quartz rocks were counted, measured (maximum length, nearest 0.5cm) and inspected visually for colonization. For spatial distribution, LCC location was recorded to nearest centimeter (x-axis along 50m transect). Underlying rock pattern was determined by photo-analyses of randomly selected 1m2 quadrats within each transect.


Mean and SE plotted as a cumulative abundance curve for the pilot transect study (Tokesun 2001, China). Rectangular belt transects were 1m × 50m. Results showed that five 1m by 50m transects were required to measure site abundance at the 0.25 precision level. Subsequent analysis revealed some sites required additional transects, such that 10 transects were used for the main study (2002). The figure demonstrates that standard plant and animal ecology methods to measure abundance can be applied to obtain statistically robust field measures of microbial abundance.

LCC climate, abundance and spatial distribution were measured across inter-site (≥1km) and intra-site (<1km) scales. The latter consisted of landscape (50–100s of m, inter-transect), area (1–50m, inter-quadrats) and rock scales (<1m to 1mm, intra-quadrat). The influence of quartz rock density (=habitat availability=quartz rocks per 50m2) and rock size on LCC abundance were measured at the transect level. The effects of rock size on colonization, along with soil moisture, temperature, LCC location and depth, were measured for individual rock scales.

Statistical analyses

anova (Devore, 2004; Box et al., 2005) was used to determine inter-site variation for percent colonization and rock density, with Pearson's correlation coefficients used to compare these parameters over multiple scales. Within-site rock size standard deviations were compared using F-tests. A two-tailed Z-test examined colonized rocks vs. overall rock proportions (dry and wet sites), with a post hoc Cochrane–Mantel–Haenszel (CMH) test used for inter-site differences in rock size. The index of dispersion (ID) tested LCC spatial randomness, with quadrat size and number chosen to ensure the validity of the χ2 test (Diggle, 2003). For sites with sufficient colonized rock samples, the scale(s) of aggregation was analyzed with the Three-Term Local Quadrat Variance (3TLQV) method (Dale, 1999). Site-level relationships between overall vs. colonized rock spatial distributions were tested with product-moment correlation analyses (Upton & Fingleton, 1985). Analyses of underlying rock patterns from photo-quadrats used two-dimensional (2-D) point pattern analyses, with calculation of Ripley's K and nearest neighbor functions and comparison with Monte Carlo simulation values (n=999) for a completely random (Poisson) process (Upton & Fingleton, 1985; Ripley, 1988; Cressie, 1993).

Logistic regression models were used to predict site-specific colonization likelihood, based on environmental variables such as moisture and temperature (Hosmer & Lemeshow, 2000). Odds ratios are computed from the coefficients of the model to assess the magnitude of the relationships. All tests were adjusted for unequal sample size and variance and transformed as needed to meet normality and heterogeneity of variance assumptions. Statistical significance was assessed at P≤0.05.


Environmental monitoring

The study sites are roughly divided by climate into hot (Tokesun and Ruoqiang) and cold (Sorkuli 01 and 03) deserts (Table 1 Supplementary). In situ TS indicated a minimum range of thermal tolerance for hypoliths of −23.8–53.8°C at the sites. Long-term MAP and soil moisture availability divided sites into dry (Tokesun, Sorkuli 01) (≤15mm MAP and ≤500hyear−1 of LWS) and wet (Ruoqiang, Sorkuli 03) (>15mm MAP and ≫500hyear−1 of LWS) categories (supplementary Table 1). Spring/summer rainfall was the sole source of LWS to hypoliths at all sites but Ruoqiang, where winter precipitation (e.g. snowmelt) also played a dominant role (supplementary Table 1). No dew or fog was recorded. Annual liquid water available to chasmoliths indicated a range of LWR from 297 to 1397h and 302 to 1839h for hypoliths (LWS). Roughly 60% of available water occurred under conditions suitable for photosynthesis (daylight, TA >−6°C) (200–922hyear−1, supplementary Table 1) (Warren-Rhodes et al., 2007).

LCC identification and colonization

Visual examination revealed all colonized rocks supported a single coccoid morphotype corresponding to Chroococcidiopsis, and the presence of an apparent extracellular polymeric substance (EPS) surrounding (either partially or fully) cells was indicated for all samples examined (Fig. 3), similar to observations for other cyanobacterial habitats (Allen, 1997; Philippis & Vincenzini, 1998; de los Ríos et al., 2004). Cracks that appeared in the EPS demonstrate its relatively thick nature, and the omission of any chemical fixing step precludes the possibility that EPS was an artifact of sample preparation. Phylogenetic analysis of the 16S rRNA gene for cultivated strains from the sites indicated that they affiliated into a clade of closely related phylotypes within the known desiccation-resistant (Grilli-Caiola et al., 1993) and arid environment inhabiting (Schlesinger et al., 2003; Warren-Rhodes et al., 2006; Pointing et al., 2007) genus Chroococcidiopsis (Fig. 4), thus concurring with visual observations. The morphological and phylogenetic data supports the validity of the field sampling as being focused upon the same types of assemblage dominated by this form-genus at each location.


(a) Scanning electron micrograph of colonized quartz illustrating extracellular polymeric substance associated with cells, scale bar=2μm. (b) Light microscopy image of Chroococcidiopsis cells from colonized quartz, scale bar=5μm.


Phylogenetic affiliation of Chroococcidiopsis from hyperarid desert locations in western China based upon maximum likelihood analysis of the complete 16S rRNA gene. Tree topology is supported by bootstrap values based upon 1000 simulations. Scale bar represents 0.1 nucleotide changes per position.

During 2001–2002, over 10000 quartz rocks and 3000m2 of desert pavement were investigated. Significant differences in percent colonization (Table 3) between all sites were observed (anova, df=3,36, F=46.7, P<0.0001), with abundance ranging from 0.37±0.16% (SE) at Sorkuli 01 to 12.6±1.8% at Ruoqiang. High spatial variability in abundance was observed not only across sites, but also within sites. For example, colonization measured within individual transects (i.e. 50m scale) at Tokesun 01 and Tokesun 02 ranged from 1% to 12%, although mean site colonization (n=10 transects) did not significantly differ for the two sites (∼10km apart) (anova, df=1,8, F=0.001, P=0.97). This observed high variability in abundance signals significant spatial patchiness in LCC distributions at multiple scales both within and across sites.

View this table:

Ecological results for the China desert sites (mean ± SE)

TokesunRuoqiangSorkuli 01Sorkuli 03
Percent colonization (abundance)0.99 ± 0.212.6 ± 1.80.37 ± 0.168.5 ± 1.6
Total no. of quartz rocks counted5198177932051802
Total no. of colonized rocks5021510142
Mean no. of quartz rocks/m210.3 ± 1.03.1 ± 0.36.4 ± 0.73.3 ± 0.4
Mean no. of colonized rocks per m20.10 ± 0.020.43 ± 0.050.02 ± 0.000.28 ± 0.05
Mean colonized rock size (cm)6.9 ± 0.45.6 ± 0.24.7 ± 0.63.2 ± 0.1
Percent of rocks ≤5cm79.693.696.595.5
Percent colonized hypolithic4674.810099.3
Percent colonized chasmolithic5425.200.7
  • Two sites within Tokesun (01 and 02) were also compared to assess intra-location variability in quartz rock size and density.

  • With the exception of these variables, or unless otherwise noted, Tokesun refers to the main study site, Tokesun 01.

LCC colonization and habitat availability

Characteristics of all quartz rocks at the sites

No significant differences in quartz rock density were observed among sites (anova, log transformed, df=4,20, F=2.4, P=0.083) and rock density was not correlated with colonization either at or within sites (n=25, ρ=−0.154, P=0.9) (supplementary Table 2). Therefore, a greater availability of quartz rock habitat did not equate to higher LCC abundance at any of the scales investigated.

All sites were dominated by small rocks ≤5cm (Table 3), although significant differences were observed for mean rock size (supplementary Table 3) and rock size distribution among sites (F-test, df=596,1318, F=8.24, P<0.0001; Cochrane–Mantel–Haenszel statistic, df=4, CMH χ2=190.2,P<0.0001, respectively). A significant correlation was shown between percent colonization and percent large rocks within sites (supplementary Table 2, ρ=0.54, P=0.0058, n=25 transects), but not across sites (ρ=0.135, P=0.83, n=5 sites). Thus, variability in rock size distributions partly explained LCC abundance at intra-site but not inter-site scales, indicating that spatial heterogeneity in rock characteristics (i.e. size) is a partial control on LCC abundance at scales <1km.

Characteristics of colonized rocks at the sites

Colonized rock size ranged from 1 to 26cm. Mean colonized rock size (Table 3) differed significantly from the mean size of all quartz rocks at each site (Z>1.96, P<0.01, all sites), with colonized rocks consistently larger than the available underlying rock size (Fig. 5). The shift to larger colonized rocks is further evident in Table 4 and Fig. 6, which show a significant and discernible trend towards larger size classes (>5cm) for colonized rocks at all sites, indicating that colonized rock size is not merely a reflection of the background quartz rock size of a site. At Ruoqiang, for example, only 6.4% of all quartz rocks were large, yet 48% of colonized rocks were large (Table 4).


Mean rock size and SE. Solid line=colonized rocks; dashed line=all rocks.

View this table:

Percent size distribution of rocks and Z-test results

SiteAll rocks (%)Colonized (%)Ztest
Tokesun 0120.459772.733−3.52*
Tokesun 0210.8101027.747−2.20*
Sorkuli 013.5131927.818−1.96*
Sorkuli 034.585836.857−4.02*
  • To compare the percent of colonized rocks that were large with the percent of all rocks that were large within a site, a Z-test was used with the Z critical=+/−1.96 for α/2=0.025 for two-tailed test.

  • * Significant at P≤0.05. These results show that, at all sites, a significantly greater proportion of all rocks were small, whereas a greater proportion of colonized rocks were large.


Frequency distribution of (a) all rocks and (b) colonized rocks by size class.

The data above indicate that cyanobacteria in hyperarid environments do not randomly inhabit available quartz rock habitat, but instead tend to colonize larger rocks within the desert pavement. This preference may, in part, explain the high spatial heterogeneity in abundance observed for all sites and the correlation between areas with higher percentages of large rocks (e.g. across and within transects) and higher percent colonization. Such an ecological response by LCC to colonizing large rocks is likely explained by the greater efficiency of larger rocks in collecting and retaining scarce water under hyperarid conditions and the variable temperature and light that affect moisture availability (e.g. undersides of rocks as moisture reservoirs, Friedmann & Galun, 1974). Interestingly, no significant differences were observed between the mean size of colonized rocks at ‘dry’ vs. ‘wet’ sites (test for unequal variances: t186=0.67, P>0.05).

Rock size also influenced LCC colonization type (i.e. hypolith, chasmolith), which varied significantly between sites (Cochrane–Mantel–Haenszel Statistic (df=6, CMH χ2=191.56, P<0.001). Multinomial logistic regression modeling (supplementary Table 4) showed larger rocks as more likely to support endolithic (within interstitial pore spaces) and chasmolithic forms, which are more prevalent at Tokesun and Ruoqiang, than the relatively smaller rocks at Sorkuli 01 and 03, where hypolithic forms dominate.

LCC spatial distribution

Site level analyses consistently showed that LCC are spatially aggregated (patchy), as is the underlying rock pattern (Table 5). However, these two patterns were not significantly correlated (Table 5). At smaller (1m2) scales, point pattern analyses for underlying rock distributions indicate greater heterogeneity, with random and aggregated spatial patterns occurring. At Ruoqiang and Sorkuli 03, where colonized rock sample size allowed analysis, the scale of LCC aggregation was 2–5m or 12–16m, or both.

View this table:

Site-level rock spatial pattern results

TokesunRuoqiangSorkuli 01Sorkuli 03
Colonized rocks
No. of quadrats (n)5010010100
Index of dispersion (ID)1.67351.97671.59262.7768
P-value of χ2 test0.0022<0.00010.1110<0.0001
All rocks
No. of quadrats (n)10101010
P-value of χ2 test<0.0001<0.0001<0.0001<0.0001
correlation coefficient, r
Z-test statistic, z0.80170.6586−0.4617−0.0995
Probability, P0.44590.52860.65660.9232
  • ID >1=aggregated spatial pattern. For χ2 test, if P≤0.05, pattern is significantly aggregated.

  • Product moment correlation tests whether the two rock distributions (colonized and all rocks) are independent, have negative association (repulsion, r<0) or positive association (attraction, r>0) Upton & Fingleton (1985).

Analysis of site-level LCC spatial patterns (Fig. 7) indicates that LCC ‘island-patches’ (as defined by Belnap et al., 2005) consisted of one to seven colonized rocks/patch, were 1–6m2 in area, and were separated by mean interspaces of 4–20m (linear distance). These site-level patterns and parameters (Table 6) were strongly tied to water availability, with patch area significantly decreasing, and inter-patch distance increasing with lower MAP/LWS levels (Table 6). These changes are manifested visually with LCC site pattern shifts from relatively abundant cover and stripes (e.g. Ruoqiang, Fig. 7, top) to mostly spots (e.g. Tokesun, Fig. 7, bottom), reminiscent of patch mosaic pattern shifts exhibited by desert plants with declining rainfall (Rietkerk et al., 2004).


Colonized rock spatial pattern maps for 10 randomly placed 1m by 50m transects (a–j) within each site: (a) Ruoqiang (23.3mm MAP); (b) Tokesun (10.0mm MAP). Black squares indicate one colonized rock per 1m2 quadrat, while numbers within the square indicate >1 colonized rock in the quadrat. Horizontal stripes, gaps (interspaces between colonized patches), and spots are evident.

View this table:

LCC patch data from transects (mean ± SE) with decreasing MAP (left to right)

Parameter/siteRuoqiangSorkuli 03TokesunSorkuli 01
No. of patches per transect (50m2)9.8 ± 0.86.9 ± 0.73.5 ± 0.50.8 ± 0.3
Mean patch area (m2)1.5 ± 0.11.4 ± 0.11.1 ± 0.11.0 ± 0.3
Mean inter-patch distance (m)4.4 ± 0.56.0 ± 0.67.6 ± 0.820.1 ± 2.5
Mean no. of colonized quartz/patch2.2 ± 0.22.0 ± 0.21.5 ± 0.31.1 ± 0.6

Because site-level spatial patterns were linked to MAP, we hypothesized that one underlying mechanism for LCC patchiness was dispersal via rainfall. To further test this hypothesis, we randomly selected small rock clusters (∼1m2) within three sites (Table 7). In each case, clustered rocks (facilitating potential short-range dispersal) had higher colonization than either non-clustered rocks or the site mean, with the disparity growing as water scarcity intensified (Table 7). Colonized rock clusters at the two driest sites had ∼20 times higher abundance than the overall site, whereas at the wettest site (Ruoqiang) the difference was less pronounced.

View this table:

Individual rock cluster studies (mean ± SE), with decreasing MAP (left to right)

ParameterSorkuli 03TokesunSorkuli 01
No. of patches in study51414
Mean no. of quartz stones (uncolonized)/patch30 ± 5.638 ± 6.619 ± 2.9
Mean patch area (m2)0.9 ± 0.31.0 ± 0.21.2 ± 0.2
Mean no. of colonized quartz/patch3.2 ± 1.96.3 ± 1.91.1 ± 0.4
Mean % colonization, individual patches9.5 ± 4.420.9 ± 4.89.1 ± 3.0
Percent of all patches colonized6010050
Mean site percent colonization8.5 ± 1.60.99 ± 0.20.37 ± 0.16
Patch-to-site percent colonization ratio1.120.324.6

Relationships between rainfall and other climate conditions and LCC abundance were further examined using the odds ratio model. Results showed that hot, wet sites such as Ruoqiang are 43.9 times more likely to be colonized than cold, dry sites such as Sorkuli 01 (Table 8). To investigate the separate influence of rainfall and temperature, we further segmented the data by MATA and MAP (or LWS) to show that LCC abundance follows the trend hot-wet >cold-wet >hot-dry >cold-dry. When MAP is held constant, colonization was greater at hot vs. cold sites [n=11984, goodness of fit (based on log-likelihood)=382.88, P<0.0001], which supports previous findings by Allen (1997) for Gobi Desert hypoliths. Logistic regression models indicated water availability (MAP, LWS or LWR), MATA and MATS to be the most significant positive influence on LCC abundance, with MAP exerting the largest effect.

View this table:

Results for odds ratio model that indicate the likelihood of colonization compared to baseline environmental conditions

Likelihood of site colonizationBaseline
RuoqiangSorkuli 01Sorkuli 03Tokesun 01
Sorkuli 0143.92127.3313.103
Sorkuli 031.6070.0370.114
Tokesun 0114.1540.3228.807
  • In the odds ratio model, based on MAP and MAT, Ruoqiang and Sorkuli 03 are classified as ‘wet’ sites and Tokesun and Sorkuli 01 as ‘dry’ sites; Tokesun and Ruoqiang are classified as ‘hot’ sites and Sorkuli 01 and Sorkuli 03 as ‘cold’ sites. Numbers in columns indicate the odds ratio results for the site listed in the column heading compared with the site listed in the row heading, e.g. Ruoqiang is 43.9 times more likely to be colonized than Sorkuli 01. The above results show the following trend in colonization likelihood: hot, wet site (Ruoqiang) >cold, wet site (Sorkuli 03) >hot, dry site (Tokesun 01) >cold, dry site (Sorkuli 01). Alternatively, a rock at Ruoqiang is 1.6 times more likely to be colonized by LCC than at Sorkuli 03; a rock at Sorkuli 03 is 8.8 times more likely to be colonized than at Tokesun 01; and a rock at Tokesun 01 is 3.1 times more likely to be colonized than at Sorkuli 01.


Lithic photoautotrophic communities in deserts provide a unique opportunity to explore key facets of microbial ecology within a relatively ‘simple’ system – namely, one lacking in plants and higher organisms. Despite this relative simplicity, a complex relationship between LCC and their environment exists, and its understanding requires the intersection of fundamental theoretical ecology frameworks and concepts, including patch dynamics, ‘trigger-transfer-reserve-pulse’ (TTRP) models, self-organization, biological feedback, hierarchy theory, and heterogeneity and scale (O'Neill et al., 1986; Grundmann, 2004; Rietkerk et al., 2004; Ludwig et al., 2005).

Our data support the following major findings:

  1. MAP (and soil moisture, as LWS) determines LCC abundance across hyperarid sites, and when MAP is held constant, higher abundance is likely at hot sites;

  2. abundance increases with rock size within (but not between) sites: however, greater quartz availability (=habitat) per se does not equate to increased abundance;

  3. LCC spatial pattern is aggregated (patchy) and not correlated with the underlying rock pattern;

  4. LCC spatial pattern is linked to MAP.

These results are explained in terms of the aridity gradient at the site level, but at smaller scales require an incorporation of various desert plant ecology models, particularly with regard to LCC spatial patterns. We suggest that three main mechanisms – rainfall dispersal, ecohydrological variables and self-sustaining biological feedbacks between LCC (via EPS) and soil water – are operating independently or in concert to create LCC patchiness.

By viewing LCC through the lens of plant ecology, namely the Trigger-Transfer-Reserve-Pulse (TTRP) framework (Belnap et al., 2005; Ludwig et al., 2005), it is possible to facilitate interpretation of the results, gain new theoretical insights into LCC survival and distribution, and identify new hypotheses for future testing. We provide a graphical adaptation of the TTRP framework that can be applied to LCC systems (Fig. 8). Within sites, ‘local’ ecohydrological factors, e.g. topography, soil/rock properties (i.e. size), and water availability partly explain LCC spatial pattern. Such local variables can operate at multiple hierarchical scales and exhibit marked temporal and/or spatial heterogeneity (Rodriguez-Iturbe, 2000; Lookingbill & Urban, 2004).


Trigger-Transfer-Reserve-Pulse diagram adapted for LCC from Ludwig et al. (2005). Precipitation (trigger) results in runoff (RO, transfer) from the inter-patch, which can be captured as run-on (RN, transfer) by LCC patches (reserves). Transfers (spatial redistribution) of water, soil and biota can follow trigger events, and large events may move these resources overland to connect patches. Water is stored in soil (ΔS) layers, especially the rock-soil interface, and in LCC extracellular polymeric substances (EPS), at rates dependent on EPS properties, soil infiltration (I) and rock/soil properties, such as soil texture (T). Soil water is lost by evaporation (E) or leaching (L). Transfers may result in pulses of LCC activity (C, N) and, possibly, colonization by LCC to new rocks and areas.

Reserves: LCC- island patch

Desert plant ecology frameworks (e.g. TTRP) (Ludwig & Tongway, 1997; Breshears & Barnes, 1999; Ludwig et al., 2005) explain the complex, coupled interactions between organisms and ecohydrological processes (Belnap et al., 2005; Bowker et al., 2006). Our data suggest that, like desert plants, LCC ecosystems function as ‘island-patches’, or ‘reserves’ (Belnap et al., 2005), amidst a sea of ‘inter-patch’ rock/soil pavement (Fig. 8). We extrapolate our results to the TTRP framework – aware that some specific aspects described require additional testing.

In China, a diversity of LCC patch patterns were measured that were reminiscent of plant patch mosaics – from abundant cover and lines/gaps at wet sites to mostly spots/solitary patches at dry sites. These spatial pattern transitions were significantly linked to MAP, with significant differences in mean patch number, area and inter-patch distance observed. Whether these changes reflect shifts in ecosystem states as described for desert vegetation (Rietkerk et al., 2004), or alternatively, simply declines in LCC abundance with MAP, is currently unclear.

Rainfall-mediated dispersal of LCC is indicated by declines in inter-patch distance, larger mean patch size and decreasing ratios of percent colonization of rock clusters to mean site abundance with increasing MAP. Further support exists in the aggregated vs. random LCC spatial pattern measured in this study. LCC dispersal during wetting events has been described previously for individual rocks (Bell, 1993; Allen, 1997), whereby EPS (see below) expands with water and cells are ejected and dispersed to new areas along a rock. We suggest a ‘stepping stone’ process, whereby LCC disperse via water through the wet soil matrix to adjacent soils and rocks to create larger scale patches, a mechanism possibly contributing to self-organized patchiness (Rietkerk et al., 2004) by concentrating biomass (and EPS) within small geographical spaces.


Rainfall is the key trigger for surface runoff, LCC activity and possibly dispersal. Its stochastic nature was evident (Warren-Rhodes et al., 2007), resulting in similarly high variability in water inputs (‘run-on’) to LCC patches. Winter precipitation (e.g. frost or snowmelt) is also a likely trigger at some sites (e.g. Ruoqiang).

Transfers: horizontal and vertical water movement, nutrients and biota

Precipitation can trigger runoff that may be received by LCC as run-on, stored in soil layers (e.g. rock/soil interface), or lost through leaching or evaporation (Fig. 8). The extent and rates of these processes depend on physical factors (e.g. event intensity, temperature, light, terrain), rock/soil properties (texture, depth, rock orientation/size), and biology (EPS) acting to control water availability (Fernandez-Illescas et al., 2001; Guswa et al., 2002; Lookingbill & Urban, 2004; Belnap et al., 2005; Bowker et al., 2006).

Such processes and factors are highlighted by the strong positive correlation of intra-site percent colonization (i.e. patch existence) and rock size. We hypothesize that micro-scale positive water concentration control mechanisms influence colonization, which can be understood via the TTRP model. Like plants, rocks obstruct water flow and collect run-on at the surface and rock–soil interface, where rainwater penetrates more deeply and may persist as reservoirs (Friedmann & Galun, 1974; Ludwig et al., 2005). Indeed, Mehuys et al. (1975) reported many important moderating micro-scale effects of surface rocks, including a milder thermal regime with less temperature oscillations beneath rocks – similar to plant canopies (Domingo et al., 2000; Puigdefábregas, 2005) – and temperature differences that facilitate condensation and lower overall water loss. Rocks thus create for organisms (relative to bare soil) ‘islands where water would be more available when the desert is dry’ (Mehuys et al., 1975, p. 441).

Soil moisture availability and retention benefits of the rock habitat will vary (temporally and spatially) based on myriad factors, including seasonally variable temperature and light, rock size/orientation, and the extent and position of LCC colonization. The 1–3cm subsurface location of many hypolithic communities likely reflects this control of soil-moisture availability, as well as light capture/protection requirements. We further suggest that large rocks enhance LCC survival in hyperarid deserts not only by collecting but also by retaining more run-on/soil moisture – reducing evaporation, accumulating more water, and having lower surface temperatures beneath them (i.e. thermal inertia effects) (Mehuys et al., 1975) – than bare ground or smaller rocks. These hypotheses require further testing, but this would explain why larger rocks, and areas within our sites with higher proportions of large rocks, had higher percent colonization. Moreover, in our study, where rocks were closely clustered (≤1m2 scales), abundance increased, demonstrating that rock spacing is also important. Rock orientation, depth and location (micro-topography) are also likely important for colonization at smaller scales owing to soil-moisture effects.

The presence of EPS, a common feature of most cyanobacterial communities, has several well-described benefits: (i) water absorption and retention (slows desiccation); (ii) soil adhesion (increases macroporosity and infiltrability); (iii) reduced evaporation; and (iv) nutrient capture and aid in biogeochemical processes (Foster, 1981; Lynch & Bragg, 1985; Grilli-Caiola et al., 1993; Mazor et al., 1996; Allen, 1997; Philippis & Vincenzini, 1998; de los Ríos et al., 2004). We suggest that EPS plays a role in biological feedbacks (and thus LCC patchiness) that concentrate soil water by capturing runoff and enhancing infiltration (via macroporosity), storage and retention of run-on in the LCC rock/soil environment. As LCC disperse, EPS accumulates to fill and stabilize soils, which further enhances the water concentration feedback. This process and its effects may extend beyond individual rocks to create larger-scale LCC patches.


Light and temperature, along with water, determine photosynthetic and metabolic rates and balances (Allen, 1997; Belnap et al., 2005). For this study, 200–922hyear−1 were available for carbon fixation. Water inputs probably initiate pulses of activity within LCC patches but likely also supply metabolic products to nearby soil microorganisms, as C accumulates at higher levels in the LCC ecosystem than in inter-patch areas (Warren-Rhodes et al., 2006). In hyperarid deserts, LCC patches thus function as islands of fertility (Schlesinger et al., 1990), concentrating C and other nutrients (which may also be present in soils via additional mechanisms, such as aeolian deposition) through metabolism and decomposition.

Feedbacks between transfer, reserve and pulse

The C and N pulses stemming from LCC activity during precipitation presumably create more biomass and EPS, which further intensifies feedback loops at the individual rock scale. Significant rainfall events probably also enable LCC dispersal, thereby increasing patch size. Whether this, in turn, leads to even greater positive EPS feedback at these larger-scales (cm to m) is presently unknown.


Much has been learned from the systematic study of desert plants and soil dynamics in landscape ecology and ecohydrology. To date, the application of such techniques has been limited to only a few soil microbial communities (Nunan et al., 2002; Barrett et al., 2004; Bowker et al., 2006) and is especially lacking for extremely dry and/or hot deserts (Belnap et al., 2005). This study represents a significant step toward understanding microbial ecohydrology and is the first to quantify LCC ecology systematically across significant geographical and climatic scales. The results provide novel insights into a well-studied micro-habitat but poorly understood landscape ecosystem. Of particular importance is the finding that significant LCC patchiness is displayed at multiple scales. The data support the idea that physical factors (e.g. rock size) and biological mechanisms (e.g. rainfall-related dispersal) offer underlying explanations of LCC spatial pattern in deserts. The data also raise the possibility that EPS-soil water feedback further contributes to LCC patchiness – i.e. that LCC patches are, in part, self-organized. Further study of LCC at landscape to finer scales is needed to test and quantify these effects (e.g. Ludwig et al., 2005) and to understand how LCC survive within the harsh and dynamic physical environments of the world's driest deserts.

Supplementary material

The following supplementary material is available for this article:

Table S1. Environmental parameters at the China study sites. All temperatures (T) in °C.

Table S2. Pearson correlation coefficients (ρ)between LCC abundance and (a) quartz rock density and (b) percent of large rocks.

Table S3. Supplementary. Mean rock sizes and Z-test results for all sites.

Table S4. Logistic regression results for colonization form and rock size.


The authors acknowledge the financial and logistical support of the National Research Council and the Institute of Microbiology, the Chinese Academy of Sciences and Hong Kong Research Grants Council (HKU 7573/05M). We thank P. Gong (University of California-Berkeley) and the China Meteorological Administration for long-term climate data, and the two anonymous reviewers for their insightful comments. This article is dedicated to the memory of Rosalie and E. Imre Friedmann.


View Abstract