^{1}

^{*}

^{2}

^{3}

^{4}

^{4}

^{5}

^{6}

^{*}

Conceived and designed the experiments: DPH. Performed the experiments: MBP WH NLHJ DPH. Analyzed the data: MBP DPH. Wrote the paper: MBP NLHJ JJB DPH.

The authors have declared that no competing interests exist.

Parasites are likely to play an important role in structuring host populations. Many adaptively manipulate host behaviour, so that the extended phenotypes of these parasites and their distributions in space and time are potentially important ecological variables. The fungus ^{2} of primary rainforest. We established that high density aggregations exist (up to 26 dead ants/m^{2}), which we coined graveyards. We further established that graveyards are patchily distributed in a landscape with no or very few

Parasites negatively affect their hosts in multiple ways. Because disease prevention and containment is crucial in many arenas such as city living, agriculture and the conservation of endangered species, we know a great deal about parasite spatio-temporal dynamics

Many examples of parasites altering host behaviour exist

Ants infected by the fungus

Ants biting the underside of leaves as a result of infection by

The purpose of this study was to systematically survey the spatio-temporal distribution of dead

In Thailand, regular surveys at 42 sites over the last 15 years have confirmed that

All dead ants found biting the underside of leaves were infected with the species

However, 21 days after fieldwork began an active trail of

Counts of dead ^{2}; 65% of the 40 cells were empty and only one cell contained more than 3 ants. Transect 2 had a somewhat higher mean density of 2.4 ants/m^{2}, but still 73% of the cells (i.e. 29 cells) contained less than 4 ants. However, within this almost ^{2} contained 14 dead ants and in transect 2, we found 10% of the cells containing more than 6 dead ants.

a. Density frequency distributions for the five plots and the two transect in 2006. Mean

The smaller scale patterns obtained from the five 100 m^{2} sampling plots within the graveyards gave a different impression than the large scale pattern encountered from the two transects. Recall that the five sampling plots were chosen after identifying the area as a high density location. Confirming that we chose graveyard locations we found much higher mean densities here, namely 8.2 ants/m^{2} in plot 5 and 5.0 ants/m^{2} in plot 1, 3.3 ants/m^{2} in plot 2, and 2.8 ants/m^{2} in both plot 3 and 4. In contrast to the transect plots, the five sampling plots also had less right-skewed density frequency distributions (

Parametric tests on log-transformed densities showed that transect plots were significantly different from sampling plots (t-test, df = 578, t = 6.13, p<0.01), confirming that the five sampling plots indeed were situated in patches where densities of _{4,495} = 33.23, p<0.01), and a subsequent Tukey test showed plot 5 to be different from all other plots and plot 1 to be different from plot 3, 4 and 5.

The correlogram found Transect 1 to have Moran's I values around zero for all distance classes, thus exhibiting no autocorrelation (

The five sampling plots were located within such graveyards and the correlograms also showed that the graveyards themselves were spatially structured. In all plots, except plot 4, positive autocorrelations were found at the lowest distance classes (1 m) (Plot 1: I = 0.17, p = 0.01; Plot 2: I = 0.17, p = 0.01; Plot 3: 0.17, p<0.001; Plot 4: −0.04, p = 0.51; Plot 5: I = 0.37 , p<0.001 ). In Plot 1, 2 and 3 Moran's I values decreased towards zero when the patches reached a range of ca. 3 m; in plot 5 patch range was 6 m (

The test statistics from the SADIE analyses (_{a}>1.39, p<0.02), while distributions in plot 2 and plot 4 were not statistically different from random (I_{a}<1.12, p>0.19). In each plot the means of local v_{i} and v_{j} values were nearly the same indicating that the plots are structured into sink (low

a. Schematic maps of the five 10×10 m sampling plots. Trees are shown in dark brown and paths in light brown. Black dots indicate cell densities of dead

2006 | ||||||||

Plot | I_{a} |
P_{a} |
mean v_{j} |
mean v_{i} |
P_{j} |
P_{i} |
% sink | % source |

1 | 1.40 | 0.02 | −1.38 | 1.33 | 0.02 | 0.04 | 21 | 10 |

2 | 1.12 | 0.2 | −1.07 | 1.16 | 0.27 | 0.14 | 11 | 9 |

3 | 1.79 | <0.01 | −1.70 | 1.58 | <0.01 | <0.01 | 22 | 19 |

4 | 0.96 | 0.55 | −0.97 | 0.94 | 0.52 | 0.60 | 8 | 4 |

5 | 2.46 | <0.01 | −2.18 | 2.15 | <0.01 | <0.01 | 36 | 27 |

2007 | ||||||||

Plot | I_{a} |
P_{a} |
mean v_{j} |
mean v_{i} |
P_{j} |
P_{i} |
% sink | % source |

1 | 2.15 | <0.01 | −2.05 | 2.12 | <0.01 | <0.01 | 37 | 19 |

2 | 1.36 | 0.03 | −1.40 | 1.31 | 0.02 | 0.04 | 5 | 16 |

3 | 2.00 | <0.01 | −1.95 | 1.71 | <0.01 | <0.01 | 35 | 28 |

4 | 1.94 | <0.01 | −1.90 | 1.93 | <0.01 | <0.01 | 30 | 29 |

5 | 1.72 | <0.01 | −1.67 | 1.51 | <0.01 | <0.01 | 30 | 19 |

I_{a} is an overall index of aggregation and P_{a} the associated probability value. P-values<0.025 indicates a statistical significant aggregated distribution (5%, two-tailed) and p-values>0.975 indicates a significant regular distribution (5%, two-tailed). V_{i} and V_{j} are indices for sources and sink areas, respectively, and P_{i}/P_{j} values indicate the probability of obtaining a test value respectively exceeding or below the observed test value after 6000 randomizations. The last two columns reports the percentage of sink and source area in a plot.

At the largest scale, the occurrence of graveyards of dead ^{2} the density of _{s} = 0.18; p = 0.04) and in Plot 1 between _{s} = 0.25; p = 0.01) and percentage of path (Spearman, r_{s} = −0.22; p = 0.01).

At the intermediate scale, analysing the 4 m^{2} subplots, significant positive correlations were found between mean _{s} = 0.47; p<0.01), temperature (Spearman, r_{s} = 0.56; p<0.01) and absolute humidity (Spearman, r_{s} = 0.61; p = 0.01). A strong correlation was also found between temperature and absolute humidity (Spearman, r_{s} = 0.87; p<0.01).

The total recount of all cells in September 2007 revealed some dramatic changes in mean density and distribution in the plots (^{2} in 2006 to 6.6 ants/m^{2} and 5.8 ants/m^{2}, respectively, in 2007 (Wilcoxon signed-rank test, Plot 3: Z = −6.81, p<0.01; Plot 4: Z = −6.22, p<0.01). Density frequency distributions were also much less skewed and had a considerable broader range than in 2006. After a year these plots had changed from low density, only slightly patchy plots to very structured, high density plots. In contrast plots 1, 2 and 5 had their mean densities reduced by 40–50% (Wilcoxon signed-rank test, Plot 1: Z = 4.64, p<0.01; Plot 2: Z = 4.66, p<0.01; Plot 5: Z = 6.91, p<0.01). Plot 1 from 5.0 to 3.1 ants/m^{2}, Plot 2 from 3.3 to 1.7 ants/m^{2}, and plot 5 from 8.2 to 4.3 ants/m^{2} and, compared to 2006, the frequency distributions had become more skewed in these plots. In all plots except plot 5 spatial correlograms showed increased levels of autocorrelation in 2007 (_{a} in all plots except plot 5, as well as greatly increased sizes of clusters (_{a} and the area of clusters was also smaller than in 2006.

Comparing the distributional patterns from 2006 and 2007, SADIE's test for spatial association found a significant overlap of the spatial distributions of dead

Overall we found that in just a single year the mean densities had changed radically within the study area. Some plots had become patchier and the existing patches had moved, whereas others showed the opposite trend, indicating that the distribution of dead ^{2}) took place in high density patches reducing the mean densities with ca 45%, but that the changes in low density patches were relatively small. After the onset of the rainy season in May the mean densities increased again but only in low-density patches, whereas high-density patches remained more or less constant. Plot 3 and 4, which originally had the lowest mean densities, were exceptions. Here both high and low density patches seemed to have an increasing influx of dead ants from March to July. Despite these striking changes, the overall mean density of the survey area did not change noticeably (2006: 4.4 ants/m^{2}; 2007: 4.3 ants/m^{2}).

Mean

As predicted, and using a range of approaches, we have shown that the distribution of dead ^{2}.

At the intermediate spatial scale (4 m^{2}), the density of dead ants within graveyards did correlate significantly with humidity, temperature and vegetation. On a large scale the location of graveyards within the forest did not correlate with such factors while at the smallest scale (1 m^{2} ) some correlations between the density and vegetation or path cover was observed. This is correlative and not causational and experimentation would be useful is teasing the relative effects apart.

An important factor to consider when examining the spatial patterns of dead ants is the location of live

A stylised representation of the 3-D habitat of the

We have shown that graveyards exist and we also wanted to test if they were ephemeral as many examples of fungal disease in insects are ^{2} after 1 year and therefore not ephemeral. However, the distributional pattern of high density areas within graveyards is not static. The total recount within plots in September 2007 revealed a significant shift in the spatial distribution of dead ants with new low and high density patches becoming established and old ones disappearing. The density in a patch will be the result of local rates of disappearance and appearance of dead

Our study has shown that the extended phenotype of the parasitic

We situated our work in Thailand because the repeated and extensive collections of fungi by BIOTEC (

Within the FDP a ca. 250×70 m survey area was delimited at an altitude of 120–140 m. Here several areas (graveyards) with dead ^{2} sampling plot was marked out (plot 5) in an area with a more open canopy. Within the 250×70 m survey area, but away from the plots, two parallel 200 m transects, 40 m apart, were also laid out, with 2×2 m plots at 20 m intervals.

We systematically turned over and checked the underside of all leaves in both sampling plots and transect plots, up to a height of 2 m (preliminary and subsequent sampling has never recorded dead ants above 2 m, details available upon request). We only checked under leaves because the extensive collections by BIOTEC have established that

The transect plots and the five 100 m^{2} sampling plots were divided into 1 m grids creating 1 m^{2} cells. All cells were georeferenced and the number of dead ^{2} plot and one in each corner. All recordings were conducted between 1000–1600 hrs. In the transects we also measured light intensity ca. 30 cm above ground. In the five sampling plots trees and paths were mapped and the percentages of ground covered by paths or trees were estimated for each cell.

Since fungal disease outbreaks in plants and insects are often ephemeral phenomenon we performed counts over time. We chose four high density cells and four low density cells from each of the five sampling plots, i.e. 40 cells, following the initial survey in September 2006. These cells were recounted in November 2006, and then January, March, May and July 2007. In September 2007 a total recount was performed on all 500 1 m^{2} cells in the five plots. Thus during 12 months we examined every leaf and counted the number of dead ^{2} of primary rainforest.

Two different methods were used for georeferenced count data in order to evaluate the spatial structure. First, we constructed Moran I correlograms for both transects and sampling plots to assess trends in the spatial autocorrelation of dead

Second, we further analysed the pattern and degree of patchiness in the spatial structure with the use of SADIE (Spatial Analysis by Distance IndicEs), also freely downloadable at

SADIE computes an index of aggregation I_{a} on the basis of the minimum distance that individuals in a sampling grid would have to move between cells for all cells to contain the same number of individuals (distance to regularity). A spatially random sample will give I_{a} = 1, while I_{a}<1 indicates a regular distribution and I_{a}>1 an aggregated distribution. Random permutations of the cell counts provide a method for statistical testing _{i} and v_{j} denote the degree to which a cell must receive or deliver individuals in order to obtain regularity. Values of v_{i} higher than ∼1.5 indicate that the cell belongs to a source (cluster) while v_{j} values lower than ∼−1.5 indicate association to a sink (gap). Values around zero represent randomness

SADIE was also used to evaluate the temporal changes in spatial distribution between 2006 and 2007

In order to detect associations between the densities of ^{2} transect plots with environmental factors such as light, humidity, vegetation cover and temperature across the landscape. For analysis on an intermediate scale, 4 m^{2} subplots were established around the humidity/temperature measuring points within the 10×10 m sampling plots giving 5 such subplots in each of the five 10×10 m plots. Mean values of

We are very grateful to the Department of National Parks, Thailand who has enabled this work. In particular we thank Dr Sarayudh Bunyavejchewin whose constant hard work has enabled research at this site. We also warmly thank Dr Dave Lohman, Dr Chaweewan Hutacharern, Dr Kanyawim Kirtikara and Prof Morakot Tanticharoen for excellent support and advice and Dr Stuart Davies of CTFS for help. We thank Dr Decha Wiwatwitaya for identifying