The authors have declared that no competing interests exist.
Conceived and designed the experiments: CGE DJG MS JBO RWT. Performed the experiments: CGE DJG MS JBO. Analyzed the data: CGE DJG MS JBO RWT HGM. Wrote the paper: CGE.
Marine diseases are of increasing concern for coral reef ecosystems, but often their causes, dynamics and impacts are unknown. The current study investigated the epidemiology of
Infection with ARBS caused increased loss of healthy sponge tissue over time and a higher likelihood of individual mortality. Hurricane Irene had a dramatic effect on
Substantial impacts on marine populations and communities have been attributed to diseases of marine organisms
In addition to disease, periodic storm events can have major impacts on coral reefs. Strong storms do not affect all reef species equally
Spatial analysis offers a powerful method to study the epidemiology of diseases within a population. Population monitoring can be used to develop a time series of disease status in individuals within a population. These spatial and temporal patterns of disease incidence can then be used to discern the process of transmission
This study was conducted on two shallow reefs (3–5 m) near the Perry Institute for Marine Science on Lee Stocking Island, Exuma Cays, Bahamas, from January 2008 to June 2012. Field monitoring was conducted at Big Point (N 23° 47.301”, W 76° 08.118”) and Rainbow Gardens (N 23° 47.798”, W 76° 08.786”), located 1.5 kilometers apart. Permission for use of the study locations was provided by the Department of Marine Resources, Ministry of Agricultural and Marine Resources of the Bahamas.
The study investigated the epidemiology of ARBS in the common Caribbean branching sponge,
In order to track the rate of progression of ARBS in individual sponges, 18 diseased individuals and their nearest healthy neighbors were marked and monitored from 2008–2009 at Big Point. Marked sponges were photographed, number of lesions counted, and measurements were made of the healthy tissue, active red bands and necrotic tissue during March 2008, July 2008 and June 2009. These data were analyzed for differential fates by indicating health status based on the presence of ARBS, as has been done in other studies investigating qualitative effects of a treatment
Permanent 10×10 m grids were established at Big Point and Rainbow Gardens reefs. Within each grid, digital images representing 1 m2 were taken using a Canon D10 underwater camera, resulting in 100 images per grid. The location of each individual
Photographs from May 2011 for Big Point (BP) and July 2011 for Rainbow Gardens (RG) were georeferenced using GPS coordinates to their locations on the patch reefs and assembled into a mosaic representing the 10×10 m grid in the program ArcMap (
The relative locations of healthy (blue dots) and diseased (red dots)
As a comparison to previous studies that used broader surveys to investigate marine diseases
To investigate whether ARBS disproportionally affected certain size classes of sponges within the population, size frequency distributions of healthy and diseased sponges were compared using a Kolmogorov-Smirnov test.
Spatial analysis techniques are used in many fields
To assess the pattern of clustering of disease within a population of
Ripley’s K function is used to analyze spatial patterns and investigate spatial dependence of features (clustering or dispersion). While many spatial statistics require selection of a specific scale, the Ripley’s K function examines patterns over a range of scales to determine the appropriate one for the specific dataset
The Ripley’s K statistic has been used in many marine disease studies
For the Getis-Ord General G and Moran’s I statistics, the sponges on the grid, represented by point vectors, were converted to Thiessen polygons to best represent the spatial relationships among sponges in the population
To address disease transmission mechanisms between sampling times, join-counts were performed on all sponges within each population
Join-counts in this study were used to assess the presence of two types of connectedness: physical contact and Gabriel (vector) contact. Transmission via direct physical contact between sponges was hypothesized to be important based on previous studies demonstrating ARBS transmission with forced contact between diseased and healthy individuals
To test for vector-based transmission, we used a Gabriel connectedness scheme originally employed in studies of plant pollinators
Join counts also allowed for calculation of a transmission rate for each potential transmission mechanism. Transmission rate was calculated as D:D / (D:D + D:H), where D:D represents the number of joins between diseased individuals and D:H represents the number of joins between a diseased individual and a healthy individual (total diseased sponge connections).
The polygon-based datasets were analyzed to identify pairs of polygons sharing boundaries and containing at least one diseased individual. The number of connections between polygons containing diseased sponges was counted for each time point. Similar to previous analyses, the original attributes describing each polygon were randomly assigned, creating a different dataset in a total of 100 iterations, and the expected random number of diseased polygon connections was compared to the observed diseased polygon connections as described previously.
Analysis of individually marked sponges demonstrated differential fates for healthy and diseased individuals. At each subsequent sampling time, the health status of the monitored diseased individuals was significantly different than that of their healthy neighbors, mostly due to a greater number of diseased individuals going missing in surveys through time (Wilcoxon Rank-Sum test, P < 0.0001 and 0.03 for July 2008 and June 2009, respectively,
Sponge length, total length of all sponges and the number of healthy and diseased sponges from each site and sampling time point are summarized in
Total sponges | Healthy sponges | Diseased sponges | Percent Diseased | Mean±SE length (cm) | Total length (cm) | |
Big Point May 2010 | 342 | 330 | 12 | 3.5 | 52±4 | 17,795 |
Big Point May 2011 | 285 | 254 | 31 | 10.9 | 46±4 | 13,063 |
Big Point July 2011 | 270 | 243 | 27 | 10 | 47±4 | 12,728 |
Big Point September 2011 | 280 | 261 | 19 | 6.8 | 38±3 | 10,633 |
Big Point June 2012 | 320 | 305 | 15 | 4.7 | 36±2 | 11,529 |
Rainbow Gardens May 2010 | 133 | 126 | 7 | 4 | 56±6 | 7,400 |
Rainbow Gardens July 2011 | 187 | 174 | 13 | 9.8 | 48±4 | 8,283 |
Rainbow Gardens September 2011 | 118 | 108 | 10 | 8.5 | 54±5 | 6,405 |
Rainbow Gardens June 2012 | 152 | 147 | 5 | 3.3 | 50±5 | 7,599 |
The dashed line represents the occurrence of Hurricane Irene.
ARBS disproportionately affected larger sponges in the population, as determined by comparing the size frequency distributions of healthy and diseased sponges (
A-E are pre-storm sampling times and F-I are post-storm sampling times. White bars represent healthy sponges and black bars represent diseased sponges.
Getis-Ord quadrats (on left) display areas where the General G value was significant for clustering. Moran’s I quadrats (on right) display areas of significantly high clustering and significantly low clustering. These statistics recognize slightly different areas as “clustered” due to differences in the ways in which they calculate spatial relationships. Also note that between sampling times, significant clusters often overlap or are immediately adjacent to one another, suggesting that transmission occurs over a relatively small scale.
Getis-Ord results are shown on the left and Moran’s I results are shown on the right. The hurricane randomized the patterns observed during pre-storm sampling times, but this pattern appeared to recover to some degree between September 2011 and June 2012 at Big Point.
Contact connectedness | Gabriel Connectedness | ||||||||||||
Observed/ Expected joins | Observed/Expected Joins | ||||||||||||
H:H | H:D | D:D | Total | T. rate | H:H | H:D | D:D | Total | T. rate | ||||
BP May 2010 | 873/843 | 98/143 | 2/4.4 | 973/991 | 0.93 | 2% | 587/604 | 41/45 | 0/1 | 628/650 | 0.99 | 0% | |
BP May 2011 | 393/328 | 106/144 | 24/14 | 524/486 | 18% | 407/425 | 100/109 | 10/7 | 517/540 | 0.5 | 9% | ||
BP July 2011 | 458/331 | 126/156 | 18/14 | 602/502 | 0.3 | 12.5% | 393/410 | 81/92 | 8/5 | 482/507 | 0.13 | 9% | |
BP September 2011 | 412/341 | 94/92 | 4/4.7 | 510/434 | 0.71 | 4% | 424/459 | 60/67 | 2/2 | 486/529 | 0.7 | 3% | |
BP June 2012 | 482/381 | 72/80 | 6/3 | 560/464 | 0.14 | 8% | 508/549 | 46/55 | 2/1 | 556/605 | 0.27 | 4% | |
RG May 2010 | 184/83 | 30/20 | 4/1 | 218/104 | 12% | 197/216 | 23/24 | 2/0.5 | 222/241 | 8% | |||
RG July 2011 | 298/121 | 66/29 | 8/1.6 | 373/152 | 11% | 225/274 | 42/45 | 2/1.6 | 269/321 | 0.45 | 4.5% | ||
RG September 2011 | 236/71 | 46/18 | 0/1 | 282/90 | 0.99 | 0% | 157/176 | 30/33 | 2/1.4 | 189/210 | 0.39 | 6% | |
RG June 2012 | 306/119 | 20/12 | 0/0.2 | 326/132 | 0.99 | 0% | 221/259 | 15/18 | 0/0.3 | 236/278 | 0.99 | 0% |
Joins are represented as observed vs. expected (based on a random distribution) number of H:H = healthy:healthy, H:D = healthy:diseased, D:D = diseased:diseased; Total = total joins in a grid. P = the proportion of random diseased:diseased joins that were greater than or equal to the observed number of diseased:diseased joins. T. rate = transmission rate, calculated as observed D:D divided by observed D:D plus H:D. BP = Big Point and RG = Rainbow Gardens. Bold numbers = significant values and Underlined numbers = trends in the results. The dashed line represents the occurrence of Hurricane Irene.
Analysis of polygon grids demonstrated a strong dependence of these results on clustering scale. At Big Point, where maximum clustering for significant contact joins was small, the 0.5 m2 grids did not adequately reflect the trends seen in the join-count data for individuals. However, at Rainbow Gardens, where maximum individual clustering was at a larger scale, the same trends were observed for both significant point joins and polygon joins.
Previous studies have documented detrimental effects of ARBS on
Biomass of
Epidemiology in marine systems has been extensively studied
The significant results of these analyses showed ARBS transmission via direct contact in three out of five pre-storm time points. Following hurricane disturbance, significant transmission by direct contact was not observed; however, these observations added support to the theory of transmission by physical contact. Previous data
While join-count statistics are useful for determining mechanistic information, these statistics are not commonly used in marine epidemiology studies. Direct comparison of three commonly used statistics in the current study indicates that the Moran’s Index was the most accurate at predicting significant transmission within the population of
Statistic | Origin | What is Tested | References in Marine Epidemiology | |||||||
Ripley's K function | Ripley 1981 | Clustering or dispersion over a range of distances | Jolles et al. 2002, Gardner et al. 2008, Zvuloni et al. 2009, | |||||||
Lentz et al. 2011, Muller and van Woesik 2012 | ||||||||||
Getis-Ord General G | Getis and Ord 1992 | Clustering of values in a given area | LeDrew et al. 2004, Roff et al. 2011, Ban et al. 2012 | |||||||
Moran's Index | Moran 1950 | Clustering via measuring spatial autocorrelation: | Van Houton et al. 2010, Ban et al. 2012 | |||||||
Feature similarity based on locations and values | ||||||||||
Join-Count | Sokal and Oden 1978 | Connections between individuals in a | Jolles et al. 2002, Zvuloni et al. 2009 | |||||||
population (Contact, Gabriel, Nearest Neighbor) | ||||||||||
Ripley's K function | Max | 5.5 m | 0.3 m | 0.3 m | 0.5 m | 0.3 m | 3.1 m | 2.3 m | 2.5 m | 2 m |
Range* | 0–7 m | 0–1.8, 2.3–6.4 m | 0–1 m | 0–6.2 m | 0–1.7, 4.4–6.1 m | 0–6 m | 0–5.9 m | 0–5.4 m | 0–6.2 m | |
Getis-Ord General G | Clusters | 23 | 31 | 19 | 17 | 26 | 8 | 40 | 4 | 1 |
P-value | 0.88 | 0.04 | 0.02 | 0.64 | 0.39 | 0.98 | 0.0006 | 0.59 | 0.53 | |
Moran's Index | Clusters | 2 | 15 | 6 | 7 | 8 | 4 | 8 | 3 | 0 |
Outliers | 6 | 10 | 9 | 12 | 7 | 2 | 3 | 4 | 4 | |
P-value | 0.9 | 0.002 | 0.47 | 0.72 | 0.02 | 0.0005 | 0.95 | 0.147 | ||
Join-Count - | Diseased joins (O/E) | 2/4 | 24/14 | 18/14 | 4/5 | 6/3 | 4/1 | 8/2 | 0/1 | 0/0 |
Direct Contact | P-value | 0.93 | 0.03 | 0.3 | 0.71 | 0.14 | 0.05 | 0.03 | 0.99 | 0.99 |
Join-Count - | Diseased joins (O/E) | 0/1 | 10/7 | 8/5 | 2/2 | 2/1 | 2/0.5 | 2/2 | 2/1 | 0/0 |
Gabriel (vector) | P-value | 0.99 | 0.5 | 0.13 | 0.7 | 0.27 | 0.45 | 0.39 | 0.99 |
Comparison of the results of spatial statistics associated with disease of
Multiple factors could affect the ability to characterize transmission mechanisms within natural populations. In the case of this study, one challenge was the ephemeral nature of the sponges themselves. As seen in the marked sponge data, the annual resample rate was 67% and 39%, for healthy and diseased individuals, respectively. Unlike corals, which are long-lived (years to decades), and leave behind a permanent skeletal record of their location, sponges have shorter life-spans (months to years), and when they die, they do not leave behind a skeleton that can be accounted for in subsequent sampling times. Thus, the ability to resample individuals and determine their proximate fate is lower than in studies investigating coral diseases. Another factor that might affect a spatial population study like the current one is the influence of sponges outside the sampling area. Our grids represent only a portion of the shallow reefs on which they are located. Thus, sponges just outside the sampling area could influence disease dynamics. Many diseased sponges at these sites occurred along the edges of the grids, but because these outside sponges were never directly sampled, we can only infer their influence on sponges inside the grid. These issues are potentially reflected in some of the insignificant statistical results found in this study, suggesting that additional factors may be influencing the spatial dynamics of the sponge population in these grids.
Analysis of the 0.5 meter polygon grids showed a strong influence of clustering scale, which is likely a reflection of a lower
While the individual connectedness models fit reasonably well, not all assumptions of the models were met. In the physical contact model, the sponge branch lengths were assumed to be equal because exact branch length of every sponge was not recorded, and this may underestimate sponge contacts. The vector model (Gabriel connectedness) assumed that the vectors, in this case spongivores, fed on
In conclusion, ARBS is a detrimental disease to individuals, but its effects at the population level were not as obvious, possibly due to the dramatic and potentially confounding effects of a hurricane impacting our study sites. Even with this limitation, spatial analysis techniques tremendously increased our understanding of the dynamics of
We thank Chris Freeman, Erica Hunkin, Lindsay O’Donahue, Jim Weston, Sylvester Lee and Cara Fiore for their extensive help in conducting the field sampling and measurements for this study. We also thank Greg Easson, Hal Robinson, Eric Vela, and Allison Innman of the University of Mississippi Geoinformatics Center and Darlene Wilcox for their help with GIS software and analysis. We thank Tori Redinger and the staff of the Perry Institute of Marine Science in the Bahamas for logistical support during this study.