Combining Forces - The Use of Landsat TM Satellite Imagery, Soil Parameter Information, and Multiplex PCR to Detect Coccidioides immitis Growth Sites in Kern County, California

Coccidioidomycosis is a fungal disease acquired through the inhalation of spores of Coccidioides spp., which afflicts primarily humans and other mammals. It is endemic to areas in the southwestern United States, including the San Joaquin Valley portion of Kern County, California, our region of interest (ROI). Recently, incidence of coccidioidomycosis, also known as valley fever, has increased significantly, and several factors including climate change have been suggested as possible drivers for this observation. Up to date details about the ecological niche of C. immitis have escaped full characterization. In our project, we chose a three-step approach to investigate this niche: 1) We examined Landsat-5-Thematic-Mapper multispectral images of our ROI by using training pixels at a 750 m×750 m section of Sharktooth Hill, a site confirmed to be a C. immitis growth site, to implement a Maximum Likelihood Classification scheme to map out the locations that could be suitable to support the growth of the pathogen; 2) We used the websoilsurvey database of the US Department of Agriculture to obtain soil parameter data; and 3) We investigated soil samples from 23 sites around Bakersfield, California using a multiplex Polymerase Chain Reaction (PCR) based method to detect the pathogen. Our results indicated that a combination of satellite imagery, soil type information, and multiplex PCR are powerful tools to predict and identify growth sites of C. immitis. This approach can be used as a basis for systematic sampling and investigation of soils to detect Coccidioides spp.


Introduction
Valley fever research has predominantly focused on the medical and epidemiological aspects of Coccidioides immitis and Coccidioides posadasii, the fungi that cause coccidioidomycosis ( [1,2], and references therein). Coccidioides spp. can have a complete life cycle as soil dwelling organisms but if the soil is disturbed, their arthroconidia can become air-borne and are able to infect a host via the respiratory tract. About 60% of infected patients report no symptoms [3]; about 25% exhibit severe flu-like symptoms, such as cough, sputum, fever, and muscle aches; the remaining 15% become very ill with pneumonia-like symptoms (e.g. pleurisy and heavier sputum) requiring medication and bed rest. In a small number of cases (about 0.5-1%), the disease disseminates beyond the lungs to e.g. the skin, bones, and/or meninges of the brain, and the disease can be fatal. Certain sectors of the population seem to be more susceptible to infection, such as the very young, persons newly arrived to the endemic areas (since immunity develops with infection), field-, and construction workers, and those with impaired immune systems [4].
Coccidioides spp. are endemic in the southern part of the San Joaquin Valley in California, southern California, the southern part of Arizona, New Mexico and Texas, most of northern Mexico, and some areas in Guatemala, Honduras, Venezuela, northeastern Brazil, Argentina, and Paraguay [5,6]. Given its geographic distribution, it is evident that C. posadasii is able to flourish in desert regions of the Americas (besides California), in contrast to its close relative C. immitis which seems to be restricted to areas in California. Of the two fungal species, it is C. immitis which afflicts the San Joaquin Valley portion of Kern County, California [7,8] which is the Region of Interest (ROI) of this study. However, population genomic sequencing of Coccidioides spp. revealed recent hybridization between both species [9], and nothing is known about the distribution and ecology of these hybrids.
It is reasonable to expect that climatic fluctuations might affect the rate at which humans become infected [4]. For example, an extended drought might decimate less heat tolerant, nonsporeforming soil microorganisms that had acted as natural antagonists to the pathogen in its natural environment. A wetter than-normal rainy season could help Coccidioides spp. bloom, and windy spells might facilitate the dispersal of its arthroconidia. The ''grow and blow'' hypothesis has first been introduced by Comrie and Glueck [10]. It has long been surmised that Coccidioides spp. are generally poor competitors [11], but that they are more heatresistant than competing microorganisms -thus, it can be expected that hot summers might favor its presence or dominance. Indeed, anecdotal evidence to these effects is well documented in the literature [12][13][14][15][16][17][18][19][20]. There have been a number of attempts at demonstrating this connection quantitatively with various degrees of success [10,11,[21][22][23][24][25]. Yet, despite extensive study, there is currently no ecologically consistent link identified between the environment and coccidioidomycosis rates [26]. The predicted warming of the climate in California will add another piece of the puzzle in the already complicated interrelationships of environmental factors that might support or suppress the growth of pathogens with environmental reservoirs [27,28]. However, occasionally, the pathogen was detected in other regions, such as the recent detection of Coccidioides immitis in soils of Eastern WA [29].
There also have been several attempts to characterize the ecological niche of Coccidioides spp. in more detail [4,[31][32][33], but we still do not have a complete description of this niche. To date, Fisher et al. [31] present the most comprehensive review of this subject. We need to direct attention to a few fundamental points about what is known in regards to this niche. First, it is important to realize that C. immitis and C. posadasii do not grow in disturbed soils [4,30,31] such as cultivated fields, gardens, etc. Second, whereas it was initially thought that Coccidioides spp. ecological niche corresponds to the Lower Sonoran Life Zone (as defined and described by Merriam [34]), or similar environments [14,35,36]. Later research [30] showed that this is not quite correct, and indeed more recent works [31,32,37,38] suggested that the fungus's niche corresponds more closely with thermic and hyperthermic soils in which temperatures can reach or exceed 22uC in 50 cm depth. Fisher et al. [31] described sites where Coccidioides spp. were suspected to have been present because humans or animals were reported to have been infected at these sites. Fisher et al. [31] also made the general observations that the vegetation at those sites ranged from sparse to relatively thick cover in lower Sonoran Deserts, Chaparral-upper Sonoran brush and grasslands, as well as Mediterranean savannas and forested foothills. Furthermore, they stated that the temperature regimes, climate conditions in general, and soil textures are the only indicative variables of the presence of Coccidioides spp. Microbial diversity in soils is highly influenced by the habitat's chemical and physical parameters. But biotic soil factors such as plant and microeukaryote diversity influence fungal and bacterial soil communities as well through root exudation (additional available nutrients), microbial antagonism (antibiotic production) and synergism, as well as through selective grazing by microeukaryotes [32,[39][40]. It is currently being discussed that the pathogen is in fact not very competitive as a soil saprophyte because it has lost the ability to produce a variety of enzymes that are involved in important biodegradation processes of soil organic matter, which might explain the difficulty to detect it in bulk soil [41].
There are few published data available about the distribution of C. immitis growth sites in Kern County, California [42] most probably because it has been very difficult in the past to isolate and identify Coccidioides spp. from soil and dust samples [4,14,30]. Recently, first attempts using molecular biological techniques to identify C. immitis in bulk soil samples from Kern County, predominantly around Bakersfield, have been performed [32]. Based on that study, it appeared that C. immitis is likely to be found in the Bakersfield area at locations that are non-agricultural and have about equal parts of sand, clay, and silt (clay loam), a pH between 7.8 and 8.5, an available water capacity of about 0.15-0.2 cm/cm, a water content of about 30% (1/3 bar), an available water supply (0-25 cm) of 4-5 cm, and a Cation Exchange Capacity (CEC7) of over 20 milliequivalents per 100 grams.
The idea of using remote sensing (RS) techniques to piece together environmental characteristics, environmental change, and their relationship to disease transmission has been used extensively in connection with other diseases such as malaria [43], cholera [44], and African trypanosomiases [45]. Even though the ecological niche of C. immitis is not well characterized, we present here a RS technique that allows the mapping of sites around Bakersfield, California, where the pathogen is suspected to grow based on data obtained in a previous study by Lauer et al. [32]. Our method utilized a location well-known for being a C. immitis growth site (Sharktooth hill [STH], Bakersfield, California) as a basis, and then examined satellite images of the ROI to find all locations with similar spectral signatures. This is similar to characterizing the growth sites by the vegetation that tends to grow in the same environment as C. immitis, using the vegetation type as a marker. This is reasonable because the vegetation type closely reflects the co-variation of the relevant physical and chemical parameters such as clay and sand content, temperature, pH, nutrients, water content, etc., and also affects the development of the microbial diversity in the soils [46].
To validate our approach, we investigated if a combination of remote sensing and soil parameter information can predict locations which might be suitable to support the growth of C. immitis, followed by a molecular biological approach to detect the fungus in these soils with a culture independent polymerase chain reaction (PCR) based method [32,47].

Material and Methods
No specific permissions were required for the soil sampling. Our field study did also not involve endangered or protected species.

Multispectral image analysis
Landsat-5 Thematic Mapper (TM) L1G corrected multispectral images were downloaded from the United States Geological Survey archive (http://EarthExplorer.usgs.gov). The satellite relayed a continuous data stream which was then framed into individual scenes each 23.92 sec (see, e.g., http://landsat.gsfc. nasa.gov/about/wrs.html). The images for path 42, rows 35 and 36: Worldwide Reference System to cover our ROI were downloaded, and then the two images were mosaicked. Most of the analysis that is presented here was performed on a spatial subset of this mosaicked image. This subset corresponds to an area approximately one million hectares that covers the San Joaquin Valley portion of Kern County. Our work mainly focused on a multispectral image taken on April 20, 2008 at 10:23 PM local time. This image was chosen because it was obtained at a date (during spring) were microbial activity and biomass in the soil is generally considered high, because of supportive environmental parameters, such as moderate temperatures and increased water content. Furthermore, this image had 0% cloud cover. Our analysis started by defining a 25 pixel 625 pixel area centered at latitude 35u 289 20.2999 N, and longitude 118u 549 37.0499 W. This location is at STH, an area where C. immitis has been repeatedly detected ( [30,31,48], this study). These 625 pixels were used to train the algorithm, and thus, they define a spectral class which is referred to in what follows as the ''STH-vegetation class''. To implement the Maximum Likelihood Classification (MLC) method distributed (Richards & Jia, 2006), TM bands 1 (0.45-0.52 mm, blue-green), 2 (0.52-0.60 mm, green), 3 (0.63-0.69 mm, red), 4 (0.76-0.90 mm, near infrared), 5 (1.55-1.75 mm, mid infrared), and 7 (2.08-2.35 mm, mid infrared) were used. Band 6 (10.40-12.50 mm, thermal infrared) was not used in our MLC scheme because the resolution was 60 m instead of the 30 m (as it is for the other bands). However, this band was used to compute surface temperatures as described in more detail below. Our MLC scheme then entailed computing, for each of the pixels in the ROI, the probability that it belonged to the STH-vegetation class. This probability was assumed to be normally distributed [49], and thus is given by where x is a vector location in pixel space, N = 6 is the dimensionality of pixel space, S the covariance matrix of the distribution, and m is the mean position of the spectral class. (m and S are computed from the training pixels). A threshold value was set at p 0 , meaning that if a pixel had a probability p §p 0 of being in the STH-vegetation class, then the pixel was put into this Growth sites (GS), accumulation sites (AS) and negative sites (NS) were determined by multiplex PCR results, nd: not determined. * Proof of rodent activity was observed in the immediate neighborhood of the sampling site. Soil disturbing activity was also observed by burrowing owls, coyotes, kit foxes, spiders or large ants at some locations. The dominant rodents observed were ground squirrels, kangaroo rats and hares. doi:10.1371/journal.pone.0111921.t001 spectral class. Otherwise pvp 0 ð Þ the pixel was simply left unclassified. Clearly, as the parameter p 0 decreased, the fraction f of pixels in the ROI which belong to the STH-vegetation class increased. This is because pixels which are less and less like the training pixels get included into this class.
It was also investigated how much p 0 needed to be reduced from 1 until the sites which tested positive for C. immitis came into the STH-vegetation class. This served to calibrate the method and as a validation step. Clearly, if the C. immitis-positive sites get included in the STH-vegetation class for p 0 v * 1, then our method is robust.
However, if our C. immitis-positive sites remain unclassified until p 0 w * 0, then our spectral class is poorly defined.
Lastly, we determined the area (km 2 ) that was characterized by vegetation that belonged into the STH-vegetation class over the sampling period and until early 2014 using landsat images and the software ENVI 5.1+IDL 8.3.

Surface temperatures
Surface temperature variations across the ROI were of interest as well. In addition to utilizing a vegetation class to assess potential Detection of Coccidioides immitis PLOS ONE | www.plosone.org sites for C. immitis growth, surface temperature variations across the ROI may also help to characterize the niche of this fungus. Landsat-5 TM-6 is an infrared band. This band (from the same April 20, 2008 image) was used as follows to examine the thermal landscape of our ROI. The same area in STH was taken as training pixels, and their average x x and standard deviation s were computed. Then, a simple parallelepiped method [49] was used to find other locations in the ROI with similar values. Thus, all pixels whose value was in the ranges x{ns,xzns ð Þ , n = 1,2,3 were put into this spectral class, and we referred to this as the ''STHthermal'' class. All other pixels were left unclassified. The image's digital numbers were converted to temperatures by applying the procedure described in the National Aeronautics and Space Administration's Landsat 7 Science Data Users Handbook (http://landsathandbook.gsfc.nasa.gov/). See also Chander & Markham [50] for details. As a result, a map was obtained where the surface temperature was close to STH at the time the image was taken.

Weather data
Precipitation data for the Southern San Joaquin Valley was obtained from the California Data Exchange Center (http://cdec. water.ca.gov/snow_rain.html). The cumulative monthly precipitation (inches) over time was assembled from 5 stations (Calaveras Big Trees [CVT], Hetch Hetchy [HTH], Yosemite HQ [YSV], North Fork RS (NFR), and Huntington Lake (HNT]). A more detailed analysis of the weather data for Bakersfield in particular was not the focus of this study.

Physical and chemical soil parameters
To determine physical and chemical soil parameters of all soil samples, the websoilsurvey database of the United States Department of Agriculture (http://websoilsurvey.nrcs.usda.gov/) was used. Furthermore, all sampling sites were characterized by using the soil series extent mapping tool from the website of the Center of Environmental Informatics (CEI) (http://www.cei.psu. edu/cei_wp/). Thus, through agricultural and environmental support tools available from the USDA and CEI websites, our sampling sites were further characterized in regard to land use and vegetation. Additional geological information was obtained as well, such as the distribution of certain soil types and series in California. By using the soil series extent mapping tool, our soil samples were linked to known soil series and soil groups that are characteristic for the Southern San Joaquin Valley and beyond.

Soil sampling sites
Soil physical and chemical parameters that could likely support the growth of the pathogen based on results of the study by Lauer et al. [32] were used to choose 13 new sites that were investigated in winter and spring 2011 (Jan-Apr). Additionally, two sites that were found to be strong growth sites of the pathogen in 2008/2009 were also investigated again in 2011. Six sites were the pathogen was not detected were included in this study as well. Sampling sites included in this study were all non-agricultural silt, clay or sandy loams that differed in regard to physical and chemical parameters. All sites were located within the Central Valley Portion of Kern County. Overall, 23 sites were investigated in this study by satellite imagery and multiplex PCR. Two additional sites from STH were investigated by satellite imagery only (reference sites). Based on information from the USDA websoilsurvey database, the soils belonged to 13 different soil map units. Samples were taken each month in 2008, 2009 and 2011 (some sites were not sampled in 2011) from three different depths (0-2 cm, 5-7 cm, and 18-20 cm), placed on ice during transport to the lab, and frozen at 2 80uC when not processed immediately. See table 1 for detailed information about all sites, including exact location, soil type, observed rodent activity and indication of the presence or absence of C. immitis. Also see the first column of table two for the year they were investigated. Our sampling sites were not chosen based on Landsat imagery. They were chosen mainly based on the percentage of clay in the soil as indicated by the USDA websoilsurvey database. About 30% of clay had been indicative of a potential C. immitis positive site based on previous research [32]. After results from the multiplex PCR approach became available, we evaluated if sites where C. immitis was detected correlate with sites indicated by Landsat imagery to fall into STH vegetation sites.

DNA extraction and multiplex Polymerase Chain Reaction (PCR)
DNA was extracted from well-mixed soil samples (two replicates) using the MoBio PowerSoil DNA Isolation Kit (MoBio Laboratories, Solana Beach, CA) following the manufacturer's protocol. The multiplex PCR approach developed by Greene et al. [47] and optimized for the detection of C. immitis from soil DNA by Lauer et al. [32] was used to determine the presence of fungi in general and specifically C. immitis in all soil samples with two primer pairs. Primer pair ITSC1A/ITS C2 (18S ribosomal intertranscribed spacer [ITS] region, 223 bp), which is specific for C. immitis, was used in combination with primer pair RDS478/ RDS482 (18S ribosomal gene, 650 bp) which amplifies 18S rDNA from all fungi. The ITS region was chosen due to its high    nd* = not determined in this study. nd** = not determined in this study, but confirmed as growth site by Swatek (1970). growth site* = sites were C. immitis was detected at least twice in a deeper soil layer during the late winter/spring (February-May). accumulation site* = the pathogen could only be detected on the surface of the sampling site and never in a deeper soil layer over a several year period.
negative site* = the pathogen could not be detected in any of the soil samples using the multiplex PCR method as described in this study. Sites were C. immitis was detected at least twice in a deeper soil layer during the late winter/spring (February-May) when the soil is moist and the soil temperature increased were referred to as 'growth sites' of the pathogen in this study, assuming that the soil parameters likely supported the growth of the fungus and thus, the pathogen could be detected consecutively over a several year period in the same location, over several growth seasons. Based on the definition provided by Fisher et al. [31], growth sites are sites where physical, chemical, and biological conditions are suitable for completion of the entire growth cycle required by the organism. Thus, it could be assumed that if the pathogen finds supportive environmental conditions, it would likely expand into deeper soil layers, and not just remain on the surface which can be more hostile due to desiccation and increased uv-radiation. In fact, the majority of the soil samples that contained the pathogen in deeper soil layers also contained the pathogen in surface layers. To the contrary, sites were termed 'accumulation sites' in this study when the pathogen could only be detected occasionally on the surface of the sampling site and never in a deeper soil layer over a several year period. This made it likely that arthroconidia had been transported to this location by the wind, but the pathogen was never able to complete its life cycle because of non-supportive environmental conditions. 'Accumulation sites' were also never positive in consecutive years in contrast to 'growth sites'. Fisher et al. [31] defined 'accumulation sites' as sites where arthroconidia of Coccidioides may have been deposited on or near the soil surface after being transported from growth sites by wind, water, organisms, or anthropogenic means. We are aware that we did not investigate the activity of the pathogen in the soil or verify its growth, and that finding the pathogen in the surface layer of the soil does not mean that it cannot grow there at all. Therefore, we have to consider that some of our results might have been false negatives.

Remote Sensing Approach
A false color map of our ROI for April 28, 2008 was generated and is presented in figure 1, with indication of all sampling sites. Sites which were similar in vegetation to site STH, a confirmed growth site of C. immitis, were indicated in yellow, whereas sites that are characterized by different vegetation types appeared in various shades of green and red (agricultural fields, housing developments with gardens, higher elevated mountain slopes etc.). Results by Landsat imagery indicated large areas west of Bakersfield as potential growth sites of the pathogen, in addition to the STH area east of Bakersfield. The city of Taft southwest of Bakersfield was completely surrounded by vegetation that is similar to the vegetation type that characterizes STH. Landsat imagery furthermore indicated small pockets of potential growth sites of C. immitis scattered throughout the Southern San Joaquin Valley and around and within the city of Bakersfield. Overall, the yellow colors indicate that about 15% of the landscape visible in the satellite image was covered with vegetation that has the same reflection pattern as the STH vegetation.
We also applied Landsat imagery to an area northwest of Bakersfield where two prisons are located near the cities of Avenal and Coalinga (Kings County and Fresno County) (Fig. 2). In this area, the incidence of coccidioidomycosis has been observed to be large among prison inmates, so one might hypothesize that C. immitis could be present in the neighboring environments. And indeed, yellow areas in the immediate neighborhood of the prisons, as presented by Landsat imagery, indicated the presence of potential growth sites of the pathogen.  Table 3. Probability that the sites fall in the STH-thermal class, as predicted by Landsat data. nd* = not determined in this study. nd** = not determined in this study, but confirmed as growth site by Swatek (1970 Overall, 4 false color maps were generated for our ROI showing results for April 2008 to April 2011 (one for each April, see Table 2). For each satellite picture we obtained one STH vegetation profile. Therefore, it did not matter if variation in the vegetation occurred. The April profiles were considered the most definitive for our work, because in early spring, the climate (soil and air temperature and humidity) still support the growth of the vegetation, and grasses and herbs which are characteristic for the STH-vegetation profile have not dried up yet, compared to the summer months. A validation of our approach is shown in figure 3. In this figure, the corresponding plot of e vs. p 0 is presented. This figure also presents the dependence of the fraction e on p 0 . It can be pointed out that when p 0 drops from 1, e changes from zero to 0.36 at p 0~0 :90, but then increases slowly with decreasing p 0 until p&0:04, where f &0:50, and jumps to f~1 for p 0~1 0 {5 . Most likely, this is due to the STH-vegetation class being quite distinct from all other possible spectral classes in the ROI, with a large distance (in pixel space) from those classes. Otherwise, discrete increases in e with decreasing p 0 would be expected as other surface types get merged into the STHvegetation class.
As a further consistency check, we also wanted to examine the extent to which the STH-vegetation and STH-thermal classes overlapped and if growth sites of the pathogen could be predicted by soil thermal data (tables 2 and 3, figure 4). Thus, we obtained the STH-thermal class by implementing a parallelepiped scheme as described before (see methods). For n~1, only sites 13, 14, 16 and 21 were included in the STH-thermal class (April 2008, data not shown). Sites 8,9,[12][13][14][15][16][17][18]21, and 22-24 were added when n~2. Site 11 was included when n~3. Site 25 (one of the strongest growth sites of the pathogen) was never included. Figure 4 (right) shows our results for n~2. We present this figure here because site 8, which we identified as a strong growth site of the pathogen, came into the STH-thermal class for this value of n (but site 8 was not in the class for n~1). By evaluating the agreement between satellite imagery (STH-vegetation class and STH-thermal class, data from 4 consecutive years), we found that both data sets disagreed in 12% (sites 6, 11 and 25). Almost all sites that fell into the STH-vegetation class also fell into the STHthermal class (see tables 2 and 3).

Soil series and soil parameters
By using the soil series extent mapping tool, we found that the soil series and soil groups in which the pathogen was detected around Bakersfield, California, belonged to the Garces (Natragid, sites CLR, Bear Mt. Rd.), Chanac (Haploxerept, site AHOF), and Pleito (Haploxeroll, sites STH1 and 2 and 3) series. These soil series are not restricted to the Southern San Joaquin Valley. See figure 5 for a distribution of these soil series in California. Soils that belonged to the Chanac soil series can also be found in western Arizona and southern Nevada. All soils were of mixed mineralogy, had a superactive cation exchange capacity, and were thermic soils with predominantly fine loamy particles. Soils that belonged to these soil series are among the dominant soils in the Southern San Joaquin Valley, especially Kern County and Kings County, but can also be found in northern and western California. The use of software such as the USDA websoilsurvey database, as well as tools available at the Center of Environmental Informatics (CEI) have been found to be very valuable in obtaining information about physical and chemical parameters of soils that could support the growth of C. immitis. Detailed information about soil type, landform, dominant parent material of the soil, as well as soil physical and chemical parameters are listed for all sampling sites in table 4. Using these tools, information about land use, vegetation, mean annual soil temperatures, and geographic setting was accessed as well and is summarized in table S1 in file S1.  Field [AHOF], see [32]), we were able to detect additional ''hot spots'' of the pathogen in 2011. These sites were located within Bakersfield city, in the southwest, northeast and northwest of Bakersfield, and near Arvin, California (see table 1). Sites where the pathogen was detected more than once in a deeper soil layer were considered growth sites, whereas sites where the pathogen was detected occasionally in the surface layer only were considered accumulation sites. Sites were C. immitis was never detected were considered negative sites and included areas within Bakersfield city, an area northwest of Bakersfield, and 2 sites near Wasco, California. Fungal DNA could be detected in all soil samples. Of all sites investigated in 2011 (two growth sites investigated in 2008/09 [CLR and AHOF], and 14 new sites, out of 16 sites altogether for 2011), 4 sites (25%) were found negative, and 12 sites (75%) were confirmed as growth sites of the pathogen. No new accumulation sites were discovered in 2011. The site at Bear Mt. Road was the strongest growth site of C. immitis for the 2011 sampling set (positive for C. immitis from Jan-Apr). For an example of multiplex PCR results, see figure 6.

Correlation between Landsat imagery and multiplex PCR results
By comparing results obtained by satellite imagery (STHvegetation class) and multiplex PCR, we found that in ,74% (17 out of 23 tested sites) the satellite imagery results and the results obtained by multiplex PCR agreed at least in one year out of four years p 0~0 :90 ð Þ . When p 0~0 :95, the agreement was ,70% (16 sites). When satellite imagery based on STH-thermal class was compared with multiplex PCR results, we found that both methods agreed only in 61% n~3 ð Þ or 65% n~2 ð Þ (table 5). Tables 2 and 3 show the probability that the sites fall in the 'C. immitis growth area' based on Landsat data in comparison to results obtained by multiplex PCR. We set Red, Green and Blue (RGB) to TM bands 4, 3, and 2 respectively. With this choice, the different depths of red indicated different plant associations. These maps were the result of implementing the MLC method with p 0~0 :90andp 0~0 :95. Two of the three sites at STH listed in table 1 were confirmed growth sites of C. immitis. STH site 2 was confirmed as a growth site by Frank Swatek (Fisher FS, personal communication based on [30]), and STH site 3 was confirmed as a growth site by multiplex PCR in this study. Sampling site 8 (Cole's Levee Rd. I), which was determined as a strong growth site of the pathogen by multiplex PCR in every year, became included in the 'C. immitis growth area' based on Landsat data when p 0 was reduced to 0.95. Sampling site 7 (Lake Webb, accumulation site, located less than 1 mile west of CLR) was added when p 0 was reduced to 0.10. Sampling sites 1, 5, 6, and 11, (near Children Center, Flood Plain, Bike Path West, and Valley Street Field) never got added for p 0 w10 {5 . This was consistent with results obtained by multiplex which confirmed the absence of the pathogen (site 6, negative site), or which detected the pathogen occasionally in surface samples only (sites 1, 5 and 11). Sites, 1, 5 and 11 were termed accumulation sites, where the arthroconidia had been likely transported to by the wind, and where the presence of the pathogen could not be detected in deeper layers by multiplex PCR. We interpret this to mean that STH was quite representative of the C. immitis ecological niche within our ROI. However, in some occasions the prediction made by satellite imagery to indicate soils that could potentially harbor the pathogen could not be confirmed by multiplex PCR. Of all 25 sites included in this study, only sites 8 and 9 (Cole's Levee Rd. I and II) fell in the STH-vegetation class in each year whenp 0~0 :90. These sites were confirmed as positive for C. immitis by multiplex PCR. Other sites that were confirmed as growth sites of the fungus by our culture independent approach fell in this class at least on one occasion out of four when  Changes in extend of areas (km 2 ) that fell into the STHvegetation class were observed for the sampling period until early 2014 and are displayed in table S2 in file S1. The year with the highest precipitation (2011) had the lowest area of vegetation that belonged into the STH-vegetation class in comparison to the years 2008 and 2009 which were characterized by a significantly reduced amount of precipitation and showed an increased area of vegetation that belonged into the STH-vegetation class (see figure  S1).

Discussion
The purpose of our study was to identify soil types in Kern County that could support the growth of C. immitis by combining Landsat imagery (based on vegetation and soil temperature), and soil parameter information (from 25 sites) with a culture independent PCR-based method to detect the pathogen. We showed that satellite imagery, combined with soil parameter information, can provide a map of locations within our ROI, where C. immitis might reasonably be expected to be found. We were able to verify the presence of the pathogen by a multiplex PCR method in about 74% p~0:90 ð Þ , when soil samples were investigated over a 4 year period. However, for about a quarter of our sites (26%), results obtained by Landsat imagery and multiplex PCR did not correlate. The reasons for this observation could be multifold. Some main factors to be considered are: 1) The amount of Coccidioides DNA extracted from the soil might have been under the detection limit of our PCR based methods (sites 22 and Detection of Coccidioides immitis 23), or 2) the resolution of the satellite imagery might not have been detailed enough (site 10, a small site of only 10 m 2 ), and 3) the distribution of Coccidioides in the soil might have been spotty, and the positive site was missed (sites that were not positive for the pathogen in all sampling years). A closer look at sites where satellite imagery and soil parameter data indicated potential growth sites for C. immitis also revealed that these sites were not uniform in regard to plant coverage, distribution, and diversity, thus, generating microhabitats for soil microorganisms that most likely would be quiet distinct, especially in and around the rhizosphere [32]. Other factors, such as fluctuation in climate and pollution of the soil might have had an impact on our analyses as well. Furthermore, it has to be considered that C. immitis might be able to persist in soils that have been converted to agricultural fields for an unknown amount of time, but its arthroconidia might never germinate and grow into vegetative hyphae. These sites could be termed dormant sites (e.g. site 20, an orchard with young almond trees). To assess these impacts on our results was not the focus of our work, but we are aware of these limitations. In previous research we have investigated the limitation of the multiplex PCR approach to detect C. immitis, (see [33] for results of primer efficiency). Briefly, we found that the sensitivity of the diagnostic PCR (ITS primer pair) was reduced compared to the primer pair that amplifies 18S rDNA fragments of all fungi (RDS primer pair).
In previous work [23][24][25], it was suggested that some environmental fluctuations are a fundamental link missing from coccidioidomycosis incidence statistical modeling schemes. One important aspect to investigate is whether fluctuations in the STHvegetation class can provide this connection, and be statistically linked to the observed variations in incidence of valley fever. In this regard, one effect to consider is the extent to which this RS approach continues to be valid through the seasons. In the spring, when plants are blooming, the different vegetation types have different spectral signatures. As the weather dries and plants wither, the spectral signatures of the relevant vegetation types may become less distinct. Thus, the vegetation on STH may not be as good a marker for C. immitis in the fall, as it is in the spring. The implicit assumption in this study is that the STH environment is the only type of environment which harbors C. immitis within the San Joaquin Valley area of Kern County. We presented in this paper arguments to support this assumption; nevertheless, it would be useful to find more similarly suitable test sites to further corroborate our findings, or to find slightly different ecotypes that support C. immitis, beside of those detected in this study. Figure 6. Example of multiplex PCR results. White arrows point on a 223 bp fragment that represents C. immitis. Site Bear Mt. Rd. shows the strongest ITS amplicons in all soil layers, whereas sites Cole's Levee Rd. and site Across CALM gave a weaker signal in some soil layers, and site Beech Str. was negative. NC = negative control. Bands that indicate the presence of the pathogen in the 2% Agarose gel were confirmed to origin from C. immitis by sequencing. doi:10.1371/journal.pone.0111921.g006 Table 5. Agreement between multiplex PCR and MLC for the STH vegetation class and the STH-thermal class to predict growth sites of C. immitis (to agree a prediction by either multiplex PCR or MLC must be confirmed at least once for the four years by the other method). We also observed changes in the extent of the STH-vegetation class over time. A comparative analysis of precipitation between 2008 and 2011 (up to early 2014) suggest that years with a reduced precipitation (drought) favor plants of the STH-vegetation class, but other factors likely play a role as well, such as development and changes in land use (see figure S1 and table S1 in File S1), which was not assessed in this study.
Compared to the STH-vegetation class data, the STH-thermal data showed considerably more variation for the four different years, as expected. The vegetation on a certain day in each year may be very similar, but soil temperatures might be more variable in different years (data for n~2 and n~3 can be seen in table 3, no data is shown forn~1). Other strong growth sites of the pathogen (sites 8, 9, and 13) were also not consistently included in all years, not even withn~3. It should also be noted that sites 6 (negative site) and 11 (accumulation site) were never included into the STH-vegetation class by satellite imagery, but were included in the STH-thermal class whenn~3. We concluded therefore, that the STH-thermal classes alone might not be sensitive enough to predict growth sites of C. immits. Site 25 was never included (thermal class) maybe because of the limited resolution of the satellite imagery, as discussed earlier. It is important to keep in mind that the TM-6 image tells us surface temperatures. It may very well be that what matters is the temperature below the surface [31]. For our ROI, all landcovers were similar on the macroscale; therefore, apparent temperatures were appropriate for comparison purposes.
To improve the value of satellite imagery data, the actual spectral reflectance profiles of various soil components could be included to complement the satellite data (for details see http:// www.africasoils.net/data/ldsf-description) in future studies. A time series analysis could also be considered, if feasible. Phenology development throughout the year can make the analysis more specific to a particular vegetation type. Niche modeling rather than automated classification could be considered as well to obtain a richer output that indicates variables of importance. However, a large dataset would be necessary that would include presence and also absence data of the pathogen in a certain type of soil at a certain time with presence or absence of a certain type of vegetation. Furthermore, it should be considered that broad band signatures over larger geographic areas and ecotones might not be precise enough to be useful in predicting growth sites of a pathogen, especially when the pathogen could be adapted to grow in a variety of different ecosystems.
Our results indicated that strong growth sites of the pathogen were likely associated with 3 different USDA soil map units (180, 131, and 205), which were all loamy sands. Several sites around Bakersfield, California, that fell into one of these map units were indeed growth sites of C. immitis, as confirmed by multiplex PCR, and were similar in vegetation compared to the STH area. These types of soils are not restricted to the Southern San Joaquin Valley, but can be found in other areas of California as well. One could hypothesize that with a drier and warmer climate, as it is predicted for California in the near future [51,52], C. immitis might be able to expand its current range. In our study, we focused on soil samples from only one County, the above mentioned Kern County in the Southern San Joaquin Valley of California, a highly endemic area for C. immitis. The soil types investigated here did not comprise all types that can be found in our ROI. Even though Kern County is a hot spot of C. immitis with the highest incidence of coccidioidomycosis documented for as long as incidence data is recorded in California, we cannot conclude that soils that predominate in this area are the ones that also predominantly support the growth of the pathogen. A more rigorous sampling framework should be attempted in the future that would include locations beyond Kern County covering as wide a range of habitats as possible to correctly determine growth sites of C. immitis, as well as determining sites that are not supporting the growth of the pathogen. Developing such a sampling plan should include stratification, replicate sampling, and determination of important chemical and physical soil parameters, including investigations in other countries where coccidioidomycosis occurs would be of value as well. The ultimate goal would then be to generate a U.S. or America-wide database of occurrence and absence of Coccidioides spp. Such a database could be useful for characterizing the ecological niche for both Coccidioides species, and could indicate a variety of supporting ecosystems, as well as being an advisory public health tool, to reduce incidence of coccidioidomycosis in Kern County and elsewhere.
In conclusion, the combination of the methods used in our research can be used to generate maps that indicate potential growth sites of C. immitis, and thus serve as a tool to further investigate the ecological niche occupied by the pathogen in the Southern San Joaquin Valley and beyond in more detail. Recent advances in computer processing and geographic information system and global positioning system technologies facilitate integration of remote sensing data, such as environmental parameters with disease incidence data, so that models for disease surveillance and control can be developed [53,54]. File S1 Supporting tables. Table S1. Detailed soil series descriptions of sites which were found to be growth sites of C. immitis. Table S2. Extend of STH-vegetation class in our ROI between 2008 and 2011 based on satellite imagery. (DOCX)