Medical and Veterinary Entomology (2014) 28, 461–464
S H O R T C O M M U N I C AT I O N
Estimating Culicoides sonorensis biting midge abundance using digital image analysis C. J. O S B O R N E 1 , C. E. M A Y O 1 , B. A. M U L L E N S 2 and N. J. M A C L A C H L A N 1 1
Department of Pathology, Microbiology and Immunology, School of Veterinary Medicine, University of California Davis, Davis, CA, U.S.A. and 2 Department of Entomology, University of California Riverside, Riverside, CA, U.S.A.
Abstract. ImageJ is an open-source software tool used for a variety of scientific objectives including cell counting, shape analysis and image correction. This technology has previously been used to estimate mosquito abundance in surveillance efforts. However, the utility of this application for estimating abundance or parity in the surveillance of Culicoides spp. (Diptera: Ceratopogonidae) has not yet been tested. Culicoides sonorensis (Wirth and Jones), a biting midge often measuring 2.0–2.5 mm in length, is an economically important vector of ruminant arboviruses in California. Current surveillance methods use visual sorting for the characteristics of midges and are very time-intensive for large studies. This project tested the utility of ImageJ as a tool to assist in gross trap enumeration as well as in parity analysis of C. sonorensis in comparison with traditional visual methods of enumeration using a dissecting microscope. Results confirmed that automated counting of midges is a reliable means of approximating midge numbers under certain conditions. Further evaluation confirmed accurate and time-efficient parity analysis in comparison with hand sorting. The ImageJ software shows promise as a tool that can assist and expedite C. sonorensis surveillance. Further, these methods may be useful in other insect surveillance activities. Key words. Culicoides sonorensis, digital image analysis, ImageJ.
The biting midge Culicoides sonorensis is the principal vector in the U.S.A. of the arboviral disease bluetongue, an economically important disease of domestic and wild ruminants (MacLachlan, 2011). Trapping of adult C. sonorensis is an important aspect of entomological surveillance. However, comprehensive longitudinal surveys conducted to characterize the seasonality, dispersal and bluetongue virus infection rates of C. sonorensis are labour-intensive. They frequently are intended not only to quantify abundance, but also to discriminate midges by parity and sex (Gerry et al., 2001; Lysyk, 2006; Mayo et al., 2012a, 2012b). The current and most dependable way of processing field collections is by using stereomicroscopy. Although this method is accurate, it is arduous and time-consuming when trap catches measure in the thousands. In our studies, 19 700 C. sonorensis were captured by carbon dioxide (CO2 )-baited traps in one evening, but other studies have documented collections of up to 1 million Culicoides when
using OVI light–suction traps (Meiswinkel et al., 2004). Time becomes critical, especially when attempting to preserve trap catches for later molecular assays. We therefore tested the merits of ImageJ, an open-source software tool available through the National Institutes of Health (NIH) (Bethesda, MD, U.S.A.) (Rasband, 1997–2012), to help streamline the processing of field-collected C. sonorensis. ImageJ image-processing software runs Java in Windows, Mac and Linux operating systems and is available for download at imagej.nih.gov (Kesavaraju & Dickson, 2012). The aim of the present study was to develop a system to digitally catalogue and efficiently estimate the abundance, parity and sex of adult C. sonorensis collected during a year-long intensive entomological survey. Additionally, this enhanced our efficiency to preserve the highest yield of nucleic acid material, in the absence of alcohol-based collections, for molecular assays. These included but were not limited to bluetongue virus quantitative reverse transcriptase polymerase chain
Correspondence: Christie E. Mayo, Department of Pathology, Microbiology and Immunology, School of Veterinary Medicine, University of California Davis, Room 4206, VM3A, One Shields Avenue, Davis, CA 95616, U.S.A. Tel.: + 1 530 752 1163; Fax: + 1 530 754 8124; E-mail: [email protected]
© 2014 The Royal Entomological Society
C. J. Osborne et al.
reaction (qRT-PCR) assays and the sequencing of virus strains, the latter of which requires high DNA/RNA yields. This technology has previously been described in the context of estimating abundances of eggs deposited by mosquitoes and total abundances of adult mosquitoes with reliable results (Mains et al., 2008; Kesavaraju & Dickson, 2012). Culicoides sonorensis specimens were trapped on a northern Central Valley California drylot dairy farm during the spring and summer of 2013 using Centers for Disease Control (CDC) style traps without light and baited with dry ice (Newhouse et al., 1966). Such farms seldom have biting midge species other than C. sonorensis, and the CO2 -baited traps without light tend to collect only Culicoides spp. and some mosquitoes (Gerry et al., 2001; Mayo et al., 2012a, 2012b). For the optimal preservation and identification of anatomical structures and parity status, all insects were collected in a 0.5% solution of a dry detergent (Alconox, Inc., West Chester, PA, U.S.A.) and deionized water. They were transferred to 70% ethanol as soon as possible, usually the next morning. The midges were then sorted by sex and parity using standard stereomicroscopy. Parity was determined by the presence of a burgundy-red pigment deposited in the abdominal cuticle (Dyce, 1969; Akey & Potter, 1979) and by changes in the pigmentation pattern of the abdominal tergites (Potter & Akey, 1978). A set of whole trap catches was set aside for the ImageJ studies. For automated counting using the Shape Descriptor tool, insects were transferred from the initial storage vials containing 70% ethanol using a plastic pipette and deposited into Petri dishes measuring 66 mm in diameter and containing 2 mL ethanol for samples of one to 200 midges, or into Petri dishes measuring 100 mm in diameter and containing 4 mL ethanol for samples of over 200 midges. Under a dissecting microscope, all non-C. sonorensis and other insects were removed. To create the two counting conditions [(a) scatter and (b) gridded distributions to compare with visual counts], the Petri dishes were first placed on a plate shaker [Thermo Scientific Titer Plate Shaker Model no.4265 (Thermo Scientific, Dubuque, IA, U.S.A.)] for 30 s at maximum speed (approximately 1100 rpm). The ethanol was then aspirated from the Petri dish to prevent disturbing the pattern of midges during transfer to the scanner. The Petri dish was then placed on a Canon CanoScan LiDE 700F (Canon Inc., Tokyo, Japan) flatbed scanner and scanned in colour at 1200 d.p.i. (scatter distribution) (Fig. 1C). The dish was removed and the midges were then arranged on the dish in a grid-like pattern so that no two midges were touching (gridded distribution) (Fig. 1A). The dish was replaced on the scanner and scanned using the same conditions as before. The resulting images were opened in ImageJ on a MacBook Pro computer running OSX 10.8.5 (Apple, Inc., Cupertino, CA, U.S.A.). The area of the Petri dish containing midges was selected (CIRCLE/FREE SELECT TOOL) and cropped (EDIT > CLEAR OUTSIDE). The image was enhanced for optimal brightness and contrast (IMAGE > ADJUST > BRIGHTNESS/CONTRAST). Both images were converted to binary as required by the automated counting tool (PROCESS > BINARY > MAKE BINARY). The Shape Descriptor tool [Gary Chinga (http://www.gcsca.net/IJPlugins.html)] was previously downloaded and installed, and launched with the following settings:
Size Range (1000–10 000 pixels), and Detect Edge Particles (On). Pixel range was determined by selecting a subsample of midges from the binary image (SELECT WAND TOOL) and the measure tool (TOOLS > MEASURE); this range may change depending on scanning conditions. Once all parameters were made in the Shape Descriptor tool, the tool was initiated and a readout of all particles found that met the conditions was generated. The total number of particles (midges) was recorded for the study. This process was repeated for a total of 10 randomly selected vials of C. sonorensis, which contained catches ranging from 12 to 911 midges with an average of 301 midges per trap; the actual number of midges was determined by a subsequent manual count using standard stereomicroscopy. For counting using the Cell Counter tool, 10 vials of C. sonorensis were randomly selected in a manner similar to the protocol for automated counting using the Shape Descriptor tool. These vials contained from 18 to 289 midges each, with an average of 115 midges per vial. The midges were transferred from the vials into a customized scanning dish; this had an aggregate of short acrylic panels adhered by silicone to a thin, rectangular sheet of plastic film measuring 140 × 90 × 10 mm. The dish was placed on the scanner and scanned in colour at 1200 d.p.i. onto an iMac computer running OSX 10.7.5 (Apple, Inc.). The resulting image was opened and the Cell Counter tool in ImageJ initiated (PLUGINS > ANALYZE > CELL COUNTER). All C. sonorensis were labelled according to sex, parity and whether or not the insect was blood-fed by directly clicking on the image using four different Counter Types. The resulting tallies were recorded by the Cell Counter tool and the time to complete the process from transfer to final tally was noted. Next, the midges from the plate were transferred to a Petri dish and manually sorted under a dissecting microscope according to the same criteria. The time from transfer to complete tally was recorded as before. The total number of midges in each category from the manual count was used as the baseline and a Wilcoxon rank-sum test was performed to assess the difference between the times required for each method. The accuracy of ImageJ was tested by calculating the difference between the ImageJ count and the manual count and dividing it by the manual count to give a percentage of accuracy for each trap tested. Automated trap enumeration using the Shape Descriptor tool for each type of arrangement method was plotted against the manual count. R2 -values (linear regression) were 99% for the gridded layout (Fig. 1B) and 84% for the scatter layout (Fig. 1D). The time to complete gridding of each trap was not recorded; however, the total time required from the transfer of midges to enumeration was approximately 20 min for catches of < 200 midges and over 30 min for larger trap catches (> 200 midges). Normal visual counting usually requires 8 min per 200 midges, but varies by person. For small and moderately sized trap catches of between 18 and 215 midges, manual counting was faster (1:03–8:30 min:s) than counting using the ImageJ Cell Counter (3:41–9:30 min:s). Although the time required to estimate abundance did not differ significantly between the methods tested, trends demonstrated that less time was required to enumerate abundance in the two larger traps (total midge counts of 275 and 289, respectively) using ImageJ (10:32–12:17 min:s) in comparison with manual counting (11:56–13:58 min:s). The accuracy of determining sex
© 2014 The Royal Entomological Society, Medical and Veterinary Entomology, 28, 461–464
C. sonorensis abundance measured using digital analysis
Fig. 1. (A) Series of images showing the scanner-captured original image, the same image converted to binary, and the resulting readout of detected particles after using the Shape Descriptor tool (gridded). (B) Linear regression of manual compared with automated counting (gridded). (C) Similar series of images as in (A) for scatter layout. (D) Linear regression of manual compared with automated counting (scatter).
and parity status from the scanned image ranged from 91.67% to 100% in comparison with standard stereomicroscopy (Table 1). Automated counting using ImageJ is a powerful and accurate method of estimating C. sonorensis abundance when a gridded layout is used (R2 -value: 99%). However, the length of time required to set up the midges for scanning and counting was longer than that required for preparation for scatter counting. Although scatter counting is not as accurate as manual enumeration (R2 -value: 84%), it remains a useful tool for making rapid approximations. Additional tools within the Shape Descriptor tool allow for the constraining of particle detection by setting a pixel size and roundness range to further refine the insects detected (i.e. discriminating from larger or smaller ‘by-catch’ for quick estimations of only midges). Additional methods to aid in optimizing the accuracy of the Shape Descriptor tool may include altering plate shapes and styles so that the midges can be scattered quickly while reducing the grouping or bunching that leads to decreased midge detection. These constraints were not tested in our study but merit further investigation. Thus, automated methods demonstrated limited accuracy in enumerating trap catches and the binary image requirement in the program removes all of the characteristics that identify sex and parity status in midges. Using the Cell Counter tool allows us to retain this information while enumerating the entire trap. The brightness of the scanner light and definition of the scanner sensor limit some resolution when attempting to identify the characteristics of individual midges and, because the midges cannot be manipulated post hoc, some midges must be categorized by making an educated guess. Nonetheless, in our study, accuracy of midge identification ranged from 91.67% to 100%. User bias should also be considered because accuracy may be lower in less experienced individuals.
It should also be noted that these techniques make sense primarily in areas where one Culicoides species is quite dominant, as is C. sonorensis in our setting. Further, the collection of specimens into liquid (as in the present study) provides superior material for parity assessment, which is far better than more typical collections of drier insects (alive or dead) in a catch bag. Whether this is desirable depends on the goals of the study in question. The determination of parity from images would be more difficult with drier material in certain settings, especially with Culicoides species that have relatively darker abdomens that make it difficult to discriminate changes in the pigmentation pattern of the abdominal tergites (Potter & Akey, 1978). Parity assessment is important for vector surveillance and modelling as it indicates possible disease transmission events and egg laying status. Advances in imaging techniques are likely to lead to improved resolution in the defining of physiological characteristics and deserve consideration for their standardizing and time-saving properties. The use of ImageJ for trap enumeration and categorization is merited by its utility in assisting trap counts, ability to create a durable entomological record, and time-saving capacity in some scenarios in comparison with manual counting using standard stereomicroscopy. For optimal nucleic acid recovery, as in the present context, the ability to take a picture and return to count after specimens have been preserved or processed for molecular assay is perhaps the greatest benefit of using ImageJ. Additionally, these techniques could be adapted for other Culicoides species. As ImageJ is an open-source software, knowledgeable users can write programs specific to their needs and share them across a wide array of scientific fields. In summary, this method may serve as an efficient technique that will complement ongoing efforts in the surveillance of adult or immature populations of Culicoides species, especially in
© 2014 The Royal Entomological Society, Medical and Veterinary Entomology, 28, 461–464
C. J. Osborne et al.
Table 1. Comparison of times and accuracy of counts of Culicoides sonorensis for both manual counting and ImageJ counting using the Cell Counter tool.
Trap/vial 1 3 4 10 14 15 16 17 19 21
Total C. sonorensis counted, n
Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ Manual ImageJ
3:28 7:08 4:20 6:27 13:58 12:17 1:50 3:58 1:03 3:41 11:56 10:32 2:49 5:13 8:30 9:30 2:22 4:18 2:57 4:25
82 82 66 66 289 290 23 23 18 18 275 274 48 50 205 205 60 60 74 75
2 2 3 1 31 25 1 1 0 1 45 39 1 1 15 15 1 1 0 0
62 60 29 32 208 200 12 12 12 11 192 188 23 21 150 150 47 46 39 36
18 20 34 33 50 65 10 10 6 6 37 46 24 28 40 40 12 13 35 39
0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0
ecosystems dominated by a single vector species (i.e. Culicoides imicola in Africa and Europe, and Culicoides brevitarsis in Australia).
Acknowledgements We gratefully acknowledge the producers, staff and herd managers of the farms at which the samples were obtained. These studies were supported by the Agriculture and Food Research Initiative competitive grant no. 2011-02872 UCD SPO #201119009 from the U.S. Department of Agriculture (USDA) National Institute of Food and Agriculture, the Center for Food Animal Health at the University of California Davis, and the USDA under the Animal Health Act, 1977, Public Law 95-113. References Akey, D.H. & Potter, H.W. (1979) Pigmentation associated with oogenesis in the biting fly Culicoides variipennis (Diptera: Ceratopogonidae): determination of parity. Journal of Medical Entomology, 16, 67–70. Dyce, A.L. (1969) The recognition of nulliparous and parous Culicoides (Diptera: Ceratopogonidae) without dissection. Journal of the Australian Entomological Society, 8, 11–15. Gerry, A.C., Mullens, B.A., MacLachlan, N.J. & Mecham, J.O. (2001) Seasonal transmission of bluetongue virus by Culicoides sonorensis (Diptera: Ceratopogonidae) at a southern California dairy and evaluation of vectorial capacity as a predictor of bluetongue virus transmission. Journal of Medical Entomology, 38, 197–209. Kesavaraju, B. & Dickson, S. (2012) New technique to count mosquito adults: using ImageJ software to estimate number of mosquito adults in a trap. Journal of the American Mosquito Control Association, 28, 330–333.
4.55 5.19 0 5.56 3.64 8.33 0 1.67 5.41
Lysyk, T.J. (2006) Abundance and species composition of Culicoides (Diptera: Ceratopogonidae) at cattle facilities in southern Alberta, Canada. Journal of Medical Entomology, 43, 840–849. MacLachlan, N.J. (2011) Bluetongue: history, global epidemiology, and pathogenesis. Preventive Veterinary Medicine, 102, 107–111. Mains, J.W., Mercer, D.R. & Dobson, S.L. (2008) Digital image analysis to estimate numbers of Aedes eggs oviposited in containers. Journal of the American Mosquito Control Association, 24, 496–501. Mayo, C.E., Gardner, I.A., Mullens, B.A., Barker, C.M., Gerry, A.C., Guthrie, A.J. & MacLachlan, N.J. (2012a) Anthropogenic and meteorological factors influence vector abundance and the prevalence of bluetongue virus infection of dairy cattle in California. Journal of Veterinary Microbiology, 155, 158–164. Mayo, C.E., Barker, C.M., Mullens, B.A. et al. (2012b) The combination of abundance and infection rates of Culicoides sonorensis estimates risk of subsequent bluetongue virus infection of sentinel cattle on California dairy farms. Journal of Veterinary Parasitology, 187, 296–301. Meiswinkel R., Venter G.J. & Nevill E.M. (2004) Vectors: Culicoides spp. Infectious Diseases of Livestock with Special Reference to Southern Africa (ed. by J.A.W. Coetzer & R.C. Tustin), pp. 93–136. Oxford University Press, Cape Town. Newhouse, V.F., Chamberlain, R.W., Johnston, J.G. Jr. & Sudia, W.D. (1966) Use of dry ice to increase mosquito catches of the CDC miniature light trap. Mosquito News, 26, 30–35. Potter, H.W. & Akey, D.H. (1978) Pigmentation associated with oogenesis in the biting midge, Culicoides variipennis: changes in abdominal tergite patterns. Mosquito News, 38, 499–504. Rasband, W.S. (1997–2012) ImageJ. U.S. National Institutes of Health, Bethesda, MD. http://imagej.nih.gov/ij [Accessed 1 August 2013]. Accepted 4 February 2014 First published online 8 May 2014
© 2014 The Royal Entomological Society, Medical and Veterinary Entomology, 28, 461–464