The use of image analysis to estimate population growth rate in Daphnia magna

Authors

  • HELEN L. HOOPER,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • * RICHARD CONNON,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • * AMANDA CALLAGHAN,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • * STEVE J. MAUND,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • MATTHIAS LIESS,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • SABINE DUQUESNE,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • THOMAS H. HUTCHINSON,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • JONATHAN MOGGS,

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author
  • and * RICHARD M. SIBLY

    1. *School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK; †Syngenta Crop Protection AG, 4002 Basel, Switzerland; ‡Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, D-04318 Leipzig, Germany; §AstraZeneca UK Ltd, Brixham Environmental Laboratory, Freshwater Quarry, Brixham, Devon TQ5 8BA, UK; and ¶Syngenta Central Toxicology Laboratory, Alderley Park, Macclesfield, Cheshire SK10 4TJ, UK
    Search for more papers by this author

Helen L. Hooper, School of Biological Sciences, University of Reading, PO Box 228, Reading RG6 6AJ, UK (fax +44 118931 0180; e-mail h.l.hooper@reading.ac.uk).

Summary

  • 1Population growth rate (PGR) is central to the theory of population ecology and is crucial for projecting population trends in conservation biology, pest management and wildlife harvesting. Furthermore, PGR is increasingly used to assess the effects of stressors. Image analysis that can automatically count and measure photographed individuals offers a potential methodology for estimating PGR.
  • 2This study evaluated two ways in which the PGR of Daphnia magna, exposed to different stressors, can be estimated using an image analysis system. The first method estimated PGR as the ratio of counts of individuals obtained at two different times, while the second method estimated PGR as the ratio of population sizes at two different times, where size is measured by the sum of the individuals’ surface areas, i.e. total population surface area. This method is attractive if surface area is correlated with reproductive value (RV), as it is for D. magna, because of the theoretical result that PGR is the rate at which the population RV increases.
  • 3The image analysis system proved reliable and reproducible in counting populations of up to 440 individuals in 5 L of water. Image counts correlated well with manual counts but with a systematic underestimate of about 30%. This does not affect accuracy when estimating PGR as the ratio of two counts. Area estimates of PGR correlated well with count estimates, but were systematically higher, possibly reflecting their greater accuracy in the study situation.
  • 4Analysis of relevant scenarios suggested the correlation between RV and body size will generally be good for organisms in which fecundity correlates with body size. In these circumstances, area estimation of PGR is theoretically better than count estimation.
  • 5Synthesis and applications. There are both theoretical and practical advantages to area estimation of population growth rate when individuals’ reproductive values are consistently well correlated with their surface areas. Because stressors may affect both the number and quality of individuals, area estimation of population growth rate should improve the accuracy of predicting stress impacts at the population level.

Introduction

Population growth rate (PGR), which measures a population's rate of growth or decline on a per capita basis over a specified time period, is widely used as a descriptor of population performance (Sibly & Hone 2002). It is central to the theory of population ecology and crucial for projecting population trends in conservation biology, pest management and wildlife harvesting (Caughley & Sinclair 1994; Caswell 2001; Beissinger & McCullough 2002). In ecotoxicology, PGR is preferred to individual endpoints such as fecundity or survivorship because it provides a single summary statistic that correctly integrates chemical effects on individual life-history traits (Forbes & Calow 1999; Stark & Banks 2003).

Invertebrate PGR have often been estimated from life table data; however, this is time consuming, especially when life tables are required at several levels of stress, as in life table response experiments (LTRE). There has therefore been interest in devising labour-saving methods that measure invertebrate PGR directly from the factor by which the population size increases over a specified time period. Walthall & Stark (1997) measured population size as the number of individuals, taking no account of population age or stage structure; however, larger individuals are often more fecund and so contribute more to population growth than smaller ones (Stark & Banken 1999). Image analysis, which can automatically count and measure photographed individuals, offers a potentially improved methodology for estimating PGR.

It is intuitively clear that larger individuals, which are generally more fecund and likely to survive and so more likely to contribute to population growth, should be given more weight in assessment of population performance. One option when counting individuals is to weight each one according to its reproductive value (RV), which measures its expected future contribution of offspring to the population (Charlesworth 1994; Caswell 2001). The sum of these individual RV gives the population's RV. Theoretically, PGR is the rate at which the population's RV changes with time (Charlesworth 1994). Moreover, this holds true whether or not the age distribution of the population is stable. The useful conclusion is that PGR can be accurately estimated from the factor by which the population's RV changes over a specified time period.

The RV of an individual of age i, RVi, can be calculated using the formula of Charlesworth (1994):

image( eqn 1)

Here i represents the individual's current age, and j represents the ages at which it will reproduce in the future, j=i, i+ 1, i + 2, …Ω, where Ω represents the age of last reproduction. Pij measures survivorship from age i to age j, Fj measures the number of offspring produced at age j, and λ is a measure of PGR. The difficulty in using this approach in the present context is that it requires knowledge of the life table and of λ (i.e. PGR), but λ is the quantity we are trying to estimate. However, RV increases with age until the age of first breeding, just as individual size does (Fig. 1). This suggests that individual size may be an adequate surrogate for RV in the calculation of PGR.

Figure 1.

Body size, number of offspring and RV plotted against age in days, and RV plotted against body size in D. magna. Data from the laboratory life-history study of the LF clone by Glazier (1992), with RV calculated using equation 1. The correlation between body size and RV is 0·84 (P < 0·001).

We used Daphnia magna Straus, an ecotoxicological indicator species, to investigate the practicality of this approach. Under favourable conditions, populations of this freshwater Cladoceran consist entirely of females. Once mature (8–10 days), offspring are produced parthenogenetically in discrete broods every 3–5 days. Somatic growth is indeterminate, as females continue to moult after the release of each brood (Green 1954). The size of newborn offspring is also highly variable (Glazier 1992; Burns 1995; Cleuvers, Goser & Ratte 1997). Taking account of the sizes of individuals is therefore potentially important in estimating PGR in D. magna.

Here we describe and evaluate the use of a novel automated image analysis system to count and measure the sizes of individuals in laboratory populations. We assess the accuracy and precision of the system in estimating PGR both from counts and also incorporating information about individual's sizes. The two estimates of PGR were compared after 14 days exposure to three common stressors, cadmium, pH and food level, chosen because of their contrasting effects on the life history of D. magna. Exposure to cadmium reduces somatic growth and fecundity (Klüttgen & Ratte 1994), low pH inhibits reproduction (Parent & Cheetham 1980) and low food levels result in fewer but larger offspring (Gliwicz & Guisande 1992). We then considered the advantages and disadvantages of the available methods for the measurement of PGR.

Materials and methods

test animals

Daphniamagna were obtained from the Water Research Centre (WRc), Medmenham, UK, and were cultured at the University of Reading, UK for at least 2 months before testing. Cultures were initiated with third brood offspring from a single female. Groups of 20 Daphnia were maintained in 1·2 L of reconstituted water, which was prepared by dissolving 195·87 mg CaCl2·2H2O, 82·20 mg MgSO4·7H2O, 64·80 mg NaHCO3, 5·80 mg KCl and 0·002 mg Na2SeO3 in 1 L of reverse osmosis water. Initial hardness (measured as CaCO3), pH and conductivity ranged from 140 to 160 mg/L, 7·9 to 8·2 and 370 to 450 µS cm−1, respectively. The water was changed at least weekly, when a standard organic extract, Marinure (Glenside Organics Ltd, Throsk, UK) was added at a concentration of 0·2 mL/L (Baird et al. 1989). Cultures were fed daily (Monday to Friday) with 0·05 mg of Alison's dried bakers yeast (Westmill Foods Ltd, Maidenhead, UK) and suspensions of unicellular green algae, Chlorella vulgaris var viridis, equivalent to 0·5–2·0 mg of carbon depending on the age of the Daphnia. Offspring were removed at frequent intervals and new cultures were initiated every 2–3 weeks. Third to fourth brood neonates (< 24 h old) were used for testing.

experimental protocol

The studies were carried out at 20 ± 1 °C. The photoperiod was controlled (16 h L: 8 h D) and lighting was provided by a 70-watt, cool white fluorescent tube, situated 10 cm directly above the test vessels. Test vessels consisted of a glass cylinder (height 22 cm, internal diameter 18·5 cm, thickness 5 mm; Harzkristall GmbH, Derenburg, Germany) with a clear plastic lid, containing 5 L of reconstituted water. The water was aerated regularly (2 h per day) via a glass tube (length 20 cm, internal diameter 3 mm) submerged 3 cm below the water surface. Black pond liner was taped to the back of each vessel so that approximately 60% of the surface was covered. While this reduced the actual illumination, the liner increased the contrast between the Daphnia and the surrounding water, which improved the quality of the photographs.

In three separate experiments, groups of 10 neonates were exposed to different concentrations of cadmium chloride (Sigma-Aldrich, Poole, UK), low pH (manipulated by the addition of dilute HCl), and C. vulgaris for 14 days. Control populations were also established for each experiment. All populations except those subject to varying food levels were fed the equivalent of 0·5 mg of carbon day−1 during the first week and double this ration during the second week. Each population was counted and photographed at the beginning and end of each test, as described below. PGR was calculated using surface area and image counts as 1/t loge (Nt/N0), where N0 is the initial surface area or initial size of the population and Nt is the surface area or size at the end of the test. Statistical analyses were performed using Minitab Version 14.

digital imaging analysis system

Using a slightly modified version of the system reported by Liess, Pieters & Duquesne (2006), high-resolution (image size 2560 × 1920 pixels) photographs of D. magna populations were obtained using a digital camera (Olympus Camedia C-5050). The photos were stored as a tagged image file format (TIFF) to avoid compression and loss of detail. The camera was fixed to one end of a rectangular lightproof box (length 0·8 m), whilst the opposite end of the box was fitted against the front surface of the test vessel. The top of the vessel was then covered so that the Daphnia were plunged into total darkness. After 60 s the cover was removed, allowing light to enter the system, which stimulated the Daphnia to rapidly migrate to the bottom of the vessel. When the majority of individuals were within the centre of the water column, a photo was taken (speed 1/30 s; F 3·2, ISO 400, over exposure +1).

Multiple photographs of the same population were taken and the best image (i.e. the clearest) was selected for analysis. Images were analysed using Zeiss KS 300 Imaging software 3·0 (Carl Zeiss Vision GmbH, Germany). Photos were first subjected to the same grey-value thresholds, to avoid distortion, and then converted to a binary image, so that objects (including Daphnia) were white and the background was black. Non-daphnid objects (e.g. image noise and impurities on the glass) were deleted and any holes in the selected objects were filled before automatic enumeration of the number of Daphnia and the surface area (in pixels) of the population.

validation of the imaging system

We assessed the accuracy of the digital imaging system by comparing image analysis counts with manual counts. Juveniles (< 3 days old) were counted manually by eye then assigned to one of three test vessels to achieve densities of 100, 200 and 440 individuals per 5 L aquarium. The three populations were then photographed and batches of 5–10 juveniles were successively removed and populations photographed until 5, 100 and 200 individuals, respectively, remained per vessel. Images were processed as described above.

Results

validation of the imaging system

The relationship between image and manual counts is shown in Fig. 2 (small filled circles). There was a good correspondence (regression analysis r2 = 0·99, n= 57, P < 0·001), although image counts were consistently lower than manual counts by approximately 30%. The relationship was described by a no-constant regression as image counts = 0·703 manual counts. Daphnids counted manually but not by image analysis may have been obscured by other conspecifics and/or by the curvature of the test vessel.

Figure 2.

Relationship between image analysis counts and manual counts of D. magna showing juveniles (< 3 days old) and populations of varying ages obtained after 14 days exposure to cadmium, low pH and food stress.

To obtain a spread of sizes and ages, populations from the stressor studies were used, as these contained varied mixtures of adults and juveniles after 14 days exposure. Population densities, assessed by manual counts and image analysis at the end of the tests, followed the same relationship as the unstressed juveniles (r2 = 0·97, n= 52, P < 0·001; Fig. 2). No-constant regression showed that image counts = 0·676 manual counts.

As swimming direction, position in the vessel and slight variations between test vessels (e.g. glass thickness) may all influence the measurement of population surface area, we assessed the reproducibility of measured surface area by analysing six photos of a single individual in six different vessels (36 photos in total). There was some variation within individual test vessels (Fig. 3), which we attributed to the swimming angle of the female, but little variation between test vessels (anova F5,30 = 1·02, P= 0·425).

Figure 3.

Mean surface area of one individual in different test vessels (n= 6 images per vessel). Bars indicate standard deviations.

comparison of methods to measure pgr

PGR (per day) was estimated for each population using both surface area and image counts (Fig. 4). There was good agreement between the two methods, although surface area estimates were slightly higher and there was a suggestion of increased scatter at low PGR values (bottom left of Fig. 4). The data fit a quadratic (r2 = 0·91, n= 81, P≤ 0·001) better than a linear regression model, the equation being: PGRarea= 0·057 + 0·699 PGRcount+ 1·029 inline image.

Figure 4.

Comparison of population growth rates of D. magna, calculated using population surface area and image analysis counts, after 14 days exposure to different levels of various stressors. Each population was initiated with 10 neonates (< 24 h old).

Discussion

Overall the image analysis system described here proved reliable and reproducible (Figs 2 and 3). The image counts correlated well with manual counts but with a systematic underestimation of about 30%. This underestimation could be corrected by a ‘calibration factor’ if required; alternatively it is possible that the discrepancy could be reduced by alterations to container shape (D. Ebert, personal communication). Image analysis is likely to be better than manual counting when densities are high, because movement of many individuals makes it difficult to keep track of which have been counted, and image analysis avoids the need for the disruptive sampling that would be needed for accurate manual counts. The systematic underestimation of counts by image analysis is of no consequence in the calculation of PGR, because PGR is calculated from the ratios of two estimates, as Nt/N0.

In practice there was good agreement between area estimation and count estimation of PGR, although the former was systematically higher than the latter (Fig. 4). The area estimates were higher because the study populations were all initiated with neonates but by the end of the test the surviving founder members were relatively large adults. In this situation, weighting individuals by their areas, as in the area estimator, will inevitably result in higher estimates of PGR than unweighted estimates, the count estimator.

Further explanation for the differences in performance of the two estimators relates to the population composition at the end of the experiments. Under control conditions (open circles in Fig. 4) most of the founders survived and each produced several dozen offspring, so that the final populations comprised a mix of large and small individuals, resulting in consistently high PGR whichever method was used. In contrast, highly stressed populations (left-hand side of Fig. 4) varied markedly in population composition. One of the stressed populations consisted of < 10 neonates, their mothers having died before the end of the test. In this instance the surface area estimate of PGR was much lower than the image count estimate (extreme left point in Fig. 4). Generally in these stressed populations some of the founders died, removing substantial RV, but there was variation between populations in reproductive success and so in the numbers of neonates. These variations in population composition produced discrepancies between the PGR estimators because neonates were given the same value as adults in the count method.

To assess which PGR estimator is more accurate, we need to know the true value of PGR. This would be obtained if individuals could be weighted by their RV, which unfortunately, as explained earlier, is not feasible in practice. However, body size correlates well with RV in the data presented in Fig. 1 (r = 0·84) and so may provide a useful surrogate for RV. It follows that the superiority of area estimation of PGR depends critically on the existence of a good correlation between body size and RV. Simulations suggest that the correlation would be little affected if all fecundities were reduced by the same amount (Fig. 5; second row, r= 0·75) or if the reproductive schedule was delayed (Fig. 5; third row, r= 0·72). However, the correlation drops if fecundity declines but body size does not (Fig. 5; fourth row, r= 0·26). If the somatic growth rate was reduced but the pattern of body growth was otherwise unchanged then there would be no effect on the correlation between RV and body size. If mortality rate is constant throughout the animal's lifetime, then its value does not affect RV or the correlation between RV and body size. The correlation is also little affected if mortality rate varies by a factor of five positively or negatively after day 11, as demonstrated in Fig. 6. Changing mortality rate from 0·1 day−1 to 0·5 day−1 (Fig. 6 top row) or vice versa (bottom row) gives r= 0·87 and r= 0·74, respectively.

Figure 5.

Analysis of number of offspring vs. age, RV vs. age and RV vs. body size for three different population scenarios (bottom three rows). The top row are the data of Fig. 1. The second row shows age-specific fecundities reduced by a factor of 5; the third shows age-specific fecundities and body growth delayed by 5 days; and the fourth shows a substantial decline in fecundity after day 11. The relationship between body size and age is assumed to be as in Fig. 1.

Figure 6.

Analysis of mortality rate against age, RV against age, and RV against body size for two mortality scenarios (rows). The relationship between body size, fecundity and age is assumed to be as in Fig. 1 (top two panels). The top row shows an increase while the bottom shows a decrease in mortality rate after day 11.

In summary, it seems from these theoretical studies that the correlation between RV and body size will generally be good for organisms in which fecundity correlates with body size. Such correlations between fecundity and body size are commonly reported in species producing more than one offspring per breeding attempt (Roff 2002). Where such correlations exist, area estimates of PGR are likely to be more accurate than estimates based on counts. This is still true when there exist large old individuals with reduced RV, as they will generally make up a small fraction of the population, so their contributions to the population's RV will anyway be small.

A similar imaging system to the one described here was reported by Færøvig, Anderson & Hessen (2002). Using their technique, image counts were equal to manual counts at densities greater than 100 individuals, although a maximum density was not stated. The method of Færøvig, Anderson & Hessen (2002) however, is more complex. Daphnia were transferred to an observation chamber that was viewed from above. Image measurements of the dorsal axis were modelled to represent body length, width and volume; the latter was the best predictor (r2 = 0·96) of individual dry weight. Dorsal measurements were obtained by subtracting two consecutive binary images from each other so that only moving objects (i.e. Daphnia) were captured. In our system, analysis of a single image was more accurate than the subtractive method, especially at high densities where individuals may overlap and consequently be excluded from the total number. Compared to the method of Færøvig, Anderson & Hessen (2002), our approach is less time consuming, avoids disturbance of the test system and involves no awkward calibration of individual dimensions.

Commercial image analysis systems are also available in which body size and other traits of individuals (e.g. swimming behaviour) can be measured. Currently, however, although these systems can count the number of Daphnia, they are limited to quantifying populations at low density. This is because the Daphnia are placed inside the ‘analyser’, usually within a microtitre plate or other small container. Enumeration of large populations is possible but, unlike the imaging system reported here, would involve the transfer of each individual and require many measurements per population.

We hope the techniques described here will be helpful wherever photographs of populations can be taken that indicate the sizes of individuals, provided these sizes correlate with fecundity. Existing laboratory studies have used the technique to estimate PGR in populations of the springtail Folsomia candida exposed to various concentrations of zinc (Noël et al. 2006a) and Ivermectin (Noël et al. 2006b).

In conclusion there are both theoretical and practical advantages to area estimation of PGR when individual RV are consistently well correlated with their surface areas. Daphnia magna is widely used to assess toxicant effects by measuring life-history parameters such as survivorship and fecundity, and more recently using count estimates of PGR. However, some toxicants may stimulate an increase in offspring number, but the offspring may be of low quality (small size, increased mortality risk and less fecund), as in Hammers-Wirtz & Ratte (2000). Count estimation during a short period of time then exaggerates population growth. In such situations ‘weighting’ individuals by their size, as in the area estimates reported here, will improve the accuracy of prediction of toxicant impacts at the population level.

Acknowledgements

Financial support was provided by NERC, Syngenta and AstraZenca (project NER/D/S/2002/00413 ‘The population and molecular stress responses of an ecotoxicology indicator species’). We are grateful to S. Wahrendorf and B. Pieters for their help with the digital imaging processing system, to Dr D. Glazier for checking we correctly used his data, and to Professor B. Charlesworth for comments on theoretical aspects.

Ancillary