Quantitative Assessment of Eye Phenotypes for Functional Genetic Studies Using Drosophila melanogaster

Abstract

About two-thirds of the vital genes in the Drosophila genome are involved in eye development, making the fly eye an excellent genetic system to study cellular function and development, neurodevelopment/degeneration, and complex diseases such as cancer and diabetes. We developed a novel computational method, implemented as Flynotyper software (http://flynotyper.sourceforge.net), to quantitatively assess the morphological defects in the Drosophila eye resulting from genetic alterations affecting basic cellular and developmental processes. Flynotyper utilizes a series of image processing operations to automatically detect the fly eye and the individual ommatidium, and calculates a phenotypic score as a measure of the disorderliness of ommatidial arrangement in the fly eye. As a proof of principle, we tested our method by analyzing the defects due to eye-specific knockdown of Drosophila orthologs of 12 neurodevelopmental genes to accurately document differential sensitivities of these genes to dosage alteration. We also evaluated eye images from six independent studies assessing the effect of overexpression of repeats, candidates from peptide library screens, and modifiers of neurotoxicity and developmental processes on eye morphology, and show strong concordance with the original assessment. We further demonstrate the utility of this method by analyzing 16 modifiers of sine oculis obtained from two genome-wide deficiency screens of Drosophila and accurately quantifying the effect of its enhancers and suppressors during eye development. Our method will complement existing assays for eye phenotypes, and increase the accuracy of studies that use fly eyes for functional evaluation of genes and genetic interactions.

Keywords: Drosophila melanogaster; human disease models; modifier screens; neurodevelopmental disorders; ommatidia; rough eye.

Figure 1

A computational strategy for automated assessment of Drosophila melanogaster eye morphology. Eye localization in a bright-field microscope image is carried out by first converting the original image (A) to grayscale (B), then morphological transformations are applied to suppress the background, followed by edge detection to identify an approximate region with ommatidial cluster (C), and finally morphological closing operation localizes the eye area (D), giving the final output image with the eye area localized (E). For detection of the ommatidial center, the original bright-field image (F) is converted to grayscale and inverted (G). Multiple filters and transformation operations further enhance the contrast of the inverted image and eliminate the noise from the ommatidial boundary (H). The centers of the bright spots due to light reflection are considered as ommatidial centers (I).

Figure 2

A conceptual representation of Flynotyper algorithm for detection of ommatidial centers, and calculation of ommatidial disorderliness is shown for different classes of eye phenotypes. Grayscale inverted and bright field images of (A) wild type control eye, (B) subtle rough eye, (C) rough eye, and (D) severe rough eye phenotypes are shown. The ommatidial centers were accurately detected in all the four categories of eye phenotypes. In the algorithm, the six neighboring ommatidia are chosen based on distance; the closest six ommatidia are considered for calculation of phenotypic scores. The six local vectors are represented by black (for grayscale/inverted) and yellow arrows (for bright-field images). The red and white arrows represent the angle between two adjacent vectors in grayscale/inverted and bright-field images, respectively. Note the differences in the lengths of the local vectors and the angles between them in different classes of eye phenotypes.

Figure 3

Analysis of Drosophila orthologs of human neurodevelopmental genes. (A–N) Representative bright-field microscope images of fly eyes displaying eye-specific knockdown of tpc1, eph, para, rk, mcph1, prosap, nrx-1, kismet, arm, caps, dpten, and dube3a genes from flies reared at 30°. Eyes of GMR-GAL4; Dicer2/+ control flies show normal ommatidial organization, while the eyes of flies with GMR-GAL4 driven RNAi knockdown of the 12 genes show disruption in the morphology of the eye. Note the variation in the severity of the eye phenotype for different genes. (O) Graph representing the mean phenotypic scores of control flies compared to the mean phenotypic scores for 21 RNAi lines with knockdown of neurodevelopmental genes (n = 9–30). The rank order of these fly lines was significantly correlated with the phenotypic scores (Spearman correlation coefficient = 0.99, P = 1.2 × 10⁻¹⁹). (P) Graph representing phenotypic scores of the 21 RNAi lines with GMR-GAL4 at 30°C is shown. The number of images analyzed for each genotype ranged from 9 to 30 (median of 20.5). Comparisons were made between each of the gene knockdowns to controls using a student t-test (*represents corrected two-tailed P < 0.001).

Figure 4

Quantitative assessment of eye phenotypes due to gene-dosage alteration. (A) Plot of mean values of gene expression of RNAi lines with eye specific knockdown at 28° vs. 30°. Note that each experiment was conducted in triplicate, and then repeated using a fresh preparation of RNA and cDNA synthesis. Gene expression of flies reared at 30° is lower than that at 28° due to increased RNAi-mediated knockdown at 30° (note that most red circles are below the black diagonal line), indicating a temperature dependent effect of the UAS-Gal4 system. (B) A graph representing phenotypic scores for knockdown of prosap, dube3a, para, pten, arm, caps, and kismet at 30° and 28° is shown. A dosage dependent increase in severity was observed for all the genes tested. Asterisks (*) show significant P-values (student t-test, corrected two-tailed P < 0.001) when phenotypic scores at 30° was compared to that at 28°. The number of images processed for each genotype ranged from n = 14 to n = 30. A complete list of statistical analysis and n numbers for each of the genotype assessed is presented in Table S6.

Figure 5

Analysis of eye images obtained from independent studies. (A) Bright-field microscopy images of representative Drosophila eyes overexpressing C9orf72 pure or RO repeats using the GMR-GAL4 driver, imaged on d 45. While three pure repeats had no effect, 36 pure repeats were toxic and 103 pure repeats showed more overt toxicity; 36 RO repeats had no effect, and 108 RO and 288 RO repeats showed a mild effect. (B) Graph representing the phenotypic scores of C9orf72 pure or RO repeats using the GMR-GAL4 driver. The phenotypic scores are concordant with the visual assessment of the eye phenotypes (Mizielinska et al. 2014). The numbers of images used for these assays were n = 4 for GMR >36 RO d45, n = 5 each for GMR > 3 d45, GMR > 36 d45, GMR > 288 RO d45, and GMR > 108 RO d45, and n = 6 for GMR > 103 d45. (C) Scanning electron microscope images of genetic modifiers of tau-induced neurotoxicity. The control listed is w¹¹¹⁸/+;gl-tau/+. All other panels, except wild type, contain one copy of gl-tau transgene in trans to one disrupted copy of the gene listed in the panel. (D) A graph representing the phenotypic scores of wild type, control, and the three enhancers and suppressors of w¹¹¹⁸/+;gl-tau/+, is shown. The number of images used for these analyses were n = 2 for par1, n = 3 each for wild type, and ksr, n = 4 each for control, cana, Klp61F, sgg, and frc.

Figure 6

Genetic modifiers of sine oculis. (A–L) Scanning electron microscope (SEM) images of adult compound eyes are shown. (A) ey-GAL4/UAS-so. Note that the eye is rough and smaller in size. (B–C) Enhancers of the rough eye phenotype due to overexpression of UAS-so with ey-GAL4. (D–E) Suppressors of the rough eye phenotype due to overexpression of UAS-so with ey-GAL4. Note that BL34665 rescues both the small eye, and the rough eye phenotype of ey-GAL4/UAS-so. (F) Wild type (WT). (G) GMR-GAL4/UAS-so. (H–I) Enhancers of the rough eye phenotype due to overexpression of UAS-so with GMR-GAL4. (J–L) Suppressors of the rough eye phenotype due to overexpression of UAS-so with GMR-GAL4. (M) A graph representing the phenotypic scores of WT, ey-GAL4/UAS-so, and the four enhancers and four suppressors of ey-GAL4/UAS-so. The number of images processed for each genotype ranged from n = 2 to n = 10. (N) Graph representing the phenotypic scores of WT, GMR-GAL4/UAS-so, and the four enhancers and four suppressors GMR-GAL4/UAS-so. The number of images processed for each genotype ranged from n = 2 to n = 10. Although the phenotypic scores distinguished the effect of most suppressors and enhancers on the eye phenotype, Flynotyper alone was not able to accurately classify BL34665 as a strong suppressor. Visual inspection of BL34665 images showed ommatidial organization and eye size comparable to that of wild type eyes.