Comments from curator
Phenotyping data comes from the authors'
GitHub repository mingwhy/AD_fly_eye, cited in the Data availability section of the paper (
journal page, also on
PMC): file
01_fly_eye_features/eye.image.scores/combine-3batch.txt, which has the 16 eye metrics of each of the 5,623 eye images of the screen (one eye per fly). The Supplementary Material of the paper has only the GWAS results (Table S1), the GO enrichment (Table S2) and figures.
These are F1 phenotypes, not phenotypes of the inbred DGRP lines: virgin females of 162 DGRP lines were crossed to males of the "R32" line, which carries
GMR-GAL4 and
UAS-Aβ42 recombined on chromosome II and
UAS-tau (0N4R isoform) on chromosome III (backcrossed for 6 generations to
w1118, kept over CyO and TM6, Tb). The F1 females carrying the three transgenes express human amyloid beta 42 and tau in the eye, which causes an age-dependent rough-eye phenotype (fused, pitted and collapsed ommatidia), a
Drosophila model of Alzheimer's disease. Each value is assigned to the DGRP mother line, and the cross is given in the phenotype names (_xR32).
The F1 females were kept at 25°C on cornmeal food, up to 20 per vial, and aged to 28 days. One eye per fly was imaged at 50× (Zeiss Stemi 508 stereomicroscope). The authors' automated R pipeline (EBImage) selects a region of interest (ROI) in the focal plane, segments the ommatidia in it, and measures each ommatidium (area, perimeter, mean, SD, minimum and maximum radius, eccentricity, position). For each image, it then computes the mean and the SD of each measure over the ommatidia of the ROI, and the mean and SD of the distance from each ommatidium to its nearest neighbour. This gives the 16 traits of the paper. Degenerated eyes have larger values, except for the minimal radius, which is smaller. Lengths and areas are in pixels of the image as processed by the pipeline (downscaled; the paper gives no conversion to µm). Eccentricity has no unit (0 for a circle). The per-image means and SDs are trimmed: the authors' Perl scripts drop 3 values at each end of the list of values of the image sorted as text, which are not always the 3 smallest and 3 largest values.
The raw dataset has
one row per imaged F1 female (one image), 5,285 rows for 162 DGRP lines:
-
*_d28_xR32: 4,936 images of flies aged 28 days, 10 to 107 per line (the paper says more than 10 images per line; 16 lines have exactly 10). The images of 9 lines in imaging groups 15 and 19 (48 images) were taken at 27 or 29 days. They are included, as in the authors' analysis, and the exact age is in AgeDays_xR32.
-
*_d4_xR32: 349 images of flies aged 4 days, for 23 lines of batch 3 (6 to 21 per line). The same cohort was imaged again at 28 days. These images are on separate rows, since an image at 4 days cannot be paired with one at 28 days.
- For each age, the 16 traits of the paper (names as in the paper: NNMean = nn mean, EccMean = ecc mean, AreaMean = area mean, PerimMean = perimeter mean, RadMeanMean = radiusmean mean, RadSDMean = radiusSD mean, RadMinMean = radiusmin mean, RadMaxMean = radiusmax mean, and the matching *SD traits), and NOmmatidia, the number of ommatidia detected in the ROI (in the authors' file, not analysed in the paper).
- Obsolete design columns from the authors' file: ImageID_xR32 (image name; the raw images, 16 GB, are available from the authors on request), Group_xR32 (imaging group: 26 groups of 8 to 48 lines, a fixed effect in the authors' model), Batch_xR32 (imaging batch 1 to 3) and AgeDays_xR32.
The column the authors' file calls cc (described as circularity in their R markdown) is the EBImage eccentricity of the ommatidia (ecc in the paper).
Not uploaded:
- The control images: the R32 line itself (241 images;
w1118 background, not a DGRP line) and the wild-type eye control (87 images of non-transgenic flies of DGRP line 441).
- 10 images labelled line 43: this is not a DGRP2 line (the authors had no genotype data for it), and the authors left these images out of their analysis. This explains why the file has 163 "lines" and the paper 162.
The authors' GWAS phenotypes are BLUPs, which are not published per line. For each trait, they fitted lmer(trait ~ group + (1|line)) to the day-28 images, with the imaging group as a fixed effect and the line (the 162 DGRP lines plus the two controls) as a random effect. They then used the line BLUPs in a PLINK GWAS with PC1-PC4 and the
Wolbachia status as covariates. The GWAS used 14 of the 16 traits: EccSD (H
2 < 0.05) and RadMinMean (not correlated with human scoring of the eyes) were left out. Recomputed with this model from the same images (the uploaded data plus the control images), the BLUPs reproduce the paper's heritabilities (15 traits with H
2 ≥ 0.05, 14 between 0.1 and 0.25). The day-4 BLUPs reproduce the paired t-tests of day 28 vs day 4 of Fig. 2c exactly (for example, area mean t = 3.0398, P = 0.0058). The DGRPool line means are not group-corrected, and their rank correlation (Spearman) with the recomputed BLUPs is 0.68 to 0.81, depending on the trait.
Also not uploaded, because the per-line values are only shown in figures or are not DGRP data: the comparison of GMR, GMR>Aβ42, GMR>tau and GMR>Aβ42;tau flies in the
w1118 background (Fig. 2b), the wild-type eyes of six DGRP lines (Supplementary Fig. S4), the RFP reporter fluorescence of 23 lines (Fig. 3b) and the tau RT-qPCR of 13 lines (Fig. 3c).