A PAR-1-dependent orientation gradient of dynamic microtubules directs posterior cargo transport in the Drosophila oocyte.

Fly strains

Stocks were raised on standard cornmeal agar medium at 21 or 25°C. MT markers used in this paper were EB1-GFP, EB1-mCherry expressed ubiquitously (provided by H. Okhura, Wellcome Trust Centre for Cell Biology, University of Edinburgh, Edinburgh, Scotland, UK), and Tau-GFP 65/167 (provided by D. St Johnston, University of Cambridge, Cambridge, England, UK). Posterior cargo markers used in this paper were osk :MCP-GFP, Staufen-RFP (provided by D. St Johnston), and γ-tubulin37C–GFP (provided by S. Endow, Duke University Medical Center, Durham, NC). PAR-1 mutant flies used in this study (provided by D. St Johnston) were w-;par-1[6323]/CyO and w-;par-1[w3]/CyO.

Tissue preparation and imaging

Flies were prepared, and ovaries were dissected and mounted for imaging as previously described in Parton et al. (2010) . Imaging was performed either on a wide-field deconvolution system (DeltaVision CORE) from Applied Precision (with a microscope [IX71; Olympus], 100× 1.4 NA objective, 16-bit camera [Cascade II; Roper Scientific], and standard Chroma filter sets), an OMX-V2 prototype microscope designed by J.W. Sedat (University of California, San Francisco, San Francisco, CA) and built by Applied Precision ( Dobbie et al., 2010 ), or a spinning-disc confocal microscope (UltraVIEW VoX; PerkinElmer; with a microscope [IX81; Olympus], 60× 1.3 NA silicon immersion objective, and an electron-multiplying charge-coupled device camera [ImagEM; Hamamatsu Photonics]). Where required, image sequences were deconvolved with the SoftWoRx Resolve 3D constrained iterative deconvolution algorithm (Applied Precision). Basic image processing was performed with ImageJ (v1.43u; National Institutes of Health).

Immunofluorescence

Flies were prepared as previously described in Parton et al. (2010) . Ovaries were dissected into PBS, pH 6.0, with 8% EM-grade PFA. After 5 min, ovaries were then transferred to 200 µl PBS, pH 7.0, with 8% EM-grade PFA, vortexed with 200 µl heptane to permeabilize the tissue, and fixed for a further 10 min. Immunofluorescence labeling was performed as in standard protocols ( Cha et al., 2002 ; Rosales-Nieves et al., 2006 ). To detect tubulin modifications, we used antibodies, previously shown to work on Drosophila tissues, against acetylated tubulin (mouse monoclonal 6-11B-I; acetylated α-tubulin; Sigma-Aldrich) and glutamylated tubulin (clone 1D5 mouse hybridoma anti–Glu-α-tubulin; Synaptic Systems) at 1:250 and 1:300 dilutions, respectively ( Warn et al., 1990 ; Januschke et al., 2006 ; Rosales-Nieves et al., 2006 ). Primary antibodies were applied overnight at 4°C. The secondary antibody donkey anti–mouse Alexa Fluor 594 (Invitrogen) was applied at 1:500 for 2 h at room temperature. The tissue was mounted in VECTASHIELD (Vector Laboratories) and imaged immediately by spinning-disc confocal microscope (UltraVIEW VoX).

Colcemid treatment

The protocol for colcemid feeding was modified from Cha et al. (2002) : flies were fed for 1 d after eclosure, starved for 1 d, and then fed colcemid in yeast paste (200 µl of 0.1-mg/ml colcemid in distilled water added to 175 µl of dried yeast) for 4–6 h. Ovaries were dissected as normal.

Tracking and analysis

Through mid-oogenesis, the oocyte rapidly increases in size and accumulates yolk in the cytoplasm. This makes imaging increasingly challenging beyond stage 8. Furthermore, the EB1 protein is freely distributed in the cytoplasm as well as being associated with MT plus ends. This, combined with autofluorescence from the yolk, results in relatively poor contrast images of MT plus ends. Confocal methods increased contrast but proved insufficiently sensitive to detect the low EB1 signal. Wide-field deconvolution images were adequate for manual tracking of EB1 trajectories but resisted automatic segmentation and tracking without additional processing. For tracking analysis, oocytes expressing EB1-GFP were imaged on a wide-field deconvolution system (DeltaVision CORE) over three z planes twice per second with a pixel size of 97 nm. Image data were optionally denoised using the patch-based denoising algorithm ND-SAFIR (N-Dimensional–Structure Adaptative Filtering for Image Restoration; Boulanger et al., 2008 ) implemented in Priism (L. Shao, University of California, San Francisco, San Francisco, CA, and J.W. Sedat). Time series were deconvolved, the three z planes were maximally projected, and time points were equalized in SoftWoRx. Preprocessed image data were exported to 16-bit TIF format using ImageJ (64 V1.41) and imported into MATLAB (MathWorks). To obtain statistically relevant data for EB1 track directionality, a robust automated tracking algorithm was developed that combined probabilistic foreground extraction and Haar-like feature identification (referred to here as probabilistic feature extraction; Fig. S4 and Fig. S5) implemented in MATLAB (v7.9). Probabilistic feature extraction and tracking comprises four components as follows: (1) local median equalization to correct for uneven illumination while preserving local contrast; (2) foreground extraction by a probabilistic temporal median filter to facilitate the discrimination of moving features from a static background; (3) identification of EB1 foci using Haar-like feature energy (adapted from Yang et al., 2010 ); and (4) linkage into trajectories using the particle-matching algorithm of Sbalzarini and Koumoutsakos (2005) with a modified cost function to take account of the linear path of EB1 tracks. For visual inspection, trajectories were output to a text file formatted for the ImageJ Manual Tracking plugin. Local median equalization. To facilitate subsequent processing steps, all pixel values are normalized according to the ratio between the median for the local neighborhood (five times an EB1-GFP feature size, typically a 25 × 25–pixel patch) and the median for the whole image stack. Probabilistic temporal median filter. To separate moving foreground features from the uneven static background, a probabilistic temporal median filter was devised. Using the median value of the neighboring few frames (where the number is defined here as W, as described in the following paragraphs) as a model for the static background has been proposed by several authors ( Lo and Velastin, 2001 ; Cucchiara et al., 2003 ). Binary segmentation of pixels as either foreground or background throws away intensity information and is prone to error in the presence of noise. We therefore calculate a foreground probability image instead. We find that this soft segmentation preserves some of the original intensity information, facilitating subsequent processing steps. For the entire normalized image sequence (from the Local median equalization section), a crude estimation of intensity variation caused by noise σ was made: for each pixel, the SD over the first W frames and the last W frames was calculated, and the mean of these values was taken as a measure of σ. W was chosen to be three times the mean time taken for an EB1-GFP particle to cross a pixel, typically n = 15. The background intensity level, I bg (x,y,t), for each pixel at position x,y and time t was estimated by calculating the median over the surrounding W time points, i.e., from t − 0.5 × (W – 1) to t + 0.5 × (W − 1). The more the intensity value I(x,y,t) for a given pixel exceeds the static background value I bg (x,y,t), the more likely it is that a moving foreground feature (i.e., an EB1-GFP particle) is crossing the pixel. Instead of applying a binary cutoff, we therefore calculate a probability value P(x,y,t), which is the probability that the pixel is not a background feature: (1) P(x,y,t) = Q( − f) in which f(x,y,t) = [I(x,y,t) − I bg (x,y,t) − kσ]/σ is the excess intensity above background normalized to 1 SD and Q(f) = 0.5 × (1 − erf(f/√2)) is the Q function, which is the tail probability of the standard normal distribution. Eq. 1 results in a foreground probability close to 1 when k = 0 and the intensity is 3 SDs or more above the background level. To produce a probability image in which the foreground features are not all saturated, k can instead be set to k > 0, and we find that a value of k = 1 produces satisfactory results for further analysis (Fig. S2 and Fig. S3). Calculation of a particle probability image. Haar-like features were first used by Viola and Jones (2004) for face detection using a small number of critical features. A complex scheme for particle detection using Haar-like features was presented by Jiang et al. (2007) , but for our foreground probability images, we find we are able to use a very simple segmentation criterion (adapted from Yang et al., 2010 ). Here, we calculate the energy for a single, square Haar-like feature representing a typical EB1-GFP particle (energy is calculated using pixel gray values as described in Yang et al., 2010 ; Eq. 1 ). For a particle of diameter 2w + 1, the half-width of the Haar-like feature window is w and the half-width of the internal feature is w − 1, in which a typical value of w is w = 2. The resulting Haar-like feature energy images can be rendered less noisy by application of one final filtering step: a particle probability score is calculated as a Gaussian weighted geometric mean of the highest Haar-like feature energy scores in the x,y,t neighborhood of each pixel. The neighborhood radius was set to v max in x,y (v max is the maximum velocity for a particle) and one time point about the current time point t. Segmentation of the resulting particle probability images can be achieved by applying a simple highest percentile threshold (T; normally T ≈1% of the image area corresponds to EB1 foci) because the probability is highest when appropriately sized clusters of locally high intensity pixels persist within a distance v max over multiple frames. Trajectory determination. Trajectory determination was performed using the particle-matching algorithm of Sbalzarini and Koumoutsakos (2005) with a modified cost function: φ i j = δ ( p i , q j ) + δ v e c ( p i , q j ) + 1 2 [ ( m 0 ( p i ) − m 0 ( q j ) ) 2 + ( m 2 ( p i ) − m 2 ( q j ) ) 2 ] , δ ( p i , q j ) = ( x ˜ p i − x ˜ q j ) 2 + ( y ˜ p i − y ˜ q j ) 2 , and δ v e c ( p i , q j ) = ( x − p i − x − q j ) 2 + ( y − p i − y − q j ) 2 , in which notation is as previously described in Sbalzarini and Koumoutsakos (2005) , and the additional term δ vec is the change in vector associated with linking point p i with point q j . The validity of the automated tracking algorithm for determining EB1 track orientation was assessed by comparison with manually tracked data. The ground truth was determined by manually assigning the start and end points of all recognizable EB1 tracks over a 100-frame time series for both the “raw” time series data and the processed, segmented image data. Automatically determined and manually defined trajectories were plotted for comparison (Fig. S3, A and B). To analyze the results of tracking, data were imported into the ParticleStats environment ( Hamilton et al., 2010 ). ParticleStats directionality analysis tools were used to generate EB1 track overlays, net local directionality maps, rose diagrams, and radial histograms and to assess directional bias in EB1 tracks by the Rayleigh test of uniformity (assuming a circular normal distribution) and Watson two-sample test for comparing two samples of circular data ( Mardia and Jupp, 2000 ). The validity of the automated tracking algorithm to report directionality was also assessed quantitatively by comparing the output from ParticleStats for the manual versus the automatic tracking data (Fig. S3 C). Online supplemental material Fig. S1 shows that MTs in the oocyte lack posttranslational modifications associated with increased stability ( Fig. 2 ). Fig. S2 shows combined probabilistic foreground extraction and adaptive nonlocal means filter for automatic segmentation to detect EB1 tracks. Fig. S3 shows validation of automated detection and tracking of EB1 foci (see Materials and methods). Fig. S4 follows a MT over time showing that dynamic MTs support a consistent net bias in MT orientation ( Fig. 4 ). Fig. S5 shows the distribution of γ-tubulin, supporting the idea that nucleation occurs at discrete foci along the cortex but is absent from the extreme posterior ( Fig. 5 ). Video 1 shows a time series of Tau-GFP labeling dynamic MTs in the Drosophila oocyte ( Fig. 1 B ). Video 2 shows the localization of EB1-mCherry to the plus ends of Tau-GFP–labeled MTs. Video 3 shows EB1-GFP marking the plus ends of Tau-GFP–labeled MTs, revealing the highly dynamic nature of MTs in the oocyte ( Fig. 2 ). Video 4 shows Staufen-RFP moving on Tau-GFP–labeled MTs at the posterior of a stage 9 oocyte. Video 5 shows a zoomed region with a single Staufen particle moving along an MT ( Fig. 3 B ). Video 6 shows MT regrowth from discrete foci after UV inactivation of colcemid. Video 7 shows a magnified view of individual EB1-labeled foci nucleating MT after UV treatment ( Fig. 5 ). Video 8 shows “inappropriate” EB1-GFP–labeled foci nucleating MTs along the cortex of the extreme posterior in a par-1 hypomorph ( Fig. 6 A ). Video 9 shows a magnified region of Video 8, highlighting individual EB1 trajectories. Online supplemental material is available at http://www.jcb.org/cgi/content/full/jcb.201103160/DC1 .

Online supplemental material Fig. S1 shows that MTs in the oocyte lack posttranslational modifications associated with increased stability ( Fig. 2 ). Fig. S2 shows combined probabilistic foreground extraction and adaptive nonlocal means filter for automatic segmentation to detect EB1 tracks. Fig. S3 shows validation of automated detection and tracking of EB1 foci (see Materials and methods). Fig. S4 follows a MT over time showing that dynamic MTs support a consistent net bias in MT orientation ( Fig. 4 ). Fig. S5 shows the distribution of γ-tubulin, supporting the idea that nucleation occurs at discrete foci along the cortex but is absent from the extreme posterior ( Fig. 5 ). Video 1 shows a time series of Tau-GFP labeling dynamic MTs in the Drosophila oocyte ( Fig. 1 B ). Video 2 shows the localization of EB1-mCherry to the plus ends of Tau-GFP–labeled MTs. Video 3 shows EB1-GFP marking the plus ends of Tau-GFP–labeled MTs, revealing the highly dynamic nature of MTs in the oocyte ( Fig. 2 ). Video 4 shows Staufen-RFP moving on Tau-GFP–labeled MTs at the posterior of a stage 9 oocyte. Video 5 shows a zoomed region with a single Staufen particle moving along an MT ( Fig. 3 B ). Video 6 shows MT regrowth from discrete foci after UV inactivation of colcemid. Video 7 shows a magnified view of individual EB1-labeled foci nucleating MT after UV treatment ( Fig. 5 ). Video 8 shows “inappropriate” EB1-GFP–labeled foci nucleating MTs along the cortex of the extreme posterior in a par-1 hypomorph ( Fig. 6 A ). Video 9 shows a magnified region of Video 8, highlighting individual EB1 trajectories. Online supplemental material is available at http://www.jcb.org/cgi/content/full/jcb.201103160/DC1 .