Recent advances in imaging equipment and probes have enabled the acquisition of huge numbers of bioimages. One researcher can easily inspect a few hundred images. However, when the number of images increases to thousands, millions, or billions, no individual could process them alone. Cell imaging techniques in particular yield large-scale systematic image data characterizing gene and protein localizations [1]. In these situations, computer assistance becomes increasingly important for image inspection. To develop and evaluate computer programs for image inspection, a benchmark image dataset is crucial. For this purpose, the benchmark dataset should not be a miscellaneous image collection but systematically-captured images designed with a precise aim. For example, the MitoCheck consortium focused on identifying human genes involved in mitosis progression and collected two days of time-lapse images of 67 average HeLa cells expressing histone-GFP with a chemically synthesized short interfering RNA (siRNA) knockdown system with 21,000 protein-coding genes [2]. This time-lapse image dataset is freely available (http://www.mitocheck.org/). In addition, Dr. Murphy’s laboratory at Carnegie Mellon University acquired HeLa cell images with fluorescently-labeled cell nuclei, nucleoli, endoplasmic reticulum (ER), Golgi bodies, lysosomes, mitochondria, plasma membrane, endosomes, actin microfilaments, and microtubules, and proposed organelle recognition algorithms for the HeLa image dataset [3]. Benchmark images of 80–91 HeLa cells for every 10 markers are freely available at Dr. Murphy’s laboratory website (http://murphylab.web.cmu.edu/data/2Dhela_images.html). Public release of these image datasets will contribute to the future development of computational image analytics.

In the plant sciences, a few fluorescent microscopic image databases have focused on intracellular structures, such as Plant Cell Imaging (http://deepgreen.stanford.edu/) [4], the Illuminated Plant Cell (http://www.illuminatedcell.com) [5], and the Plant Organelles Database (http://podb.nibb.ac.jp/Organellome/) [6]. These databases have provided beautiful pictures of plant cell structures and cellular dynamics and have potential value as resources for model analysis [7]. However, these image databases contain images of different plant cell types, complicating analysis.

Guard cells of plant stomata have become a model system for characterizing signal transduction mechanisms from environmental perception to turgor movement [8]. Previous biological studies of guard cells have proven that intracellular structures are required for healthy stomatal movement [9–11]. To comprehensively visualize guard cell intracellular dynamics during stomatal movement, we have already released microscopic images of 50–60 pairs of Arabidopsis guard cells fluorescently labeled with 18 kinds of organelle markers in the Live Images of Plant Stomata (LIPS) database [12]. However, with the first version of the database, visualizing intracellular three-dimensional configurations and/or mining biologically meaningful information such as the relationship between intracellular configuration and stomatal apertures was inconvenient.

We have updated the LIPS database with additional datasets. The original serial optical sections are still available as LIPS dataset I, and volume-rendering and aligned-image datasets are newly released as LIPS datasets II and III, respectively. In addition, a database table, named LIPService, was newly established to easily inspect the relationship between intracellular configuration and cell status/morphology. In this article, we describe the updated content and utility of LIPService. This database will serve as an image data mining tool, a web-based educational resource, and a benchmark dataset for plant intracellular structures in plant guard cells.

Arabidopsis thaliana plants were grown on soil-vermiculite in growth chambers at 23.5°C, at 60% relative humidity, and with a 12/12-h light/dark cycle using 100 μmol m^-2 s^-1 white lights. Epidermal strips of fully-expanded rosette leaves of 4–5-week-old plants were placed onto the inverted platform of an fluorescent microscope (IX70; Olympus, Tokyo, Japan) equipped with an oil immersion objective lens (UplanApo 100×/1.35 oil; Olympus), a spinning disc confocal unit (CSU10; Yokogawa, Tokyo, Japan), and a cooled CCD camera (CoolSNAP HQ; PhotoMetrics, Tucson, AZ, USA). GFP/YFP imaging used 488 nm excitation lasers (HPU-50101-PFS2; Furukawa, Tokyo, Japan) and 524–546 nm emission filters (FF01-535/22-25; Semrock, Rochester, NY, USA). For RFP imaging, 561 nm excitation lasers (85-YCA-025-040; CVI Melles Griot, Albuquerque, NM, USA) and 524–546 nm emission filters (FF01-590/20-25; Semrock) were used. Serial optical sections (0.5 μm intervals) were taken with Metamorph 7.0 software (Molecular Devices; Sunnyvale, CA, USA). All raw images were 500 × 500 pixels (32 × 32 μm). These fluorescent markers were chosen based on reliability and validity in previous plant cell biological studies. All transgenic plants used in the LIPS database had healthy growth and stomatal movements with a diurnal cycle [12]. We did not capture obviously abnormal guard cells with artifacts because of mechanical damage in sample preparation. To make volume-rendering data, we used an ImageJ plug-in 3D volume viewer (in ImageJ menu Plugins-3D-3D Viewer; http://3dviewer.neurofly.de/) and converted data into animated GIFs with ImageJ (http://rsbweb.nih.gov/ij/). For aligned images, we used our original KBI ImageJ plugins. For detailed procedures, please refer to our protocol [13]. The images and HTML files are located on our in-house web server. The search function is implemented with Google Custom Search (http://www.google.com/cse/). The LIPService uses Google Fusion Tables (http://support.google.com/fusiontables/).

In this article, we describe how to use the LIPS database website and additional datasets. The newly released volume-rendering dataset II will be useful for education. In addition, the aligned image dataset III provides a useful benchmark dataset for developing an automated plant organelle recognition/classification. Using the LIPS dataset, in fact, we have already reported a framework for image data mining of intracellular dynamics during stomatal movement [12] and improving an automated classification of open/closed stomatal states [16]. We believe the updated LIPS database is a model service to allow biological researchers without specialized knowledge of database system development to easily publish their own imaging data.

LIPS database is freely available at http://hasezawa.ib.k.u-tokyo.ac.jp/lips/index.html. The latest information is continuously released from the LIPS Twitter account (https://twitter.com/lipsDb).