High-resolution computational imaging of leaf hair patterning using polarized light microscopy


  • Authors disclose that there is no conflict of interest.

(e-mail grotewold.1@osu.edu).


The leaf hairs (trichomes) on the aerial surface of many plant species play important roles in phytochemical production and herbivore protection, and have significant applications in the chemical and agricultural industries. Trichome formation in the model plant Arabidopsis thaliana also presents a tractable experimental system to study cell differentiation and pattern formation in plants and animals. Studies of this developmental process suggest that trichome positioning may be the result of a self-forming pattern, emerging from a lateral inhibition mechanism determined by a network of regulatory factors. Critical to the continued success of these studies is the ability to quantitatively characterize trichome pattern phenotypes in response to mutations in the genes that regulate this process. Advanced protocols for the observation of changes in trichome patterns can be expensive and/or time consuming, and lack user-friendly analysis tools. In order to address some of these challenges, we describe here a strategy based on polarized light microscopy for the quick and accurate measurement of trichome positions, and provide an online tool designed for the quantitative analyses of trichome number, density and patterning.


Trichomes are specialized structures that differentiate from the aerial epidermis of many plants. Across different species, trichomes vary in shape and size, and fill different biological roles, ranging from the production and secretion of phytochemicals [e.g. in mint (Mentha longifolia) and tobacco (Nicotiana tabacum)] to UV and herbivore protection [e.g. in the air plant (Tillandsia variabilis) and stinging nettle (Urtica dioica), respectively] (Wagner et al., 2004). Because of their secretory and structural properties, trichomes also play important commercial roles in the pharmaceutical, chemical and even textile industries, given that cotton (Gossypium herbaceum) fibers are ovule trichomes (Peter and Shanower, 1998).

The model plant Arabidopsis thaliana (Arabidopsis) serves as an excellent system to study trichome development and de novo pattern formation. Whereas other developmental processes such as root hair formation appear to rely on pre-existing positional signals to determine cell fate, trichome patterning in Arabidopsis appears to be self-forming, regulated by networks of counteracting positive and negative transcription factors (Ishida et al., 2008; Marina Pesch and Hülskamp, 2009). Understanding the precise mechanisms underlying the formation of these patterns is the focus of a number of studies centered on several key regulatory players and their inter-regulatory interactions, including protein–protein and protein–DNA (Digiuni et al., 2008; Zhao et al., 2008; Morohashi and Grotewold, 2009).

A critical aspect of understanding trichome development and patterning is the ability to correlate phenotypic changes in trichome patterning with relevant mutant alleles. The resulting phenotypes can often be pleitropic or subtle, and can even show opposing effects with varying gene dosage (Larkin et al., 1999; Kirik et al., 2005). There is therefore a need to not only qualitatively describe trichome phenotypes, but also to accurately quantify the changes in trichome patterning across multiple genotypes, and in different genetic backgrounds.

Traditional methods for measuring differences in trichome numbers and density have largely relied on visual inspection and manual trichome counts (Hulskamp et al., 1994; Schellmann et al., 2002; Maes et al., 2008). Although simple in methodology, these techniques can be challenging in practice. The repetitive process of manually counting hundreds of often small and colorless trichomes across multiples leaves, replicates, and genotypes can be time consuming, and prone to human error. New techniques using 4D confocal microscopy (Bensch et al., 2009) and micro X-ray tomography (Kaminuma et al., 2008) provide highly accurate and automated trichome counts, as well trichome positional information which can be used for the quantitative analysis of trichome distribution and patterning phenotypes. One of the major strengths of these new techniques is the application of computer-aided analysis tools to interpret data and produce sophisticated modular comparisons of trichome distributions and densities across different leaf sections and developmental time points (Kaminuma et al., 2008; Genaev et al., 2012). However, despite providing clear advantages in data collection and analyses, these techniques have not yet been widely adopted for trichome patterning studies. Equipment and technical expertise requirements, as well as long data collection times, make these very sophisticated methodologies impractical for many laboratories. Another challenge is that much of the accompanying analysis software developed for these techniques is not readily available for use by others in the field, thus discouraging broader adoption of the technique.

Here, we describe an approach for the study of trichome patterning that provides highly quantitative positional analyses while also addressing the need for a low-cost and accessible method for trichome positional studies. Our method repurposes a previously described polarized light microscopy technique (Ballard, 1916) that exploits the polarizing (birefringent) properties of trichomes to produce high-contrast images of the trichome positions on leaves (Potikha and Delmer, 1995; Bischoff et al., 2010). In order to process these images, we developed analysis software that automatically extracts trichome positions from leaf pictures and produces quantitative data describing trichome counts, density and relative distances. We describe the use of this technique on Arabidopsis trichome development mutants, and provide the protocols and online analysis software necessary for the application of these tools to other mutants and/or species. We anticipate that the tools described here will significantly advance the high-throughput analysis of trichome numbers and distribution in a number of plant species.


High-contrast trichome imaging using polarized light microscopy

To increase the detectability of trichomes in leaf imaging, we used polarized light microscopy (PLM) on Arabidopsis leaves. PLM is a contrast-enhancing technique that exploits the birefringent properties of specimens by viewing them between cross-polarizing elements placed immediately after the light source and before the objective lens of a light microscope (Figure S1A). By turning the second polarizing filter (analyzer) at a perpendicular angle to that of the light source, only light that is polarized by birefringent tissue is allowed through the analyzer and into the objective, giving birefringent sections of a sample a bright appearance (Figure S1B, C).

To determine the effectiveness of using PLM for trichome patterning analysis, we grew Arabidopsis plants and visualized them under our microscope set-up (Figure S2). Leaves from different genotypes were collected in developmental order and were processed to clear chlorophyll before being placed on microscope slides for imaging and storage (see 'Experimental Procedures'; Gudesblat et al., 2012). The resulting material produced extremely high-contrast images of illuminated trichomes over dark backgrounds: ideal for trichome analysis (Potikha and Delmer, 1995; Bischoff et al., 2010; Figure S1C; Movie S1).

trichomenet: computer-aided trichome analysis

In order to process the high-contrast trichome images obtained from PLM of Arabidopsis leaves, we developed a custom-built software tool capable of storing, detecting and analyzing trichome positional data: trichomenet. trichomenet was designed to function on all major computer platforms through a standard web browser, and is available to the public for online use at http://www.trichomenet.com or for download and set-up of a private local server, with instructions available at the same web address.

Image upload

To begin using the system, leaf images collected during PLM (Figure 1a) are uploaded via a graphical user interface to a trichomenet server, where users can organize their images according to experimental variables (categories: e.g. genotype, treatment, developmental stage, etc.), and name them according to leaf and replicate number. Once uploaded, trichomenet offers users the option of either performing manual or automatic marking of trichomes on each leaf.

Figure 1.

trichomenet analysis workflow. Cleared Arabidopsis leaves are inspected under polarized light microscopy. (a) Images of leaves are uploaded to the trichomenet server. (b) Trichomes are either manually or automatically marked, and then saved as coordinates over the image (red and blue circles for marginal and laminal trichomes, respectively). (c) trichomenet superimposes leaf trichome positions and creates density heat maps and other analyses, including trichome density, next-neighbor distance and all trichome distance analyses. An option for raw data output is provided for data manipulation and additional analysis outside the software.

Trichome marking

Manual trichome marking in trichomenet is a guided process by which users are shown each of their uploaded leaf images and prompted to click on various leaf features, including leaf tip, marginal and laminal trichomes. As users click over each feature, different colored circles appear to mark the xy locations of each mouse click on the leaf picture (leaf tip in green, and marginal and laminal trichomes in red and blue, respectively). In this manner, trichome positions are extracted from each image and stored as a Cartesian coordinate that is later used for interleaf comparisons and analyses (Figure 1b).

trichomenet also offers users the option for automatic trichome marking via processing tools in ImageJ (fiji, http://fiji.sc/wiki/index.php/Fiji). The image-processing algorithm in fiji scans the leaf pictures and recognizes trichomes based on pixel brightness and object size cut-offs (Abramoff et al., 2004; Schindelin et al., 2012). Users are given the option to change the detection limits of automatic trichome marking via a built-in sensitivity slider to adjust for images with varying trichome brightness and contrast. Although automatic marking is highly dependent on image contrast and quality, with a little adjustment to the sensitivity slider our experiences were fairly positive, achieving marking accuracies as high as 80–90% compared with manual methods. A more detailed comparison between the manual and the automatic marking methods is discussed below.

Trichome positional analyses

Trichome density

The strength of trichomenet as an analytical tool comes from the ability to compile trichome positional data from multiple leaves for robust trichome count comparisons and sophisticated patterning analyses. The most visual examples of these are the whole leaf trichome density heat maps. trichomenet uses the coordinates of marked leaf tips to digitally superimpose leaves, collapsing trichome positional coordinates from multiple leaves into a single plane. The resulting data set is then divided into discreet regions (boxes) on an xy plane in which local trichome density is calculated and a color is assigned to each box based on a user-defined heat index. The resulting virtual leaves created by this analysis provide a snapshot of trichome patterning and density over the entire leaf surface for comparisons across different leaf numbers and genotypes (Figure 1c).

Trichome distances

In addition to reporting whole-leaf trichome density values, trichomenet also uses compiled trichome xy positioning data to output distribution graphs of trichome next-neighbor distances and distances between all trichomes (Clark and Evans, 1954; Larkin et al., 1996). These distance values, originally used to establish the non-random distribution of Arabidopsis trichome patterning (Larkin et al., 1996), can provide a quantitative measure of distribution phenotypes (Kaminuma et al., 2008), distinguishing between short- and long-range effectors (Pesch and Hülskamp, 2011). trichomenet calculates and presents the distances as histograms on each analysis page, and compiles the relevant values for export into excel or other graphing software. The outputs from trichome distance analyses are shown in more detail below.

User-defined options and data output

In order to provide flexibility throughout the analysis process, trichomenet offers users a variety of visual and data analysis options. For example, the heat map generator in trichomenet allows users to define the length and width of the box regions used to reconstruct leaves, as well as the option to include box edges, trichome position marks, leaf margins, heat index values, and numerical trichome densities in each figure. Users can also define the bin sizes and data ranges for trichome next-neighbor histograms and distance between all trichome histograms. There is also the option of restricting analysis to laminal trichomes, as well as the option of using marginal trichomes as leaf edges for heat map generation. Finally, even though results for each of these analyses are viewable on each summary page, users have the option of exporting all the analysis data as a comma separated value (.csv) file, including the raw trichome counts, xy positions and distance calculations (Figure 1c).

Comparison of automatic and manual trichome marking methods

To evaluate the effectiveness of the automatic trichome marking system in trichomenet, we compared the trichome marking results of leaves analyzed by both the automatic and the manual methods. Representative pictures of a Landsberg erecta wild-type leaf (Ler) included in the comparison highlight the advantages and drawbacks of automatic marking approaches (Figure 2a). Although the image analysis in trichomenet seems to have properly recognized most of the leaf trichomes, it has also erroneously marked a few bright dust particles and some petiole trichomes purposefully excluded from manual analysis (examples marked with white arrows in Figure 2a). Additionally, automatic marking is not able to distinguish between marginal and laminal trichomes (distinguished by red and blue dots, respectively, in the manual marking of Figure 2a), but instead marks all visible trichomes equally, thus eliminating the option for separate (marginal/laminal) analyses. Interestingly, the automatic marking in trichomenet (light-blue dots) also recognized a few trichomes that were missed by our human screener, and were not included in the manual marking (examples marked with red arrows in Figure 2a), highlighting an advantage of the deterministic computer marking (for a larger superimposition of results from both methods, see Figure S3).

Figure 2.

Comparison of manual and automatic trichome marking approaches. (a) Representative picture of an unmarked Arabidopsis wild-type Ler leaf (left), a manually processed leaf (center) with marginal and laminal trichomes marked separately (red and blue, respectively), and an automatically processed leaf (right, light blue). Automatic marking is relatively robust, but incorporates artifacts such as dust particles and petiole trichomes that are filtered by manual methods. (b) Table of trichome counts between genotypes comparing manual and automatic marking methods. The errors introduced by automatic marking are smaller than the biological differences between genotypes, suggesting that automatic marking could be used for phenotypic screens.

Overall, the trichome count numbers from automatic marking methods were similar to the manual counts, showing approximately 20% or less difference between the two (Figure 2b). In each case, the error introduced by opting for automatic trichome marking was much smaller than the actual biological differences in trichome counts between leaves of the same genotype. These results suggest that automatic marking methods, though not perfect, could be a viable screening tool for large-scale trichome studies: for example, for uncovering mutants among the populations routinely grown at the Arabidopsis Biological Resource Center (ABRC). Given the relative advantages and shortcomings of each of the trichome marking approaches, the methods have been programmed to work together, using for example, automatic marking for the majority of trichomes, followed by manual marking to fix any mistakes.

Analysis of the ttg2-1 trichome patterning mutant

To demonstrate the functionality of trichomenet, we examined the known trichome developmental mutant TRANSPARENT TESTA GLABRA 2 (ttg2-1), and compared it with its Ler wild-type background. The ttg2-1 mutant encodes a WRKY transcription factor functioning downstream of the GL1-GL3-TTG1 trichome initiation complex (Ishida et al., 2007; Zhao et al., 2008). Plants with defects in the TTG2 gene do not have marginal trichomes, and develop fewer and unbranched laminal trichomes (Johnson et al., 2002). Despite their developmental arrest, trichomes of ttg2-1 mutant leaves showed strong birefringence under PLM, making them good candidates for analysis (Figure S4). Heat maps of both genotypes showed lower laminal trichome densities in ttg2-1 mutant leaves when compared with Ler, particularly in the basal regions of leaves, where the higher relative growth is likely to exacerbate the low-density phenotype (Figure 3a; Remmler and Rolland-Lagan, 2012). Whole-leaf trichome densities differences were also graphed, showing the laminal trichome density in ttg2-1 to be approximately 50% lower than in the Ler wild type, thereby confirming results in other studies (Figure 3b; Johnson et al., 2002).

Figure 3.

Trichome density and heat-map analysis of Ler and ttg2–1 leaves. (a) Heat maps of ttg2–1 and Ler wild-type leaves show differences in trichome densities over the leaf surface. The mutant ttg2–1 appears to be less densely covered in trichomes, particularly over the basal regions of leaves. (b) Whole-leaf trichome density values show that ttg2–1 leaves are approximately 50% less trichome dense than Ler wild-type controls.

To investigate the patterning differences between the mutant and wild type, we compared trichome distance distributions. Because of the lower number of laminal trichomes in ttg2-1 mutant leaves, our initial comparison of trichome distances showed ttg2-1 distance distributions of significantly less amplitude (Figure 4a). To better compare the trichome distance distributions between plants with different trichome densities, we exported the results from trichomenet and produced normalized distributions (i.e. distributions as a fraction of the total number of distances) for ttg2-1 and Ler leaves. These normalized graphs showed a shift in ttg2-1 distance distributions towards longer distances between trichomes (Figures 4b and S5). To confirm that the ttg2-1 shift towards longer trichome distances was a consequence of lower trichome numbers, and not an effect on trichome patterning, we analyzed simulated leaves generated by randomly removing trichome data points from Ler leaves to create leaves with a trichome density equal to that of ttg2-1 leaves. The results from 1000 rounds of simulations showed a shift towards a larger trichome distance distribution similar to that of ttg2-1 leaves, suggesting no additional trichome patterning phenotypes for ttg2-1 plants other than reduced density (Figure 4c).

Figure 4.

Trichome distance analysis of Ler and ttg2–1 leaf 7. trichomenet produces trichome distance distributions based on coordinate distances between marked laminal trichomes. (a) A comparison of ttg2–1 and Ler trichome distance distributions show lower raw counts because of a lower trichome number. (b) Normalized distribution curves show a ttg2–1 shift towards longer trichome distances. (c) Distance distributions comparing ttg2–1 with simulated Ler leaves with equal trichome numbers suggest a shift in distance is simply a consequence of lower trichome numbers. Results from other leaves are presented in the Supporting Information.

Application to other plant species

To determine whether the method described here for selectively detecting leaf hairs can also be applied to plant species other than Arabidopsis, we conducted leaf clearing and PLM experiments on leaves of a variety of plant species (Figures 5 and S6). The technique appeared to illuminate trichomes of many species, including cotton (Gossypium herbaceum), maize (Zea mays), salvinia (Salvinia natans) and squash (Cucurbita maxima), but showed background in the vasculature for other species such as mint (Mentha sachalinensis), drosera (Drosera capensis) and geranium (Pelargorium graveolens) (Figure 5), or incomplete clearing that stopped light from properly penetrating the tissue, as observed in oregano (Origanum vulgare) and tomato (Solanum lycopersicum) (Figure S6, geranium). These results suggest that trichomenet analysis tools, originally designed for Arabidopsis, could also be adapted for use in other species.

Figure 5.

Polarized light microscopy of plant species: pictures of leaves from various species processed with leaf clearing methods and viewed under polarizing light microscopy (see 'Experimental Procedures'). Leaves from cotton, maize and salvinia are good candidates for trichome visualization with polarized light. Tomato and oregano leaves have strongly birefringent vasculature, which complicates the process of trichome marking, thus making them poor candidates for this protocol. Information on more species is presented in the Supporting Information.


We have described a technique for trichome patterning studies that generates highly quantitative trichome counts and positional data. We demonstrate that polarized light microscopy techniques can yield high-contrast trichome images on cleared leaves of Arabidopsis and many other plant species. This approach overcomes some of the limitations of previous approaches by providing a low-cost and relatively quick method for trichome position marking: our results were obtained through an inexpensive modification of a standard light dissection microscope using parts available from any camera store (see Figure S2 and 'Experimental Procedures'). In addition, we provide a free analysis tool that is directly accessible on the web and available for download and private hosting.

trichomenet analysis of ttg2-1

We demonstrate the functionality of trichomenet using the ttg2-1 Arabidopsis trichome development mutant as an example. Our results show that in addition to having under-developed single branched trichomes, ttg2-1 mutant leaves are also approximately 50% less dense in trichomes than their Ler wild types. Our trichome distance distribution analyses indicate, however, that ttg2-1 trichome patterning is largely unaffected, exhibiting only changes brought on by a lower trichome density.

Manual versus automatic trichome marking

In our comparison of the automatic and manual trichome marking methods, we found automatic approaches to be reasonably accurate in detecting birefringent trichomes (Figure 2a), with marking errors that were smaller than the trichome count differences among replicate leaves of a genotype (Figure 2b). However, without proper care, unsupervised automatic marking methods also have the potential for larger errors. Images with low contrast or with contaminating dust can produce erroneous results, and should be inspected before proceeding to other analyses. These errors, however, are avoidable with proper leaf processing and imaging. Given the strengths of each trichome marking system, our recommendation is to take a ‘hybrid’ approach by processing leaves through automatic marking to mark the bulk of trichomes and then compensating for any errors manually.

Although the automatic marking functionality in trichomenet requires leaf clearing and polarized light microscopy, we have found that images of green tissue immobilized with double-sided tape on glass slides followed by manual marking may also be used. However, we note that this approach does not preserve leaves and thus requires all image data to be collected and analyzed on the same day.

To provide the tools for users to customize and/or add new features, we have released the trichomenet source code under a General Public License in which improvements based on our code are similarly made available to the public. In this manner, we hope to provide a platform for other trichome researchers to rapidly develop and share their analysis techniques with the plant science community. We invite researchers to assess whether trichomenet could be incorporated into their workflow, and have provided example data on the website for users to test software features.

Experimental Procedures

Plant growth

To ensure even germination, seeds were surface sterilized with a 60-s wash in 70% EtOH followed by a 10-min soak in 50% bleach solution. Seeds were washed between three and five times using sterile distilled water and placed in a 4°C chamber in the dark for 3 days for seed stratification. Seeds were then placed on soil or, alternatively, can be plated on MS plates (MS salts, B5 vitamins, pH 5.7, 0.7% Phytagar and 2% sucrose) and grown at 24°C in continuous white light for a week before transferring them to soil. Plants were grown using Sunshine™ (http://www.sunshineadvanced.com/) peat-based potting mixture under 16-h light/8-h dark cycles at 22°C in Conviron® (http://www.conviron.com/) growth chambers, with a light intensity of around 120 μmol m−2 s−1 for 4 weeks. This technique also works on plants as young as 2 weeks old. However, plants older than 4 weeks do not work well as leaves become too large to photograph with our dissecting scope and become too curled to flatten for microscopy.

Leaf clearing

Leaves were harvested and cleared as described by Gudesblat et al. (2012). Briefly, leaves were placed in a series of solutions, including: 95% ethanol (1–2 days), 1.25 m NaOH:EtOH (1 : 1 v/v) solution for 2 h at 60°C, followed by 85% lactic acid. To speed up the process, multiple leaves were processed in parallel using multi-well plates and a mosquito screen filter (Figure S7). Leaves were mounted on glass slides with the abaxial side facing up and covered with glass slips. Leaves with curling were placed adaxial side up on cover slips and unfolded with tweezers before placing the cover slip on a fresh slide. Leaves can be stored for many weeks in slide boxes or can be imaged immediately.


Pictures of cleared leaves were taken as described by Bischoff et al. (2010). Essentially, pictures were taken on a Nikon SMZ 1500 dissecting scope (http://www.nikon.com) modified with polarizing filters (12.5–mm-thick Polaroid® Polarizing Material, http://www.amazon.com) placed under the glass stage and attached to the top of the objective. Although several relatively inexpensive polarizing kits are available from microscope manufacturers, it is also possible to construct one's own at a fraction of the cost by purchasing polarizing filters from any camera or online store (see Figure S2 for pictures). All leaf pictures were taken at the same magnification.


trichomenet was built by our laboratory and is released in an online public and private downloadable from at (http://www.trichomenet.com) under a General Public License. The online tool follows Web 2.0 standards, and is coded in html 5 and javascript, making it compatible with Mac and PC as well as tablet computers. The source code is available from the same URL.


We would like to thank Dr Rebecca Lamb for helpful discussions throughout this project and her suggestion of polarized light microscopy as an imaging method for trichomes. We would also like to thank Joan Leonard for her help in gathering samples of plant species other than Arabidopsis. Finally, we would like to thank Dr Miroslava Zhiponova for sharing her leaf-clearing protocol with us. Support for this project was in part provided by NSF DBI-1049341 to E.G. and J.B.