Plasticity and robustness of pattern formation in the model diatom Phaeodactylum tricornutum


  • Mathieu Vartanian,

    1. Biomineralization and Morphogenesis Group, CNRS UMR 8186, Ecole Normale Supérieure, 46 rue d’Ulm, 75230 Paris Cedex 05, France
    Search for more papers by this author
  • Julien Desclés,

    1. Biomineralization and Morphogenesis Group, CNRS UMR 8186, Ecole Normale Supérieure, 46 rue d’Ulm, 75230 Paris Cedex 05, France
    Search for more papers by this author
  • Michelle Quinet,

    1. Biomineralization and Morphogenesis Group, CNRS UMR 8186, Ecole Normale Supérieure, 46 rue d’Ulm, 75230 Paris Cedex 05, France
    Search for more papers by this author
  • Stéphane Douady,

    1. Laboratoire Matière et Systèmes Complexes (MSC), UMR 7057 CNRS & Université Paris-Diderot, 10 rue Alice Domon et Léonie Duquet, 75205 Paris Cedex 13, France
    Search for more papers by this author
  • Pascal J. Lopez

    1. Biomineralization and Morphogenesis Group, CNRS UMR 8186, Ecole Normale Supérieure, 46 rue d’Ulm, 75230 Paris Cedex 05, France
    Search for more papers by this author

Author for correspondence:
Pascal J. Lopez
Tel: +33 1 44 32 35 35
Fax: +33 1 44 32 39 35


  • • Understanding the morphogenesis of mineralized structures found in shells, bones, teeth, spicules and plant cell walls is difficult because of the complexities underlying biomineralization, and the requirement of accurate models for pattern formation.
  • • Here, we investigated the spatial and temporal development of siliceous structures found in a model diatom species, Phaeodactylum tricornutum, for which the entire genome has been sequenced and transformation is routine.
  • • Analyses of pattern formation revealed that the process of silicification starts from a ‘π-like’ structure that controls the spatial organization of a sternum upon which regular instabilities are initiated and developed. Detailed analyses also demonstrate that morphogenesis of silica is nonuniform. We also tested the sensitivity of pattern formation to perturbation of proton pumps, and found that selective inhibitors of H+-V-ATPases affect silica biomineralization both quantitatively and qualitatively. Morphometric analyses of valves purified from isogenic populations of cells show that the morphometric noise of several traits is under exquisite regulation, explaining why the overall valve pattern is reproducibly maintained.
  • • Altogether our analyses demonstrate that silica morphogenesis is a robust but nonuniform process, and allow us to propose a model for the dynamic growth of materials within a spatially controlled geometry.


Biomineralized structures, most commonly made of calcium carbonate, calcium phosphate and silica, are found in several kinds of organism and have various roles. For example, silica structures have been proposed to be used by some marine unicellular photosynthetic organisms as defense against grazers (Raven & Waite, 2004), and to help protect against virus infection. In higher plants they may also protect against desiccation (Neumann, 2003). Studies of the formation of mineralized structures should therefore offer opportunities for understanding intriguing aspects of cell and developmental biology in both unicellular and multicellular organisms and could help to decipher their evolutionary origins. However, the biology underlying biomineralization and the process leading to the formation of mineralized patterns are not yet completely understood because of the chemical and biological complexity of the systems (Ben-Jacob & Garik, 1990; Wilt, 2005), and because of the phenotypic plasticity frequently encountered in natural populations of cells or organisms (West-Eberhard, 1989). Such plasticity, which refers here to the ability of a genotype to exhibit alternative morphological characteristics and plays an important role in evolution by governing or modifying developmental pathways, can hinder the description of the morphotype of a whole population. Nevertheless the degree of alteration of a pattern in response to intrinsic or extrinsic factors is expected to provide valuable information about the robustness of the pattern formation process and to determine to what extent a pattern is maintained between individuals of a same population.

The extent of phenotypic plasticity of the aesthetic silicified thecae of diatoms is of major importance because species classification still remains largely on features and pattern of their silica shells, even if the estimation of the diversity of diatom populations has now gained by more systematic analyses of phylogenetic information (Medlin & Kaczmarska, 2004). In addition, because of their specific lifestyle, substantial variations in cell shape and size have frequently been observed in natural populations of diatoms or in cultures (Theriot & Stoemer, 1984; Mann, 1999). For example, variations in the patterns of valve have been observed among seasonal populations of the same species (Gallagher, 1982). Salinity was shown to alter the valve morphology of several kinds of freshwater and marine diatoms, and to modify the degree of silica polycondensation (Schultz, 1971; Geissler, 1986; Vrieling et al., 1999b, 2007). Control of phenotypic plasticity in diatoms and how the pattern is buffered against intrinsic or extrinsic perturbation remain key questions. In other words, it remains particularly important to understand precisely the robustness of the morphogenesis process in diatoms. However, such questions will remain difficult until more quantitative morphometric measurements with appropriate biological species and clonal cultures are developed, and until accurate models for pattern formation are proposed.

To explain the formation of the three-dimensional network of finely assembled silica nanoparticles that form the diatom valve, several generally accepted models based on nonequilibrium growth have been developed. Diffusion limited aggregation models (DLA) were proposed to explain the formation of the dendrite-like structures often observed in both centric and pennate diatoms (Gordon & Drum, 1994; Parkinson et al., 1999). Such models involve diffusion of silica particles from the margin of the silica deposition vesicle (SDV), the acidic compartment (Vrieling et al., 1999a) within which the deposition process occur, and precipitation on a uniform nucleating structure. More recently, another model was proposed to explain the formation of the repetitive areolae structures arranged in hexagons found in Coscinodiscus species (Sumper, 2002). The repetition of a phase separation process, but in slightly different conditions (i.e. decrease in the concentration of silicic acid or pH), may lead to a decrease in the size of the micelles and formation of smaller pores (Lutz et al., 2005). Physical models of phase-separating fluids, under the control of confinement and local external fields, have recently been tested and shown to generate a wide variety of diatom-related patterns (Lenoci & Camp, 2008). These models predict that the pattern formation has to be sensitive to the reactivity of the silica particles/silicic acid (Iler, 1979; Coradin & Lopez, 2003) and the organic molecules (i.e. long-chain polyamines, silaffins or silacidin) (Kroger et al., 1999, 2000; Lutz et al., 2005; Kroger, 2007; Wenzl et al., 2008), implying a pH dependence. In vitro experiments showed that the nature and speed of nanoparticle formation and assembly in the presence of some LCPA/silaffins were pH-dependent. Recently, a computer evolution model that also integrates the importance of cytoskeletal elements was developed to model morphogenesis of the valve (Bentley et al., 2005).

Aspects of pattern formation in diatoms have also been studied from early events up to the completion of the mature valve, using combinations of scanning and transmission electron microscopy techniques performed from pennate (bilaterally symmetrical) and centric (radially symmetrical) species (for reviews, see Volcani, 1981; Pickett-Heaps et al., 1990; Round et al., 1990; Schmid, 1994; Cox, 2002). For centric species, many interesting examples describing the fine silica structures and the formation of the valve pattern exist, including a recent study for the model species Thalassiosira pseudonana (Hildebrand et al., 2006), for which the genome has been sequenced and proteomic and genetic approaches have been applied to identify cellular and protein components involved in silicification (Frigeri et al., 2006; Poulsen et al., 2006, 2007; Sumper & Brunner, 2008). As previously mentioned, many studies of the valve morphogenesis also exist for the pennate species. In Navicula pelliculosa, transmission electron microscopy of thin sections and whole mounts of forming valves showed that development was centrifugal, raphe ribs being formed first, followed by the initiation from these of transapical ribs and pores (Reimann et al., 1966; Chiappino & Volcani, 1977). Another well-documented example corresponds to a series of studies performed by Pickett-Heaps and co-workers on Pinnularia (Pickett-Heaps et al., 1979a,b), which brought information on the variation in the position of cytoplasmic organelles and microtubules during the valve morphogenesis process.

Phaeodactylum tricornutum is an important diatom species because it is the first pennate diatom for which the complete genome information is available (Bowler et al., 2008) and molecular tools exist (Lopez et al., 2005; Montsant et al., 2005). P. tricornutum is also an interesting model to study the development of the silica valve, because, contrary to most diatoms, it does not have an absolute requirement of silicic acid for growth (Lewin et al., 1958; Borowitzka & Volcani, 1978; De Martino et al., 2007), allowing one to control the beginning of the silicification process, and to obtain valves that have the same life history, by simply regulating the availability of silicon in the growth medium. This singularity should also facilitate the identification of genes and pathways involved in silica nanopatterning. Moreover, the valve pattern of P. tricornutum is also relatively simple, making it an interesting paradigm to compare its development to current models on pattern formation, and to analyze the robustness of the pattern formation process.

Materials and Methods

Culture conditions

The selected P. tricornutum Bohlin species for which we could reproducibly obtain a large proportion of oval cells (over 95%) is the clone 1090-1a from the Culture Collection of Algae at the University of Göttingen. The diatom cells were cultured axenically in an artificial seawater medium in plastic flasks, at 16°C and under a light : dark regime (16 : 8 h). Cultures were performed in polycarbonate bottles in an enriched artificial seawater medium, as follows: NaCl, 410 mm; MgSO4·7H2O, 20 mm; MgCl2·6H2O, 20 mm; KCl, 10 mm; KNO3, 3 mm; CaCl2, 9.9 mm; Na2EDTA, 0.03 mm; glycylglycine, 3 mm; Fe-NH4-citrate, 1.9 µm; H3BO3, 8.1 µm; ZnSO4·7H2O, 1.04 µm; CuSO4·5H2O, 0.41 µm; Na2MoO4, 0.49 µm; CoCl2·6H2O, 0.42 µm; MnCl2·4H2O, 0.5 µm; Na2tartrate·2H2O, 6.5 µm; O2Se, 21.6 nm; KBr, 33.6 µm; SrCl2·6H2O, 3.75 µm; NaF, 10 µm; NiSO4·6H2O, 0.65 µm; Na3VO4, 9.8 nm; CrK(SO4)2·12H2O, 0.4 nm; K2HPO4, 0.3 mm; thiamine-HCl, 0.1 µg l−1; B12, 50 µg l−1; biotin, 50 µg l−1; Tris, 8.25 mm; pH was adjusted before autoclaving to 8.0 by adding HCl. When present, silicic acid was added in the middle of the light period to a final concentration of 150 µm. We also verified, by the molybdate yellow-blue method (Iler, 1979), that the concentration of soluble silicon in the medium did not vary by > 10% during the whole experiments. Bafilomycin A or concanamycin A (Sigma-Aldrich, Lyon, France) was used at the indicated final concentrations.

Constructs and generation of transgenic diatoms

To get accurate cell size measurements, transgenic diatoms expressing a membrane localized EYFP were developed. pEYFP-Mem (Clontech, St-Germain-en-Laye, France) vector encodes a fusion protein that consists of the N-terminal 20 amino acids of neuromodulin and a yellow-green fluorescent variant of the enhanced green fluorescent protein (EYFP). The neuromodulin fragment contains a signal for post-translational palmitoylation that targets EYFP to the plasma membrane. An EcoRI-XbaI fragment encompassing the EYFP-Mem reporter was cloned into the polylinker region of the pPha-T1 plasmid (Zaslavskaia et al., 2000). P. tricornutum cells were transformed with this new vector (pPhaT1-EYFP-Mem) by microprojectile bombardment (Zaslavskaia et al., 2000), and transgenic diatoms were selected on Phleomycin (Cayla) and screened for EYFP expression.

Labeling of silica valves and image acquisition

The Lysosensor DND-160 (Invitrogen Life Technologies Corporation, Carlsbad, CA, USA) was used (1 µm) to stain the newly silicified valves (for more details, see Desclés et al., 2008). Briefly, fluorescent microscopy was carried out with a Leica DM-IRB microscope coupled to a Z-stage piezo-controller. The filter set used is for m-EYFP excitation at 500/20 nm and emission at 535/30 nm, and for DND-160 excitation at 360/40 nm and emission at 535/30 nm or 445/40 nm. For the ratiometric determination, cells were first preloaded with DND-160 and silicic acid and then treated with concanamycin A. For each cell (n = 48), the fluorescence intensity of the two vacuoles was determined at both 535 and 445 nm. The decrease in the intensity at I535 and of the intensity ratio I535/I445 was verified using a z-test or t-test (P < 0.001). We used Fisher's test to estimate from fluorescence image analyses the correlation between the length of the raphe and the length of the cell.

Valve purification and TEM images

Exponentially growing cultures in artificial seawater medium without silica were incubated for 1.5, 3, 16 or 48 h with sodium metasilicate with or without concanamycin A. Concanamycin A concentration of 0.2 µm had no effect on cell growth but higher concentrations led to a decrease in growth. At the different times, the cells were first fixed with formaldehyde and the organic material was oxidized by 3% potassium permanganate with an excess of H2SO4 at 92°C for 10 min, and then treated with 16% HNO3 (v/v) and 48% H2SO4 (2 : 1, v/v) for 1 min. The suspension was neutralized by adding Tris-HCl buffer (1 m, pH 8), then carefully filtrated and washed with ethanol. One drop of the cleaned material was placed on a 300-mesh carbon-coated copper grid (Electron Microscopy Sciences, Hatfield, PA, USA) and observed the same day or within 2 d with a Philips Tecnai 12 electron microscope.

Morphometric and statistical analyses

Morphometry of valves was assessed by manual measurements from TEM images of the valve. We first made measurements without any selection of the valves’ traits for 107 and 206 samples for treated (0.2 µm concanamycin A) or untreated cells, respectively. Statistical analyses of the data distribution were made using these, excluding c. 10% of the valves that were clearly outside the distribution and that corresponded to incomplete or to aberrant forms. The final number of samples were n = 95 (0.2 µm concanamycin A) and n = 174 (untreated) and corresponded to five and four independent experiments, respectively. The frequency distribution obtained for the different traits, except for the R asymmetry index, calculated by dividing for each valve the longest by the shortest raphe, could be fitted to a normal distribution (Fig. 4b). The distance separating two adjacent transapical-costae (λmC) was measured from independent valve preparations of the same preparation and from independent experiments. λmCC was determined from 25 valves prepared from untreated or concanamycin A-treated cells with a total of n = 370 or n = 341 measurements, respectively. For each valve, we measured from 10 to 20 transapical-costae (or 10 to 15 transapical-costae for concanamycin A-treated cells) along the entire raphe (from the middle region and up to the edge of the raphe region). The size of the cross-costae (λmCC) was determined from 25 valves prepared from untreated or concanamycin A-treated cells with n = 284 or n = 240 measurements, respectively. Small variations were observed depending on whether the measurements were performed close to the raphe or at the margin of the transapical-costae, with the pores from the latest region tending to be slightly larger. However, such a tendency in the deposition process should need further analyses to be confirmed. For each trait, the means were compared by two-sided t-test (P < 0.001), and confirmed using nonparametric Wilcoxon–Mann–Whitney test. To estimate the allometric relationships, we used simple linear representation, as already proposed by D’Arcy Thompson (Thompson, 1942), but we also found that other conventional allometric equations, such as Huxley's power function or variations of this type (Gould, 1966), did not significantly improve the fit of the data, and will not change our interpretations.

Figure 4.

Morphometric variability of the valve pattern. (a) To quantify the valve metrics, six traits were measured from TEM images: the total length of raphe (R) was calculated by adding the size of the left (lR) and the right (rR) raphe, the width at the central nodule (WCN), the overall surface area (S), the spacing between the transapical-costae (λmC) and the distance between cross-costae (λmCC). (b) Frequency distribution of the variation in six traits measured in two different culture conditions, untreated (pink) and 0.2 µm concanamycin A (blue), for n = 174 or n = 95 different valves, respectively. The distribution of λmC corresponds to n = 370 or n = 341, and that of λmCC corresponds to n = 284 or n = 240, from untreated or concanamycin A-treated cells, respectively. The probability density functions of the normal distributions are represented.

Results and Discussion

Valve morphogenesis

Phaeodactylum tricornutum is unusual because it can form three major morphotypes and because silicification is restricted to one valve of oval cells (Fig. 1), and to the girdle bands of all other morphotypes (Reimann & Volcani, 1966; Borowitzka & Volcani, 1978). However, even if the structure of the silica shell has been studied in the past and some information on the valve formation in the oval cells has been reported (Borowitzka & Volcani, 1978), information on very early stages of valve formation and on the variability of the fine structures of the complete valves were missing. To this end we first screened for growth conditions and P. tricornutum strains to have cultures that contain exclusively (over c. 95%) oval cells. The selected strain (see the Materials and Methods section) was cultured in polycarbonate bottles in an enriched artificial seawater medium free of silicic acid (Si). Fluorescence analyses of the valve formation revealed that the addition of Si induces within the first few hours the formation of the valve in exponentially growing oval cells; at least for a large proportion of cells in the population (also see later discussion). Therefore, to reconstruct the different developmental stages and to build a coherent model for valve morphogenesis, numerous TEM images of isolated valves were obtained from cells harvested at different times (1, 1.5, 3 and 16 h) following Si-replenishment, and arranged in a probable time sequence (Fig. 2a–f).

Figure 1.

Electron micrograph of the Phaeodactylum tricornutum silica valve. The central nodule (cn) separates the two branches of the raphe (r). The higher magnification highlights the position of the raphe fissure (rf), and the pores (p) that are distributed along the transapical-costae (C) and formed by siliceous connections, called cross-costae (CC), between the transapical-costae.

Figure 2.

Developmental stages of the valve. Valve formation can be separated into three different phases: early, intermediate and late. The early phases, deduced from valves harvested at 1–1.5 h post-Si-replenishment, correspond to the formation and elongation of ribs, that is, the formation of the raphe sterna. (a) The primordial silica structure observed corresponds to a ‘π-like’ structure. (b) Elongation of one silica rib from the top of the π-like structure. (c) One primary longitudinal rib from the top of the π will elongate and then turns back sharply upon itself while the second raphe rib is initiated at the opposite side of the π. (d) Enclosure of the raphe slits (i.e. completion of the raphe fissure) occurs after the extending longitudinal rib, from the top of the π-like structure, meets the shorter rib that was initiated at its base. (e) During the intermediate phases, the raphe thickens while the previously initiated transapical-costae grow together and cross-costae appear. (f) The late phases correspond to valve pattern densification and thickening. The TEM images from (a) to (d) correspond to valves extracted 1 or 1.5 h after Si-replenishment, and in (e) and (f) to 3 and 16 h, respectively. Bars, 1 µm.

The earliest stage of a developing valve that we could unequivocally identify (from valves isolated at 1 or 1.5 h post-Si-replenishment) resembles a ‘π’-like structure (Figs 2a, 3a), a structure that corresponds to the future central nodule. Anterior and simpler structures must exist but they are difficult to identify and might be confounded by broken parts. From this π-like structure, the elongation of the primary raphe ribs appears always to begin from the top of the structure that would correspond to growth along one axis of the central nodule (Fig. 2a,b). The observed growth of the ribs strongly resembles a tip growth which is initiated in a particular order and occurs at the extremity of the growing sternum (Figs 2b, 3b). Even though we could not determine the exact timing for the growth of each side of the primary ribs, we often found, in valves extracted at 1, 1.5 or 3 h after Si-replenishment, that one side of the primary rib was more developed than the other one (Fig. 2b,c and Supporting Information Fig. S1). At a certain point the primary ribs will U-turn and join the secondary ribs which have grown following a certain delay on the other side of the future raphe (Fig. 2b,c and Fig. S1).

Figure 3.

Details of silica structures found in developing or mature valves. (a) Higher magnification showing several germ-like structures (i.e. the future transapical-costae) on the top of the π-like structure. (b) Tip-growth-like mechanism observed for a silica rib. (c) Germ-like structures (arrows) and U-turn of a primary rib. (d) The Voigt discontinuity (arrow) corresponds to the position where opposing strains of the silica deposition meet and fuse. This step precedes the enclosure of the raphe. (e) Denser pattern of the raphe edge or pole. (f) Fractal-like structure observed at the edges of the transapical-costae essentially in the mature valve. The TEM images from (a) to (d) correspond to valves 1 or 1.5 h after Si-replenishment, and in (e) and (f) to 16 h. Bars, 200 nm (a, b, e, f); 500 nm (c, d).

Analyses of numerous TEM images also allowed us to propose that one side makes the U-turn before the other, and that the elongation of the secondary ribs probably starts just before the primary rib has U-turned. We estimated that the primary and secondary ribs will face each other at about two-thirds of their length (Figs 2d, 3c). This position, where the silica ribs meet and fuse, has been observed for another naviculoid raphe sternum, called the Voigt discontinuity (Mann, 1984). A similar position for the Voigt discontinuity can also be estimated from previously reported pictures of early stages of raphe formation in P. tricornutum (Borowitzka & Volcani, 1978). Such a precise mechanism, which argues for a highly controlled topology of the growing SDV, ensures that the raphe is basically complete at an intermediate stage of development and can subsequently control further deposition of material (Fig. 2e–f), that is, elongation of the transapical-costae (lateral outgrowths of silica from the sternum, also named virgae; Cox & Ross, 1981) and formation of the cross-costae (also named vimines; Cox & Ross, 1981). Such a scenario and some of the silica structures observed here are slightly different from the ones proposed for other pennate diatoms. For example, in N. pelliculosa, the arms of the secondary ribs are much more separated from each other and the central nodule is more developed (Chiappino & Volcani, 1977). However, it was reported for this latest species that the primary ribs will U-turn before the secondary ribs start to grow. Other hypotheses have also been proposed to explain the bilateral asymmetry in the development of the raphe sternum (Mann, 1984). The existence of different scenarios suggests that the exact ontogeny of the raphe sternum formation could vary, at least to some extent, in different pennate species (also see Cox, 1999).

We found that lateral initiation of regularly spaced fingers that correspond to the transapical-costae is also determined very early during silica deposition and before the raphe structure is completed (Figs 2c, 3b). It is noticeable that these transapical-costae grow homogeneously toward the edge of the valve, and are very regularly spaced (Fig. 3d). This is inconsistent with a growth pattern that would occur by a free homogeneous diffusion limited aggregation (DLA) growth, because DLA processes are characterized by growth of branches that interact with each other. Indeed, branches can start with a regular period (most unstable wavelength), but then the strong nonlinear interaction makes branches compete, resulting in irregular fractal patterns that are commonly obtained from DLA processes. By contrast, transapical-costae development resembles a spatially controlled growth, that is, a directional growth, where the growth, and thus the interaction, is spatially limited within an expanding front. Only in this case can the fingers grow parallel and maintain a regular spacing. Such a process could be achieved most simply if the growth not only occurs within the SDV, but preferentially at its margin, and that the SDV, small at the start, only slowly enlarges, concomitantly with the valve. In directional growth deposition, the speed of growth could also be the main parameter for the development of regularly spaced transapical-costae. Under a certain threshold, only homogeneous growth would occur. Such a process could also explain the fractal-like structures that are often observed at the edges of the transapical-costae (Fig. 3e), reminiscent of a local DLA growth, and the smaller growth of the discs at the raphe margin (also named raphe pole), below the instability threshold, resulting in the formation of plain and smaller discs (Fig. 3f). These latter structures could result from a slower growth of the deposition front and/or an increase in the deposition rate of material.

Finally, the striking fact in the formation of cross-costae is that it does so by enclosing the spaces between the transapical-costae, by joining one costae to the other, while a characteristic of DLA and directional growth is to create fingers that never reconnect. This indicates that there is still more silica to deposit behind the growing front, in contrast to DLA and directional growth, where the whole material originates from outside the growing front (Fig. 3e,f). That there remains silica to deposit after the first growth of the transapical-costae suggests that there more material incoming, and the resulting effect recalls a phase separation process, occurring between the already created first transapical-costae. Indeed, phase separation can direct the deposition of condensing silica to form complex arrays of pores (Lenoci & Camp, 2008), such as the one observed here.

Morphometric noise and allometric variations

To describe more precisely the pattern and its variability, we developed quantitative approaches by first measuring lengths and distances (metric descriptions) of the P. tricornutum valve. Six different traits were considered: the length of the two raphe slits (lR and rR), the total length of the raphe (R) (R=lR+rR), the width at the central nodule (WCN), the overall surface area of the valve (S), the spacing between different transapical-costae (the wavelength λmC) and the spacing between the cross-costae (the wavelength λmCC) (Fig. 4a).

To understand the natural variability of the pattern within a clonal population grown in homogeneous conditions, we measured the different traits. We found, from 174 valves from five independent experiments (see the Materials and Methods section), the values R = 5.38 ± 0.44 µm, WCN = 1.61 ± 0.17 µm, and S = 7.94 ± 0.91 µm2 (Fig. 4b). We then evaluated the morphometric noise strength (variance/mean), which is a measure of the penetrance of each trait in the population. For these three main valve traits, we found that the variation is, in fact, very small (< 0.1) (Table 1). Another interesting criterion is the slight ‘left-right’ valve asymmetry, that is, one raphe slit often being slightly shorter than the other one (Fig. 4b). However, no relationship was obtained between this asymmetry and the total raphe length, suggesting that this fluctuating asymmetry results from random deviation of the process or from asymmetry in the cellular organization. We then found that the transapical-costae wavelength (λmC) was 116 ± 3 nm (n = 370), showing that the control of their formation is very precise (Fig. 4b, Table 1). Finally, a less controlled parameter was the pore width or spacing (λmCC), measured to be 89 ± 10 nm (n = 284), giving a morphometric noise of c. 1.2 (Fig. 4b, Table 1). Such a significant noise is probably the result of relatively weak control of pore formation, and variation of the pore size along the costae (i.e. with wider pores often found at the base of the cross-costae and more heterogeneous and smaller pores at the edge).

Table 1. Strength of morphometric noise, defined as the quantity variance divided by the mean (σp2/<p>) of valve traits
  R (µm)WCN (µm) S (µm2)λmC (nm)λmCC (nm)
  1. R, total length of the raphe; WCN, width at the central nodule; S, surface area of the valve; λmC, spacing between different transapical-costae; λmCC, spacing between the cross-costae.

0.2 µm concanamycin A0.170.070.340.081.34

We then monitored by fluorescent microcopy the formation of the silica shell, by adding to exponentially growing Si-starved P. tricornutum cultures, both silicic acid and the LysoSensor DND-160 (see the Materials and Methods section). This latest fluorescent dye and others have been used to label newly synthesized silica frustules (Shimizu et al., 2001; Leblanc & Hutchins, 2005; Desclés et al., 2008). In P. tricornutum the newly synthesized raphe structure can be seen as a rod-shaped valve (Fig. 5a, untreated). Because the length of the valve can vary with cell size, we established a new system that would allow us to measure cell length in parallel. We transformed P. tricornutum with a plasmid expressing an EYFP-Mem protein fusion, which localized to the plasma membrane (see Fig. 5a and the Materials and Methods section). Such localization of the EYFP reporter at the cytoplasmic membrane argues that posttranslational palmitoylation occurs in diatoms. Following Si-replenishment, we found that after 16 h (c. one generation time) typically c. 30% (29.4% ± 2.1, n = 919) of the cells had made a valve (Fig. 5b, [concanamycin]= 0 µm). Such a low number of labeled valves could be explained if valve formation is a stochastic and optional process, so not all the cells commit themselves to the cell wall silicification process. Cells could also lose their valve, as proposed by Borowitzka & Volcani (1978), therefore decreasing the apparent number of fluorescent valves in the population. We measured the average size of the raphe to be 5.96 µm ± 0.45 (n = 102), as measured from end to end in fluorescent images (Fig. 5a, red channel), and that of the cells to be 7.76 µm ± 0.36 (n = 102), measured using the maximum length of the cell (Fig. 5a, green channel). We calculated that the average valve size corresponds to 76.8 ± 4.8% of the length of the cell (Fig. 5c, [concanamycin]= 0 µm). These results revealed that fluorescence image analysis gives a good estimate of the length of the raphe (compared with the result obtained by TEM image analysis; Fig. 4b), and that the length of the raphe correlates with the length of the cell (also see the Materials and Methods section).

Figure 5.

Quantitative inhibition and qualitative alteration of valve formation by a V-ATPases inhibitor. (a) To quantify the valve formation by fluorescent microcopy, a transgenic cell line expressing a EYFP reporter localized at the cytoplasmic membrane was cultured in the presence of LysoSensor DND-160 (red). The inserts correspond to bright field images; bars, 5 µm. (b) Silica-starved cells were incubated for 16 h with both silicic acid and DND-160, and the proportion of newly silicified valves was assessed. The total number of cells counted in four independent experiments correspond to n = 919, n = 613, n = 816, n = 1453, n = 1472 and n > 1500 at 0, 0.1, 0.2, 0.4, 0.6, 0.8 µm concanamycin A, respectively. No silicified valves were observed above 0.8 µm. (c) Effect of concanamycin A on the cell length (green) and valve length (red). The data correspond to n = 102, n = 104, n = 88, and n = 32 cells grown without inhibitor, or in the presence of 0.1, 0.2 and 0.4 µm concanamycin A, respectively. (d) TEM images of organic-free valves. Note the shorter, thinner and bent valve upon concanamycin A treatment.

Valve pattern formation in the presence of V-ATPase inhibitors

To test for the stability of pattern formation against internal perturbations, we thought that a perturbation of the activity of proton transporters could lead to a modification of the intracellular pH (at least partially), and result in alteration of the polycondensation process. Among the different transporters involved in the control of organelle and cytosolic pH, it is known that the vacuolar (H+)-ATPases play important roles (Schumacher, 2006; Forgac, 2007). Moreover, several pharmacological inhibitors of the V-ATPase exist (Huss et al., 2002; Bowman & Bowman, 2005; Wang et al., 2005), allowing us to address the role of the multisubunit complex on valve formation. We therefore tested the effect of two high-affinity macrolide antibiotics on valve formation.

In the presence of bafilomycin at concentrations varying from 0.1 to 1 µm, no valve could be observed (not shown). We then measured for concanamycin A the dose–response relationship and obtained Ki = 0.14 µm for the inhibition of valve formation (Fig. 5b), inferring the importance of V-ATPases in the biomineralization process. For P. tricornutum, in order to ascertain the inhibitory effect of concanamycin A, we also measured the modification of the ratio of valve size to cell size and found a drastic decrease on the average valve size in the presence of concanamycin A (Fig. 5c). We also found that concanamycin A affects valve morphology, not only by reducing the average valve length, but also by creating a bend (compare, in Fig. 5a and d, the untreated and +0.2 µm samples). We verified that concanamycin A also induces a modification of the intravacuolar pH by exploiting the pH-dependent dual emission of DND-160 (Diwu et al., 1999; Lin et al., 2003). DND-160 is a fluorescent dye that selectively labels acidic organelles (such as vacuoles and lysosomes) of living cells, and the two distinct emission peaks can be used to monitor the pH fluctuations of live cells in ratio measurements. Ratiometric imaging of P. tricornutum cells showed that in the presence of 0.2 µm concanamycin A, the fluorescence emission intensity of the vacuole at I535 decreases by a factor of 1.5 ± 0.1 compared with cells grown without inhibitor, and that the intensity ratio I535/I445 decreases by a factor of 2.4 ± 0.5 (not shown). This result demonstrates that inhibition of the V-ATPase activity leads at least to a rise in vacuolar pH. Such alkalinization of intracellular compartments in the presence of V-ATPase inhibitors has also been reported in several kinds of organisms (Kluge et al., 2003; Forgac, 2007; Marshansky & Futai, 2008).

Finally, we tested the consequences of inhibition of V-ATPase activity on the frustule formation process with another pennate diatom, Cylindrotheca fusiformis, and with two centric diatoms, Thalassiosira weissflogii and T. pseudonana. For these three species, we found that concanamycin A also inhibited frustule formation (including both valve and girdle bands), although for the centric diatoms, the concentrations required for inhibition were 10 times lower than that needed for pennates (not shown). Altogether our results demonstrate that in diatoms the inhibition of V-ATPases affects the frustule formation process both qualitatively (i.e. change in geometry) and quantitatively.

Robustness of the valve formation process

To test the consequences of V-ATPase inhibitor on valve formation, we chose 0.2 µm concanamycin A because at this concentration no effect was observed on cell growth (not shown). At higher concentrations, a decrease in growth rates was visible and at 0.8 µm the cell stopped growing. Moreover, at concanamycin A concentrations > 0.4 µm, the number of valves being made was too low for statistical analyses, and they often displayed abnormal morphologies (see later).

Statistical analyses of TEM images from 95 valves purified from cells grown in the presence of 0.2 µm concanamycin A revealed that the mean value of R decreases by 65% compared with untreated cells (Fig. 4b). We also found that concanamycin A increased raphe asymmetry (Fig. 4b). Surprisingly, we found a much more profound effect on WCN, which decreased by a factor of c. 2.1 (1.61 ± 0.17 vs 0.76 ± 0.23 µm), and an even more drastic effect on S, which decreased by a factor of c. 2.7 (7.94 ± 0.91 vs 2.92 ± 0.99 µm) (Fig. 4b). These results reveal that the formation of the raphe and the formation of the transapical-costae and cross-costae might not necessarily be linked, at least regarding the extent of their development.

To further substantiate the modifications in valve morphology, we estimated the allometric relationships between the raphe length (R) or the costae length at the central nodule (WCN) and the surface area (S). To calculate the correlation between R and S, we used a simple linear representation (see the Materials and Methods section), and found that both the slope and the intercept were similar for untreated or concanamycin A-treated cells (Fig. 6a). The modification of the overall valve geometry was clearly seen by analyzing the relationship between WCN and S, where the slope and the intercepts were affected by concanamycin A treatment (Fig. 6b). Compared with untreated cells, the noise strength of R, WCN and S was increased by factors of 4.7, 4.0 and 3.3, respectively (Table 1). Perturbation of the activity of V-ATPases thus creates a high-noise state, demonstrating that intra-organelle and/or intracellular pH are important parameters in silica biomineralization. Altogether our data demonstrate that V-ATPase inhibitors lead to a qualitative modification of the size and the geometry of the valve by mostly affecting the extent of transapical-costae development. However, contrary to the other traits measured, concanamycin A did not significantly affect λmC and λmCC. Indeed, we found that the distances between transapical-costae or cross-costae were unchanged, even when looking at the nanometric scale (Fig. 4b), suggesting that the parameters that control their formation are less sensitive to inhibitors of V-ATPases.

Figure 6.

Allometric relationship of valve traits. (a) Linear representation of the relation between the raphe length (R) and the surface area (S). (b) Linear relation between the width at the central nodule (WCN; i.e. the essentially transapical-costae development) and the surface area (S). The pink diamonds (n = 174) correspond to valves from untreated cells, and the blue circles (n = 95) to valves extracted from cells grown with 0.2 µm concanamycin A. The red or dark blue line corresponds to the fitted linear equations.

Finally, extensive analysis of TEM images from concanamycin A-treated cells revealed aberrant valves that can be useful to understand pattern formation (Fig. 7). To explain the pattern of aberrant valves, one could hypothesize that the effect of the drug is not to affect the general patterning, but other features such as flux intensities (of Si and/or H+?). Wider transapical-costae and an overall decrease in the development of the costae networks indicate more detachment-redeposition events (Fig. 7a–c). We also observed that concanamycin A can impair raphe sternum development. Remarkably, in these cases, where raphe enclosure did not happen at all, transapical-costae could nonetheless develop, even though the cross-costae formation was profoundly affected (Fig. 7a,b,d). In a few other cases, particularly for concanamycin A concentrations higher than 0.2 µm, the whole geometry of the valves was affected (Fig. 7e,f). It therefore seems that when the π-like structure is not completed, the whole flux of material can be redistributed, leading to local irregular growth of more DLA-like fingers, and the formation of an unrecognizable shape. Our data suggest that, at low concentrations (0.2 µm), concanamycin A might only perturb the deposition process (Si transport, intra-SDV pH, transport of protein/LCPA??), leading to a decrease in the development on the valve, with the development of the transapical-costae and cross-costae being the most affected. However, at higher concentrations of concanamycin A, the formation of the raphe will also be affected, leading to modifications of both the overall geometry of the valve and the deposition process.

Figure 7.

Aberrant valves from concanamycin A-treated cells. (a) Development of transapical-costae on incomplete raphe sternum. Note the larger and more rounded shape of the transapical-costae. (b) Development of transapical-costae before complete raphe enclosure. In these aberrant kinds of valve, instead of growing together in the same direction, the transapical-costae invade the free space like diffusion limited aggregation fingers. (c) Formation of wider transapical-costae and very little development of cross-costae. (d) Detail showing that costae can develop perpendicular to the raphe sternum but in the two directions. (e) Slight modification of the valve geometry. (f) Complete deformation of the π-like structure. The valves were purified from cultures treated with 0.2 µm concanamycin A in (a), (b), (d) and (f), or 0.4 µm (c) and (e). Bars: 200 nm (a, f); 500 nm (b–e).

Altogether, our analyses indicate that the overall biomineralization process is robust, at least considering any molecular fluctuation involved in this process and considering the general patterning parameters. However, the formation of some structures seemed to be less controlled than others, suggesting that slightly different deposition processes might occur, or that the reacting species could be different for the different parts of the valve.

A model for valve morphogenesis

Our observations of the pattern formation of the valve of P. tricornutum and of its stability suggest that previous models cannot completely or accurately explain the formation of the whole valve pattern and that different, new models or combinations of models are required. We therefore propose here a new minimal model for pattern formation of the P. tricornutum cell wall, and possibly other raphid pennates.

We postulate that the SDV is initially small and only grows with the valve itself. Moreover, in this primordial SDV, there is an as-yet-unknown structure in the place of the future slit region that elongates on both sides of the π-like structure (Fig. 8a). This ‘central fixing structure’ is probably already present at the original growth of the two legs of the π, and could play a key role in pinning both SDV sides to the cytoplasmic membrane, attaching it to the diatom cell surface plane. The quick tip growth of the raphe ribs, as well as the further thickening of the whole structure suggest that a ‘central fixing’ structure could not only help to fix the SDV to the cell membrane but also favor the docking of vesicles. Such vesicles, which could eventually correspond to silicon-transport vesicles (STVs) (Schmid & Schulz, 1979), will provide simultaneously the necessary deposition ingredients (proteins/LCPA and/or silica/silicic acid?) as well as membrane to extend the SDV membrane (silicalemma). This fixing structure could also control the influx, by being the best site for vesicle targeting and fusion. Vesicles at the center of the forming raphe have already been observed by TEM images from several kinds of diatom (Pickett-Heaps et al., 1990). Such convergence of vesicles at this ‘organizing center’ would also explain the transapical relief of the valve (along the distal to proximal differentiation), rapidly decreasing from the raphe to the margin of the transapical-costae, as well as the slower and full growth of the lateral discs, where the same amounts of material all along the raphe spread in a much larger space at the poles of the raphe (Figs 2f, 3f).

Figure 8.

Proposed model for valve formation. (a) During the early phases from a π-like structure, silica deposition occurs inside the silica deposition vesicle (SDV) by a tip growth mechanism along a ‘central fixing’ structure (dotted). (b) In the intermediate and late phases, transapical-costae develop by a directional growth mechanism. During valve pattern densification and thickening, a phase separation process leads to the formation of the cross-costae that will define the future pores (areolae). The arrows correspond to the putative gradient fields. AA’ and BB’ correspond to representations of our model in cross-sections.

What might be the exact nature of such a ‘central fixing’ structure? Previous studies on valve morphogenesis have revealed the existence of a structure present at the cytoplasmic side (inner) of the raphe, named the ‘raphe fiber’. It was proposed that this raphe fiber is implicated in shaping the curvature of the raphe (Pickett-Heaps et al., 1979b), or in inhibiting locally the deposition of silica (Edgar & Pickett-Heaps, 1984). Microtubules (MTs) that extend from the microtubule center (MC) have also been found associated close to the raphe fissure and to the raphe fiber complex (Pickett-Heaps et al., 1990). An ‘organizing center’ (OC) composed of a diffuse collection of dense particles plus some vesicles and which plays a role in MT growth and organization was proposed to contribute to valve morphogenesis in a centric diatom (van de Meene & Pickett-Heaps, 2002). Here, we speculate that the ‘central fixing’ structure plays a dual role in both preventing silica deposition (at least during the early stages) and directing the deposition process. Even if such a structure was not observed in previous studies of P. tricornutum (Borowitzka & Volcani, 1978), its existence is compatible with the observed development of the valve pattern and with previous studies on the role of cytoplasmic structures.

To complete our model, one simply needs to assume that silica (in the form of silicic acid?), could also be imported inside the SDV through the whole silicalemma in contact with the cytoplasmic membrane. Hence, to explain the observed directional growth of the transapical-costae, as opposed to growth by free homogeneous DLA, a simple assumption is that the flux of silica comes preferentially from the margin of the SDV (Fig. 8b). Finally, the growth of the cross-costae could indicate a change throughout the whole SDV deposition process allowing deposition between the transapical-costae (Fig. 8b). For instance, a large influx (of silicic acid?) through the cytoplasmic membrane toward the outer side of the silicalemma could increase the overall concentration enough to generate silica precipitation by a phase separation process, therefore leading to the formation of the pores (Fig. 8b).


Our model for valve formation in P. tricornutum corresponds to an extension of the reaction-diffusion process that will operate in a growing volume that is nonhomogenous and spatially controlled. Our analyses suggest that, from a π-like structure, valve formation occurs through the dynamic growth of the SDV itself and by continuous influx of materials through a particular spatial geometry. In the regulation of the spatial geometry, we propose an essential role for an attachment structure, as yet unknown. Such a structure would secure both the position and the extension of the raphe, and guide the geometry of the fluxes by helping the proper docking of new materials preferentially along the raphe. Our interpretation of the development of the silica pattern in P. tricornutum strengthens the idea that the robust control of the spatial geometry of the deposition process is likely to be a key factor in the establishment of the pattern. Our data also argue for a tight regulation of the dynamic of the SDV expansion. Such interpretations are in agreement with a previously proposed concept of a pattern center (Pickett-Heaps et al., 1990), or generation of the pattern as the valve grows or the SDV expands (Lacalli, 1981). Moreover, we also suggest that even for the relatively simple and lightly silicified valve of P. tricornutum, pattern formation might arise from sequential deposition events.

Our experiments also provide the first evidence that the inhibition of V-ATPases affects pattern formation in diatoms, therefore extending the role of V-ATPases to the silicon biomineralization process. V-ATPases are ubiquitous, ATP-dependent proton pumps that both acidify intracellular compartments and pump protons across the plasma membrane. Such proteins are also known to be essential for vesicular trafficking along both the exocytotic and endocytotic pathways of eukaryotic cells (Forgac, 2007; Marshansky & Futai, 2008). Further investigations will be needed to specify whether V-ATPases are directly or indirectly involved in silica deposition in diatoms. For example, it will be of particular interest to test if V-ATPases could directly contribute to the regulation of pH homeostasis of the silica deposition vesicle, and to address the exact role of the vacuole in silica biomineralization. Alternatively, it will be important to address the role of V-ATPases in transport and/or targeting of material to the SDV.

Finally, we believe that combining developmental and physiological analyses with pattern modeling in a model diatom species provides a quantitative and testable framework to better estimate the interplay between intra- and extracellular conditions, the valve shape and plasticity.


Research in PJL's laboratory was supported in part by the Europe Network ‘Marine Genomics’, the STERP project ‘Diatomics’ (LSHG-CT-2004-512035) and the CNRS Program ‘Interface physics-chemistry-biology’.