Морфологическое разнообразие клеток животных приводит к значительным различиям в клон-специфичной экспрессии и контроле цитоскелетных регуляторов. Клетки в культуре широко используются для характеристики морфогенетических событий, напр., динамики и организации филаментозного актина и микротрубочек в прилипших и подвижных клетках. Немногие клеточные системы метазоа позволют использовать генетический анализ для идентификации набора генов, вносящих вклад в построение клеточной формы.

RNA interference (RNAi) революционизировала функциональный анализ генов, идентифицируемых с помощью геномного секвенирования [1-3]. Некоторые факторы делают клеточные культуры Drosophila очень удобными для такого RNAi функционального геномного анализа формы клеток животных. Доступность хорошо охарактеризованных геномных последовательностей Drosophila упрощает полученипе ген-специфических double-stranded RNAs (dsRNAs) [4]. Более того, геном Drosophila кодирует гомологи свыше 60% генов болезней человека [5] и лишен некоторого генетического перекрывания (redundancy), наблюдаемого у позвоночных. RNAi в клетках Drosophila эффективна, т.к. редуцирует или элиминирует экспрессию гена-мишени, позволяет выявлять фенотип, связанный с частичной или полной потерей функции, при простым добавлении dsRNA в культуральную среду [6]. Наконец, хорошо-разработанная генетическая техника для Drosophila производить сравнения между фенатипами культивируемых клеток при потере функции и фенотипами тканей, наблюдаемыми у соответствующих мутантрных мух.

Чтобы разработать базирующийся на клетках подход для изучения функций генов, участвующих в морфогенезе, мы разаработали высоко-производительную RNAi скрининг-методологию для клеточных культур Drosophila с разнообразным клеточным поведением (Figure 1a). Этот подход состоит из следующих ступеней: первое, оформление и синтез библиотеки ген-специфических dsRNA; второе, инкубация клеток Drosophila с dsRNAs с помощью 384-well assay plates (в бессывороточной среде или с трансфекционными реагентами в зависимости от линии клеток); и третье, optional индукция клеточного поведения, сопровождаемая определением люминисцентных или флюоресцентных сигналов, используемых в plate reader или automated microscope.

Здесь описывается отладка функционального RNAi подходя для изучения клеточной морфологии. Используя полученные с помощью автоматизированной микроскопии изображения, мы визуализировали фенотипические изменения, получаемые в результате reverse-functional анализа после обработки клеток Drosophila в культуреген-специфическими dsRNAs. Мы получили возможность наблюдать и охарактеризовать широкий круг фенотипов, с изменениями организации цитоскелета и формы клеток, и исходя из этого идентифицировали наборы генов, необходимые для определенно круглой или плоской клеточной морфологии.

Мы адаптировали RNAi технологию к культуре клеток Drosophila для использования высоко-производительного скрининга по определению генов, отвечающих за морфогенез клеток. Чтобы идентифицировать гены, ответственные за характерную форму двух морфологически отличающихся линий клеток, производили RNAi скрининг в каждой из линий с набором double-stranded RNAs (dsRNAs), нацеленных на 994 предполагаемых регуляторов клеточной формы. Используя автоматизированную флюоресцентную микроскопию для визуализации актиновых филамент, микротрубочек и ДНК, мы выявили морфологические фенотипы для 160 генов, одна треть из которых ранее не была охарактеризована in vivo. Гены со сходными фенотипами соответствовали известным компонентам путей, контролирующих организацию цитоскелета и форму клеток, это привело нас к предположению о сходных функциях у ранее неохарактеризованных генов. Более того, мы оказались способными выявлять гены, действующие внутри специфических путей, используя co-RNAi скрининг для идентификации dsRNA супрессоров изменения клеточной формы с помощью Pten dsRNA.

Classification of RNAi cell morphology phenotypes

We detected a broad spectrum of distinct defects in cytoskeletal organization and cellular morphology, including subtle effects in the localization and level of actin filaments and microtubules (see Table 2, Figure 3 and Additional data file 2). To classify the results, phenotypes were scored using defined descriptions assembled under one of seven major categories, denoting visible defects in actin filaments, microtubules, DNA, cell shape, cell size, cell number and cell viability (Table 2). We were able to further define subcategories that describe specific morphological attributes (see Materials and methods section for more details). Some descriptions were interdependent and therefore redundant; for example, cell shape was determined by a combined assessment of the actin and microtubule organization.

Using this system, a total of 417 phenotypic annotations were assigned to 160 genes, ranging from zero up to six annotations per gene in one cell type (Table 2, Figure 4). A comparison between the two RNAi screens revealed that 41% (65/160) of the genes were identified with phenotypes in both Kc₁₆₇ and S2R⁺ cell types. This overlapping set identified many genes that are known to control important cell-biological functions common to all cell types, such as cell-cycle progression and cytokinesis, and genes that may reflect a hemocyte origin (Figure 2 and see below). In comparing the two cell types, nearly twice as many of the genes were found to have a detectable RNAi phenotype in S2R⁺ cells (146/160 genes, or 91% of the total) as in Kc₁₆₇ cells (79/160; 49% of the total). Genes identified in S2R⁺ cells also had a greater mean number of phenotypic annotations assigned to them (2.0) than in Kc₁₆₇ cells (1.2; see Figure 4). This was due in part to the ease of detecting overt phenotypes in the larger S2R⁺ cells but may also indicate a difference in the number of genes required to maintain a flat versus a round cellular morphology (see below). Interestingly, the relative importance of a gene in the two cell types, as determined by RNAi, did not strictly correlate with the relative levels of expression. Furthermore, RNAi was shown to deplete the protein in cases in which there was no measurable phenotype in our assay (see below; and data not shown).

We also noted cases in which morphological defects were accompanied by a decrease in cell number. An RNAi-induced phenotype was accompanied by a notable decrease in cell number (estimated as fewer than half the normal number of cells per image) in 43% of cases (68/160 genes; see Additional data file 2). Less than 1% of the genes screened caused a catastrophic reduction in cell number (an estimated fewer than 100 cells per image) three days after the addition of dsRNA (6/994 genes, listed as having a cell viability defect in Additional data file 2). One example of this class of genes was a known inhibitor of apoptosis, D-IAP1 [13]. These data demonstrate that under these conditions, severe cytotoxicity is not a major obstacle for cell-based RNAi screens, even if many of the genes are essential for Drosophila development.

RNAi phenotypes with common cytoskeletal defects

Changes in actin organization and cell shape were the most commonly observed phenotypes (94 and 105 out of 401 phenotypes, respectively). In some instances, specific dsRNAs led to defects in F-actin with related morphological consequences in both Kc₁₆₇ and S2R⁺ cells (22 genes). For example, both cell types displayed RNAi phenotypes characterized by an elevated accumulation or a polarized (asymmetric or uneven) distribution of F-actin (13 genes). These phenotypes identified genes encoding proteins thought to limit the rate of actin-filament formation [14], such as twinstar (encoding cofilin) and capping protein beta, as well as previously uncharacterized Drosophila genes, such as Pak3 and CG13503 (Figure 3b,g). Conversely, dsRNAs targeting several known regulators of actin-filament formation compromised cortical F-actin in both cell types (9 genes). In addition, actin-rich protrusions were observed in both cell types following dsRNA targeting of CG5169 (Figures 3c,h), a Drosophila gene encoding a homolog of a Dictyostelium kinase thought to regulate severing of actin filaments [15]. Thus, one class of cytoskeletal regulators has similar functions in two morphologically distinct cell lines, irrespective of their characteristic shape. In addition, a significant proportion of the genes implicated in cell-cycle progression (65%) or cytokinesis (50%) exhibited similar RNAi phenotypes in both cell types.

RNAi phenotypes affecting distinct cell shapes

To identify genes that specify different cell shapes, we focused on morphological phenotypes that were restricted to either Kc₁₆₇ or S2R⁺ cells. Indeed, 78% of the morphological phenotypes observed were detected in only one of the two cell types. Kc₁₆₇ cells frequently adopted a unique, bipolar spindle shape in response to specific dsRNAs (21 genes), reminiscent of the cell-shape change induced by actin-destabilizing agents or ecdysone (Figure 1). This shape change was usually associated with the formation of discrete F-actin puncta and opposing microtubule-rich processes and was seen in cells treated with dsRNAs targeting genes known to promote actin-filament formation (such as those encoding Cdc42 and SCAR) [14] and others known to affect microtubules (for example, par-1) [16]. These observations suggest that actin filaments and microtubules play antagonistic roles in Kc₁₆₇ cells, with the contractile actin cortex opposing the formation of microtubule-based processes. Although Kc₁₆₇ cells exhibited a marked tendency to take on a bipolar morphology, various gene-specific manifestations of this phenotype were distinguishable. For example, a single, microtubule-rich extension formed directly opposite from a single, large, actin-rich protrusion in Kc₁₆₇ cells treated with dsRNA targeting the gene for the Hsp83 chaperone (Figure 3d). In addition, a large and flat bipolar morphology was induced in Kc₁₆₇ cells treated with dsRNAs targeting the puckered gene encoding JNK phosphatase (Figure 3e), CG7497, encoding a predicted G-protein-coupled receptor kinase, and the Pten gene encoding phosphatidylinositol (3,4,5)-trisphosphate (PIP₃) 3-phosphatase (see below).

One major behavioral difference between the two cell types used in this study is the ability of S2R⁺ cells to adhere to and spread over the substratum. As a result, subtle changes in cytoskeletal organization could be visualized in S2R⁺ cells, such as polarized (uneven) F-actin accumulation (in response to dsRNA targeting Abl-encoded kinase), actin stress-fiber formation (the RhoL-encoded GTPase) and the loss of cortical actin filaments (dsRNA targeting CG31536, encoding a predicted Rho guanine-nucleotide exchange factor (GEF) with a FERM actin-binding domain; Figure 3i). Of particular interest were genes required for the spreading process characteristic of S2R⁺ cells. S2R⁺ cells rounded up and detached from the plate in response to dsRNAs targeting 37 different genes, 20 (54%) of which had no visible effect on Kc₁₆₇ cells. Four genes identified in this way had known functions in cell-matrix adhesion [17] (see Figure 5c), including an enigmatic adhesion molecule that contains an integrin-ligand RGD sequence (Tenascin-major) [18], both ? and ? integrin subunits (inflated and myospheroid) and a focal-adhesion cytoskeletal anchor (talin) [19], as well as focal adhesion kinase (FAK56D, with a slightly different defect in cell spreading). This set also included novel genes (CG4629, encoding a predicted kinase; Figure 3j). The remaining 17 genes that, by RNAi, affected both S2R⁺ cell spreading and Kc₁₆₇ cell morphology may identify those that indirectly affect the cell-adhesion process (for example, S2R⁺ cells rounded up as a consequence of RNAi-induced arrest in mitosis; Figures 2 and 6).

The set of genes identified by RNAi defects in cell spreading suggested that S2R⁺ cells utilize focal adhesion complexes to flatten on the substrate. An implication of this finding is that Kc₁₆₇ cells may be unable to spread on the substrate because they fail to express adhesion-complex components. Surprisingly, quantitative PCR (qPCR) of reverse-transcribed mRNA revealed a 2.4-fold enrichment of ?PS integrin (mys) expression in Kc₁₆₇ cells versus S2R⁺ cells (adjusted cross-point difference of 1.2 cycles; see Materials and methods section; data not shown). Furthermore, ?PS integrin/Mys protein was detected in both cell types, with slightly elevated levels in untreated Kc₁₆₇ cells versus S2R⁺ cells, and similarly depleted in both upon treatment with mys dsRNA (Figure 7). We extended the analysis to other adhesion-complex components identified in the screen and discovered by qPCR that both ?-integrin (if) and Rap1 (R) were also expressed in Kc₁₆₇ cells, although at slightly lower levels than in S2R⁺ cells (adjusted cross-point differences of 1.0 and 0.3 cycles, respectively). In contrast, S2R⁺ cells exhibited a nearly 4.6-fold enrichment of talin expression relative to that in Kc₁₆₇ cells (adjusted cross-point difference of 2.3 cycles). Moreover, Mys levels were sensitive to the loss of Rap1 by RNAi in S2R⁺ cells (Figure 7). This analysis demonstrated that although many of the same adhesion complex components are expressed in both the round Kc₁₆₇ and spread S2R⁺ cells, the genes function differently in the two cell types, so that integrin-mediated adhesion has little impact on the morphology of Kc₁₆₇ cells.

Furthermore, Kc₁₆₇ cells adhered but remained round even when plated on an adhesive concanavalin A substrate that induced round S2 cells to flatten [20] (data not shown), although Kc₁₆₇ cells do flatten when actin-filament formation is compromised (Figure 1). Thus, spreading of Drosophila cells probably requires both integrin-mediated adhesion and reorganization of cortical F-actin. This is supported by the fact that S2R⁺ cells rounded up when treated with cofilin dsRNA because of an accumulated excess of cortical actin filaments. Integrins may, therefore, function to mediate substrate adhesion in both cell types, while the levels of additional gene products (such as talin, cofilin and phosphoinositide (PI) 3-kinase activity) determine whether or not the cell will spread.

Genes with common phenotypes share morphogenetic functions

The results from RNAi screens in both cell types were combined to generate a phenotypic profile for each gene. Genes with similar phenotypic profiles were involved in common morphogenetic functions, as indicated by several distinct sets of genes known to interact in pathways or complexes. In both cell types, dsRNAs specific for the pebble gene encoding a Rho-GEF, the Rho1 gene encoding a GTPase, and the CG10522 gene encoding citron kinase led to enlarged cells with multiple nuclei, indicative of a failure to form and constrict the actin contractile ring necessary for cytokinesis (Figure 5a). While Rho1 and pebble (and five other identified genes; see Figure 6) have already been shown to function in Drosophila cytokinesis [3,10], we identified CG10522 in the RNAi screen as a potential novel Rho1-effector required for cytokinesis [21]. RNAi targeting of members of a different group of genes resulted in a profound loss of actin filaments in both cell types, identifying known regulators of F-actin formation. In Kc₁₆₇ cells, dsRNAs targeting the Cdc42-encoded GTPase, enabled-encoded actin-binding protein, and SCAR-encoded regulator of Arp2/3 complex [14], each led to a reduction in F-actin, the appearance of microtubule-rich protrusions and cell flattening (Figure 5b). In S2R⁺ cells, RNAi of Cdc42, enabled or SCAR similarly reduced the levels of F-actin, compromising the ability to form lamellipodia (as in Figure 3i, and data not shown). Ena protein was effectively depleted upon ena RNAi in both cell types (Figure 7).

The screen profiles also identified clusters of genes with phenotypes unique to a single cell type, such as the set of matrix-adhesion genes required for S2R⁺ cell spreading, as noted above (Figure 5c). Three dsRNAs caused S2R⁺ cells to assume a unique, amorphous shape. This striking phenotype identified Ras85D, Downstream of Raf1 (encoding mitogen-activated protein (MAP) kinase kinase, or MEK) and kinase suppressor of Ras, all interacting components of the well-characterized MAP kinase signaling pathway [22] (Figure 5d). Thus, on the basis of phenotype alone, groups of genes were identified that function in the same cellular process, complex or pathway. In classic Drosophila genetic screens a similar logic was used to group genes on the basis of common mutant cuticle phenotypes, identifying genes that act together to control different aspects of embryonic development [23].

A co-RNAi screen identifies modifiers of the Pten-dsRNA-induced cell shape phenotype

Part of the success of using Drosophila as a model genetic system has relied upon modifier screens to identify novel components acting in related processes or molecular pathways of interest [24]. Using an analogous approach in cell culture, we designed an RNAi screen to identify genes that modify a specific RNAi-induced cell-shape change. Pten, a human tumor suppressor gene, is a lipid phosphatase that dephosphorylates PIP₃, acting in opposition to PI 3-kinase [25] to control many cellular processes including growth, adhesion, migration and apoptosis [26]. In the initial screen, Pten RNAi was found to polarize Kc₁₆₇ cells, inducing microtubule extensions and a flattened, bipolar shape (Figure 8b). A lower concentration of Pten dsRNA caused a visible but less severe asymmetric microtubule phenotype (Figure 8c) that was used for a co-RNAi screen to identify Pten modifiers.

By screening for dsRNAs that modified the asymmetric microtubule distribution seen in response to Pten RNAi, 20 of the 229 dsRNAs targeting predicted kinases were identified as visible suppressors of this phenotype. These included dsRNAs corresponding to seven genes that were not identified in screens in untreated Kc₁₆₇ cells: Akt1, CG31187, LIM-kinase 1, MAP kinase activated protein-kinase 2, Pi3K92E, slipper and wee. Importantly, two of these encode known positive regulators of the pathway: Pi3K92E and Akt1 [6] (Figure 8d,e). One suppressor, CG31187, encodes a predicted diacylglycerol kinase that may act directly in the phosphoinositide cycle [27]. It is possible that other genes identified as RNAi suppressors may rescue the Pten-morphology phenotype indirectly by modifying actin-filament organization (LIMK1 [28]). These results demonstrate that modifier screens, like those used to identify new components of specific pathways in classical genetic systems, can now be carried out in cell culture using RNAi-screening technology.

Несмотря на ограниченное знание молекулярных механизмов, используемых для поддержания морфологии клеток Drosophila в культуре мы идентифицировали свыше 100 генов с видимым фенотипом потери функции, которые затрагивают специфические аспекты организации цитоскелета метазоа, хода клеточного цикла, цитокинеза и формы клеток. Т.к. и Kc₁₆₇ и S2R⁺ клетки, по-видимому,используют сходный набор генов, чтобы регулировать актиновые филаменты в клеточном кортексе и при цитокинезе, но клетки S2R⁺ распластываются по субстрату, используя интегринами обеспечиваемую адгезию, а клетки Kc₁₆₇ нуждаются собственно в контроле PI 3-kinase пути, чтобы сохранять свою округлую форму. Итак, некоторые гены обусловливают сходные RNAi фенотипы в обоих типах клеток, тогда как др. клеточно-специфические функции, частично отражающие разные механизмы, используемые для создания округлой или плоской морфологии клеток. Более того, функциональные последствия снижения экспрессии индивидуальных генов не коррелируют с их уровнями экспрессии в двух типах клеток. Скорее всего, что функция генов детерминируется сетью функциональных взаимодействий большого числа белков. Т.о., наш анализ делает возможным генетическое описание двух типов клеток, которое выявляет потенциальные механизмы, с помощью которых их контрастирующие клеточные формы м. б. созданы. Эта же технология м.б. легко применена к модифицированным клеточным линиям или условиям для разнооюразных исследований клеток по большой геномной шкале. Сравнение разных RNAi скринов м.б. неоценимым в выявлении сложности путей, в которых наборы генов м. функционально взаимодействовть, чтобы обеспечить разные поведения клеток.