A high-throughput template for optimizing Drosophila organ culture with response-surface methods

The growth and development of the Drosophila wing imaginal disc has served as a ‘flagship’ model organ for a diverse range of studies on morphogen activity, pattern formation, growth control, epithelial cell-cell interactions, pattern formation and compartment boundary formation (Affolter and Basler, 2007; Crick and Lawrence, 1975; De Celis, 2003; Eaton, 2003; Neto-Silva et al., 2009; Schwank et al., 2011). Wing disc development lasts ~96 hours, during which the disc grows from 50 to 50,000 cells (Fain and Stevens, 1982; Garcia-Bellido and Merriam, 1971). The wing disc is geometrically simple, consisting of two epithelial layers that become folded in a stereotypic pattern during development (Ashburner et al., 2005; Ursprung et al., 1972).

Progress in Drosophila genetics continues to permit ever-higher degrees of sophisticated analyses of many developmental processes that are difficult to study in other systems (Belacortu and Paricio, 2011; Bischof et al., 2007; Elliott and Brand, 2008; Pearson et al., 2009; Yagi et al., 2010). However, the majority of studies to date have relied on measuring properties from populations of fixed, immunostained tissues owing to the difficulty of immobilizing and imaging living larvae and the lack of in vitro methods that recapitulate wing disc growth and development.

Pioneering studies on the primary culture of wing discs and wing disc derived cells by Wyss, Milner, Robb and Miyake in the late 1960s until the 1980s described different medium compositions that allowed for the transient in vitro maintenance of wing discs fragments (Robb, 1969; Wyss, 1982) and the establishment of stable disc-derived cell lines (Currie et al., 1988; Ui et al., 1987). Recently, several studies have employed short-term culture and live imaging of wing discs (Aldaz et al., 2010; Gibson et al., 2006; Kicheva et al., 2007; Landsberg et al., 2009; Mao et al., 2011; Ohsawa et al., 2012). However, there have been no recent reported efforts to develop media specifically for Drosophila organ culture, perhaps owing to the inherent difficulty involved in such a project. In this study, we decided to follow a multidisciplinary approach towards developing a culture medium that supports high levels of proliferation in freshly explanted discs by applying mathematical tools such as design-of-experiments and response-surface methodology to accelerate the exploration of the experimental parameter space and to account for multifactorial interactions (Mandenius and Brundin, 2008).

To overcome a bottleneck in the number of dissected wing discs that could be prepared to test multiple media formulae, we hypothesized that a cell line that was derived from wing imaginal discs called Clone.8 (Cl.8) could be used as a proxy for wing disc proliferation (Currie et al., 1988). In the first step of the optimization pipeline, concentrations of media components were varied based on mixing experiments and response-surface methodology (RSM) where proliferation was the chosen response variable. Application of ANOVA results in a fitted polynomial model of the data. In the second stage, an optimized culture medium formula for Cl.8 proliferation was subsequently tested in a disc culture bioassay that scores the number of mitotic cells that can be observed in fixed discs after a certain culture period. Finally, a third, quality control, stage eliminated artefacts that might have arisen during the previous stages by directly monitoring cell divisions with a flexible live-imaging setup.

Our optimization pipeline enabled us to identify crucial medium components that are required to sustain cell division in vitro for significant periods and at high growth rates (>20% growth in under 12 hours, supplementary material Movie 1). Live-imaging studies with the improved medium provide a clear demonstration of frequently observed spatiotemporal clusters of mitotic cells, transient T1 transitions and cell rearrangements, which were not noted in previous live-imaging studies. The frequency and functional importance of patterned mitotic clustering has been a point of uncertainty and observations of cell rearrangements hint that the tissue is more dynamic than previously thought (Adler and MacQueen, 1981; Aegerter-Wilmsen et al., 2010; Milán et al., 1996). Our work allows the study of growing wing discs in vitro and provides a pipeline to further improve culture conditions in an effort to recapitulate wing disc development in vitro. Finally, the medium development pipeline presented here can be modified to optimize other model organ systems efficiently and rationally, given severe technical constraints of acquiring the large numbers of organs needed for high-throughput assays that cover a large experimental parameter space.

Cl.8 cells were maintained in the M3-based reference medium, M3RM medium (elsewhere called Cl.8 medium), which contains both inactivated mammalian serum and fly extract (Fig. 1A; supplementary material Table S1) in an incubator at 25°C. For maintenance, cells were passaged at a 1:3 or 1:4 dilution every 3-4 days in T25 plates (Corning). The cell growth assays were initiated with 40,000 cells per well (1.18×10⁵ cells/cm²) in 96-well plates (Greiner Bio-one Cellstar) in 100 μl of culture medium and incubated for 4 days. Each independent experiment was performed in triplicate (three wells) with random assignment of culture conditions with a fourth well containing medium as a blank. Fetal bovine serum (AG biochrom) was heat-inactivated at 56°C for 30 minutes. Fly extract was prepared from yw flies according to the recipe of Milner and colleagues (Currie et al., 1988).

Staged 4-day-old larvae (96-100 hours after egg deposition) were washed once with 70% ethanol and twice with PBS. Wing discs were dissected in a horizontal flow hood directly in culture medium drops on siliconized glass (Sigmacote, Sigma). Care was taken to not rupture the gut during dissection in order to preserve sterility. Dissected discs were transferred to non-adhesive 24-well (Hydrocell, Nunc) plates with a micropipette. Non-adhesive wells proved to be slightly beneficial for long-term culture compared with traditional tissue culture-treated wells. The disc cultures were stored either in a 25°C incubator or in a Styrofoam box at room temperature (22°C).

Wing disc cells dramatically expand their apical area during M-phase, enabling rapid visual detection of dividing cells with a YFP-expressing septate junction marker (LAC::YFP) (Kyoto Stock Center). For wing disc proliferation experiments, the number of large, M-phase cells was counted from z-stack confocal images within a quadrat focused on the wing pouch after a given period of culture followed by fixation with paraformaldehyde solution. Investigation of the dynamics of cell division on the apical surface relies on a line with GFP inserted into the E-Cadherin locus (Huang et al., 2009). The quadrat was defined as 135 μm², corresponding to the field of view available on the Zeiss 710 with a 63× oil microscope set to a digital zoom of one. For 3rd instar wing discs (96-120 hour AEL), the number of total counted mitoses observed in discs that are fixed immediately is relatively independent of disc size. For example, for staged discs at 96 hours AEL, we observed 32±13 mitoses/quadrat (n=19); for staged discs at t=120 hour, we counted 36±12 mitoses/quadrat (n=38). Thus, the bioassay is relatively robust to variations in disc size.

Approximately 10 μl of the growth medium containing 1-2 wing discs was transferred by a 20 μl pipette to an uncoated glass bottom dish with a 20 mm diameter glass bottom (Mattek). The disc was positioned with a tungsten needle with the peripodial membrane facing downwards. Next, a Millicell polycarbonate cell culture insert (8.0 μm pore, Millipore #PI8P01250) was placed on top of the culture drop after clipping the feet (which leaves ~60 μm gap between the plastic rim and the filter). Filter paper dampened with PBS surrounded the rim of the glass bottom dish and the lid was sealed with parafilm. Both a single-point laser confocal microscope (Zeiss 710) and a spinning-disc confocal microscope (Andor) were used for obtaining time-lapse images with the spinning-disc confocal providing superior results during long-term live-imaging experiments, probably owing to reduced phototoxicity.

The sources and concentrations of the reference serum-containing medium are listed in supplementary material Table S1. Other reagents include: Schneider's Drosophila Medium 1× (Invitrogen, #21720, lot numbers are listed in the text), IPL41 (Invitrogen, #11403), Sf900 II SFM (Invitrogen, #10902), TC-100 (Invitrogen, #13055) and D22 (US Biological, #D9600).

Unless otherwise indicated, cell counts were estimated using the CyQUANT Direct Cell Proliferation Assay Kit (Invitrogen). Fluorescence measurements were acquired with a plate reader (Promega Glomax MultiDetection system). Cl.8 cells are semi-adherent cells, do not show contact inhibition and continue to grow as clusters after reaching confluency. To provide a more uniform signal, cells were manually resuspended before the fluorescence reading and then spun down for 1 minute at 470 g to remove air bubbles. Background intensity blanks were subtracted from raw intensity readings of each of the triplicate conditions. A calibration curve showed a linear relationship between background subtracted fluorescent intensity readings and known cell numbers counted from a hemocytometer (supplementary material Fig. S1).

Experimental design and data analysis were performed using Design-Expert software (version 8.0.5, Stat-Ease, MN, USA).

Improving culture conditions for wing discs requires developing high throughput methods to quantify how the medium composition affects disc vitality and growth. Execution of such a systematic analysis of nutritional requirements with explanted wing discs is hindered by the bottleneck posed by the delicate manual dissection step (~2-3 minutes/larvae for an experienced operator attempting sterile dissection) and the lack of a sensitive quantitative assay measuring either wing disc health or proliferation. Thus, the development of a high-throughput optimization pipeline became the first challenge in improving the reference culture medium (Fig. 1).

To overcome the dissection bottleneck, we chose to use the Cl.8 cell line as a high-throughput amenable proxy for wing discs and the Cl.8 maintenance medium as the starting medium composition (here called M3RM) (Fig. 1A,B; supplementary material Table S1). Cl.8 cells were derived close to 30 years ago and their physiological requirements could have drifted significantly from primary disc cells. However, a recent report by modENCODE suggests that Cl.8 cells maintain a transcriptional profile similar to that of cells near the intersection of the A/P and D/V boundaries in the wing disc (Cherbas et al., 2011). Cl.8 cells have been shown to be very sensitive to exogenous growth factors and, in addition to fetal bovine serum (FBS) and insulin, also require a fly extract that contains heat-stable proteins. For example, previous studies on Cl.8 cells were instrumental in identifying the family of imaginal disc growth factors and adenosine deaminase-related growth factors (Kawamura et al., 1999; Zurovec et al., 2002). Although other wing disc cell lines exist and are now available, Cl.8 cells historically have been the most frequently employed in biochemical studies, including in our lab, and this availability resulted in the initial selection of Cl.8 cells. Pilot experiments further supported our decision to use Cl.8 cells as a proxy for wing disc growth.

Additional efficiency in exploring the parameter space was achieved with multifactorial screens based on design-of-experiments methodology, an approach frequently used in the optimization of processes (Jeon et al., 2010). Multifactorial screens can provide efficient identification of key medium components. In this study, we looked at varying the composition of basal medium by mixing commercially available media in different proportions while keeping the other components constant, including insulin, fly extract (FEX) and fetal bovine serum (FBS). Based on these results, we then employed a further round of response-surface methods (RSM), which provide a statistical model of Cl.8 proliferation as a function of the best basal medium and concentrations of growth supplements. The optimal composition for Cl.8 cells was then tested on wing disc cultures. For this step, we developed a proliferation bioassay to compare different pools of explanted wing discs (see Materials and methods). Finally, we developed a live-imaging method that enables observation of cell proliferation over an extended period of time and can easily enable media exchange for long-term culture experiments (Fig. 1D).

First, we performed a blending screen of six different basal media for insect cell culture, while keeping the supplements (FBS, FEX and insulin) at the reference concentrations in M3RM (supplementary material Table S1). Of the media tested, M3, Schneider's and D22 are frequently used for the maintenance of Drosophila lines (Echalier, 1997). IPL-41, TC-100 and Sf900 II have been successfully used in the cultivation of Drosophila S2 cells (Batista et al., 2008; Bovo et al., 2008). The media-blending experiment was designed as an efficient method to search the parameter space of concentrations of essential nutrients such as amino acids, salts, sugars and vitamins by varying the relative proportions of the basal media in the culture medium mixture.

A simplex centroid design for fitting a quadratic model was used resulting in 96 runs with a total of four blocks (each block was a 96-well plate) (Scheffe, 1963). The simplex centroid design includes the vertices (each of the undiluted basal media), the center points of the edges (50:50 mixes of two different basal media), as well as triple blends (three basal media in proportions of 1/3, 1/3, 1/3) and the overall centroid (all six media in equal proportions). Summaries of the experimental design, ANOVA analysis and model coefficients are presented in supplementary material Tables S3-S5 and notes on the data analysis are provided in Appendix S1.

Surprisingly, Schneider's medium, which is not routinely used to culture Cl.8 cells, performed much better than any of the other basal media (including M3) or any blend without requiring any type of adaptation (Fig. 2A,B, ~60% higher density). All blends containing Schneider's medium showed a dilution effect in relation to the amount of Schneider's medium in the blend, and both D22 and Sf-900 II were toxic for Cl.8 cells.

One explanation for the difference in Cl.8 proliferation between Schneider's and M3-based formulations could be changes in the synergistic effects of supplements in the different basal media. Preliminary experiments in our lab showed that FEX was crucial to growth in M3 medium but that insulin was surprisingly dispensable (data not shown). To quantitate the effects more clearly, we employed response-surface methods (RSMs) to produce topological maps of proliferation as a function of supplements for both M3 and Schneider's basal media. An IV-optimal design distributed over three blocks served as the final design in which the basal media were treated as categorical variables (design points shown in Fig. 2C,D). The concentration of each supplement ranges from complete absence to twice the concentration of the standard supplement concentration (supplementary material Table S6). Experimental details and data analysis notes are provided in supplementary material Appendix S1 and Tables S7-S9.

Graphically exploring the response-surface model parameter space provided several insights (Fig. 2E-H). Notably, Cl.8 cells proliferated more in Schneider's than in M3 media only when insulin was added to Schneider's medium (Fig. 2H). These results imply that Cl.8 cells had not lost the inherent ability to respond to insulin, but rather that the response of insulin depends on the composition of the basal medium.

Furthermore, higher levels of FEX appeared to partially replace the need for FBS as a growth supplement (Fig. 2G; supplementary material Table S9). These experiments demonstrate that response-surface methods can be an effective tool for exploring how extrinsic signals interact with each other and are useful for revealing local and global maxima within the experimental design space. Thus, by employing RSMs we were able to explain the differences in Cl.8 growth rates in Schneider's or M3-based formulations. In addition, we could use the RSM results to predict optimal culture medium formulations for the different variables that we tested.

Wing discs cultured in the reference medium (M3RM) showed a quick decline in proliferation (Fig. 3A). Based on the Cl.8 cell optimization data, we formulated a new culture medium called Wing Disc Medium 1 (WM1). The new medium eliminates the use of FBS (to rely on a single fly-based ‘serum’) and increases FEX to 5% with a slightly higher concentration of insulin (3.1 μg/ml) in a Schneider's medium background. WM1 more than doubled the proliferation rate of wing disc cells (compared with M3RM) after both 5 hours at 25°C and 18 hours at room temperature (~22°C) (Fig. 3B,C). Therefore, WM1 sustains cell proliferation in wing discs at rates closer to in vivo rates for significantly longer periods. Based on this quantifiable improvement, we are now robustly performing live-imaging assays of wing discs cultured in WM1 with high proliferative activity to examine a large range of experimental variables.

To illustrate the potential of WM1, we found that the improved formula enables observations of discs showing more than 20% proliferation under high-resolution imaging conditions (one example is shown in Fig. 4A,B). Strikingly, cell divisions do not follow a random distribution; rather, cell divisions are often grouped in spatiotemporal clusters (Fig. 4A). This observation reinforces previous work by Milán et al. on spatial clustering and highlights a new level of cell division organization at the temporal level (Milán et al., 1996). Several examples of spatiotemporal clusters are presented in Fig. 4B. Here, we considered that cells were part of a cluster if they divided in sequence along a given direction. Selected clusters are color coded such that groups of direct neighbors that form a cluster share the same color. When direct neighbors divided along different directions we chose two different colors. A more-detailed quantitative analysis of mitotic clusters in the growing wing discs is the subject of another study.

As a second demonstration of the improved method, time-lapse analysis also revealed multiple instances when cell topology and polygon count frequently changed between timeframes. This counters a common assumption that there are rarely cell rearrangements during wing disc growth (Gibson et al., 2006). We found that cell topology at this stage of development is very dynamic. Four edge vertices appear to be inherently unstable and often oscillate between two possible configurations through T1 transitions. We also observe that short edges are equally unstable and that in positions in which several short edges are in close proximity this can even lead to cell intercalations during the latter stages of 3rd instar discs (Fig. 4C).

Hemolymph constitutes the external milieu of imaginal tissues and provides necessary requirements for tissue homeostasis, growth and differentiation. One of the main challenges in obtaining an optimal culture medium for wing discs stems from the largely undefined complexity of hemolymph. Basal insect cell culture media were developed partly by mimicking the major ion and amino acid composition of hemolymph. The differences between different culture media reflect the given state of knowledge on hemolymph composition and iterative experimentation. Other unknown metabolites are provided by yeast extract, FBS or FEX. This study shows that the differences in media composition can have strong effects. For example, the composition of the basal medium affects the strength of the proliferative response to insulin.

The importance of fly extract for supplementing the basal medium raises the question of what are the active factors necessary for wing disc growth in vitro and in vivo? We have confirmed by ultrafiltration that the crucial components of fly extract are of large molecular weight, most likely proteins (data not shown). Recently, ferritin was identified as a component of fly extract that stimulates Cl.8 growth (Li, 2010). We have found that mammalian ferritin can stimulate growth, but is not sufficient for replacing fly extract for long-term culture of Cl.8 cells in serum-free media (data not shown).

Importantly, nonlinear interactions between the large number of culture medium components makes identification of optimal concentrations a nontrivial ‘systems-level’ task. Our results point to the necessity of redeveloping basal media for Drosophila cells and organs in a fully documented, rational and quantitative manner, and in functionally characterizing the multifactorial interactions between growth factors found in hemolymph.

Based on the importance of the Drosophila homologs of insulin to the regulation of wing size, we were surprised that insulin showed such a weak effect on Cl.8 growth. Furthermore, it seemed to contradict the earlier findings of Milner and colleagues on related cell lines (Cullen and Milner, 1991), but appeared to agree with observations of insulin by Bryant and colleagues (Kawamura et al., 1999). We initially attributed the difference to changes in the Cl.8 cell line over time (over 20 years have elapsed between the studies). However, the finding that Cl.8 cells still respond to insulin in other basal media such as Schneider's (and IPL-41, data not shown) has added an interesting twist to the characterization of the behavior of the cell. The contextual response of Cl.8 cells to insulin highlights the crucial influence of the basal medium composition on results obtained from in vitro culture. Response-surface methods were thus instrumental in identifying Cl.8 cells as a possible model to study conditional insulin resistance, and thus raises many additional questions that remain to be addressed in future studies.

One of the surprising observations made from live-imaging experiments of wing discs in WM1 is the extent of spatiotemporal clustering of cell divisions, a phenomenon first described by the lab of Garcia-Bellido (Milán et al., 1996). The extent to which cell divisions are clustered, and thus possibly coordinated, has remained somewhat controversial given the difficulty in interpreting clusters from fixed images (Aegerter-Wilmsen et al., 2010). In general, we find that live-imaging of discs that show higher numbers of dividing cells also frequently exhibit more prominent clustering of cell division, highlighting that these phenomena are related to how actively a disc is dividing. These clusters could have been missed in previous live-imaging studies if insufficient levels of proliferation were sustained by the medium employed. Most cells in late 3rd instar wing discs are stalled in G2 (Neufeld et al., 1998). The observed cell division clustering in the wing disc suggests that the decision to start M-phase might be regulated at some level between neighboring cells, perhaps through gap junction communication (Bryant and Fraser, 1988; Jursnich et al., 1990). Indeed, a proper appreciation of spatiotemporal clustering might be required to properly interpret some experiments that rely on clonal analysis (Griffin et al., 2009). As cell divisions are often clustered, this phenomenon might occur more frequently than expected, which will be examined in future work.

The observed dynamics of epithelial remodeling in 3rd instar wing discs was another unexpected surprise given the assumption in the field of low levels of cell rearrangements (Gibson et al., 2006). Sub-optimal media conditions may have prevented observation of the full dynamics of organ growth, suggesting that the forces generated by cell proliferation and growth fade away quickly, masking the extent to which rearrangements and translocations normally happen. Our results highlight that cell topology during wing disc development shows unexpected dynamics that deserve further characterization (Classen et al., 2005; Griffin et al., 2009). Therefore, our results underscore the need to consider the medium composition and performance as an important variable in live-imaging studies of wing imaginal discs.

This study represents the first rational, multifactorial analysis of the growth requirements of Drosophila wing imaginal disc culture, and provides an optimization pipeline that can be adapted to improve wing disc culture techniques further. By identifying a crucial role for fly extract and an important synergy between insulin and Schneider's medium, the average permissive period for cell division of cultured wing discs is extended by more than 100% compared with the reference medium used in establishing the Cl.8 cell line. WM1 allows robust cell proliferation to take place for more than 18 hours at room temperature. This time window could be appropriate for the study of many questions of interest, such as the role of mechanical forces in wing growth development. Future improvements to WM1 will likely require the development of an optimal basal medium that improves on the capacities of Schneider's medium, the fractionation of fly extract to identify its crucial components, as well as a systems-level analysis of the adaptation of wing discs to culture conditions.

This culture method provides an efficient means to combine genetic, mechanical and pharmacological techniques for studying organ development and physiology. The strategy of identifying a proxy for the initial stages of culture optimization followed by a more limited optimization step with explanted organs might be a useful approach for other model organ systems. This method could serve as a starting point for the future development of culture techniques, allowing the full recapitulation of wing disc development in vitro.

We thank Florenci Serras for helpful guidance at the start of the project; J. Bischof, G. Hausmann and G. Reeves for providing critical comments on the manuscript; and S. Luschnig, C. Lehner, B. Handke, E. Hafen, G. Schwank, E. Brunner, W. Boll, D. Brunner and N. Dube for helpful feedback during the course of the project.