An Imaging Flow Cytometry-based approach to ...

Methods 82 (2015) 74–84

Contents lists available at ScienceDirect

Methods journal homepage: www.elsevier.com/locate/ymeth

An Imaging Flow Cytometry-based approach to analyse the fission yeast cell cycle in fixed cells James O. Patterson a, Matthew Swaffer a, Andrew Filby b,c,⇑ a

Cell Cycle Laboratory, London Research Institute, Cancer Research UK, 44 Lincoln’s Inn Fields, Holborn WC2A 3LY, UK FACS Laboratory, London Research Institute, Cancer Research UK, 44 Lincoln’s Inn Fields, Holborn WC2A 3LY, UK c Flow Cytometry Core Facility, Newcastle Biomedicine, Newcastle University, Newcastle-upon-Tyne NE1 7RU, UK b

a r t i c l e

i n f o

Article history: Received 26 November 2014 Received in revised form 28 March 2015 Accepted 8 April 2015 Available online 4 May 2015 Keywords: Fission yeast Cell cycle Imaging Flow Cytometry

a b s t r a c t Fission yeast (Schizosaccharomyces pombe) is an excellent model organism for studying eukaryotic cell division because many of the underlying principles and key regulators of cell cycle biology are conserved from yeast to humans. As such it can be employed as tool for understanding complex human diseases that arise from dis-regulation in cell cycle controls, including cancers. Conventional Flow Cytometry (CFC) is a high-throughput, multi-parameter, fluorescence-based single cell analysis technology. It is widely used for studying the mammalian cell cycle both in the context of the normal and disease states by measuring changes in DNA content during the transition through G1, S and G2/M using fluorescent DNA-binding dyes. Unfortunately analysis of the fission yeast cell cycle by CFC is not straightforward because, unlike mammalian cells, cytokinesis occurs after S-phase meaning that bi-nucleated G1 cells have the same DNA content as mono-nucleated G2 cells and cannot be distinguished using total integrated fluorescence (pulse area). It has been elegantly shown that the width of the DNA pulse can be used to distinguish G2 cells with a single 2C foci versus G1 cells with two 1C foci, however the accuracy of this measurement is dependent on the orientation of the cell as it traverses the laser beam. To this end we sought to improve the accuracy of the fission yeast cell cycle analysis and have developed an Imaging Flow Cytometry (IFC)-based method that is able to preserve the high throughput, objective analysis afforded by CFC in combination with the spatial and morphometric information provide by microscopy. We have been able to derive an analysis framework for subdividing the yeast cell cycle that is based on intensiometric and morphometric measurements and is thus robust against orientation-based miss-classification. In addition we can employ image-based metrics to define populations of septated/bi-nucleated cells and measure cellular dimensions. To our knowledge, this is the first use of IFC to study fission yeast and we are confident that this will provide a springboard for further IFC-based analysis across all aspects of fission yeast biology. Ó 2015 Elsevier Inc. All rights reserved.

1. Introduction The eukaryotic cell cycle is a highly regulated and ordered process that drives cellular expansion and renewal. During a single cycle, the genome is faithfully duplicated and subsequently segregated across the plane of cytokinesis to give birth to two daughter cells with an identical complement of DNA. Although the eukaryotic family is highly diverse, including both single and multicellular organisms, the overarching principles and regulatory mechanisms of the cell cycle are conserved from yeasts to humans. Failure to correctly regulate cell cycle entry, progression and exit ⇑ Corresponding author at: Flow Cytometry Core Facility, Newcastle Biomedicine, Newcastle University, Newcastle-upon-Tyne NE1 7RU, UK. E-mail address: andrew.fi[email protected] (A. Filby). http://dx.doi.org/10.1016/j.ymeth.2015.04.026 1046-2023/Ó 2015 Elsevier Inc. All rights reserved.

can have catastrophic outcomes, particularly in mammals where such dis-regulation drives cancer development [1]. A detailed understanding of the eukaryotic cell cycle is therefore central to current and future cancer research. Fission yeast (Schizosaccharomyces pombe) is a single cell, rod shaped eukaryote that grows by tip extension and divides by medial fission. The fission yeast cell cycle, like that of other eukaryotes, can be sub-divided into G1, S-phase, G2 and mitosis. Cell growth predominantly occurs during G2 which constitutes 70% of the cycle time [2]. Fission yeast is highly amenable to genetic manipulation and over the past 40 years has provided profound insights into how the eukaryotic cell cycle is controlled at the molecular level [3]. Importantly, many of these findings are directly applicable to understanding how human cancers arise and potentially how they could be treated or prevented.

J.O. Patterson et al. / Methods 82 (2015) 74–84

Conventional Flow Cytometry (CFC) is a powerful analytical tool for the measurement of cellular phenotype and function based on the measurement of fluorescence and scattered light as cells pass in suspension through a laser interrogation point. CFC is a mainstay for the analysis of the mammalian cell cycle (Fig. 1A) based on the fluorescent detection of DNA stains and cell cycle-associated markers. Fluorescent DNA binding dyes are routinely used to measure the relative levels of DNA within cells to distinguish between G1 (2N), S-phase and G2/M (4N) cells (Fig. 1B). CFC has played a fundamental role in the understanding of the mammalian cell cycle, especially in the context of dis-regulation and cancer. In contrast, CFC-based dissection of the fission yeast cell cycle is less straightforward than for mammalian cells because cytokinesis occurs after DNA replication (Fig. 1C). As a result, G1 (2 1C) cells and G2 (1 2C) cells have the same overall 2C DNA content and cannot be distinguished by measuring the total integrated fluorescence (pulse area) of a DNA binding dye (Fig. 1D). It has been elegantly shown that other DNA-derived pulse parameters such as width can be utilised to distinguish G1 and G2 cells based on the assumption that cells with 2 1C nuclei should orientate through the long axis and as such have an increased time of flight through the laser beam that translates into an increased pulse width [4]. While this approach has merit, we feel that a truly accurate analysis of the fission yeast cell cycle requires an image-based approach due to the fact that not all cells will orientate correctly through the laser interrogation point and may rotate around the x, y and z axes (Fig. 2B). This will likely result in a degree of miss-classification because the measured width (time of flight) may not reflect the true cellular arrangement of the DNA content, particularly if the cells rotate along the x and z axes. The orientation of the yeast through the flow cell may also be affected by length so there is a strong possibility that the resulting miss-classification could affect cells in particular phases more than others, skewing the inferred cell cycle distributions. Furthermore, the method described by Knutsen et al requires validation using a total of three different cytometric platforms; namely analytical flow cytometry, cell sorting and microscopy to confirm if CFC derived pulse data and the microscopical images correlate for the majority, if not all, of the cells [4]. To this end we sought to develop an approach that used a single, unified cytometric platform and chose Imaging Flow

75

Fig. 2. Schematic depicting issues relating to hydrodynamic focusing in CFC. (A) Cell tipping results in the miss-classification of cell cycle stages by CFC. (B) Demonstration of the possible axes of yeast rotation within the flow cell.

Cytometry (IFC) as it maintains the high-throughput, multi-parameter, relative-quantitation of CFC, while providing the spatial information of a fluorescent microscope. Briefly, cells in suspension are flow-focused prior to interrogation by various excitation lasers and a bright-field LED source. The emitted, transmitted and scattered light is collected by a microscope objective (20, 40 or 60, user selectable) positioned behind the flow cell and spectrally decomposed through an array of long pass dichroic mirrors angled to reflect light of a given wavelength range onto a specified location on a 12-bit CCD camera operating in time delay integration mode (TDI). The result is a system that can image hundreds of moving objects per second, producing up to 12 spatially registered, spectrally decomposed fluorescent, bright-field and dark field images per event [5]. We have successfully applied IFC to a number of mammalian cell studies where the hybrid nature of the technology uniquely provided all the necessary cytometric data from a single, unified platform [6–8]. Here we show that IFC analysis of fission yeast preserves the high-throughput, objective analysis afforded by CFC. However, due to added benefit of multi-spectral imagery it can dissect more cell cycle stages with greater accuracy than CFC and was robust against mis-scoring issues related to rotation of cells about the z axis, while also allowing cells rotated around the x axis to be excluded from the analysis, none of which is possible by CFC. It can therefore replace laborious and often subjective manual microscopy-based approaches for analysing the progression of synchronised cultures or identifying and measuring septated/bi-nucleated cells. It also proved particularly powerful for the analysis of mutant yeast strains with altered cellular dimensions where CFC-derived pulse width was unable to resolve putative bi-nucleated cells. Here we show that IFC can be used to reliably quantify all parameters in a single assay.

2. Materials and methods 2.1. Yeast strains, culture and preparation

Fig. 1. Stages and DNA distribution histograms of a typical proliferating mammalian (A, B) and fission yeast (C, D) cell cycle.

The strains used in this study are listed in Table 1. Cell culture conditions and growth media were as previously described [9]. Haploid fission yeast cells (PN1) were grown in EMM4S media. 5 106 cells where harvested during exponential growth, fixed in 1 ml 70% ice cold ethanol and stored at 4 °C. Cells were then washed and re-suspended in 1 ml 50 mM sodium citrate, treated

76


Table 1 The details of the strains used in this study. Strain

Mating type

Genotype

Source

PN1 MS278 PN369

hhh-

WT (972) cdc2-asM17::bsd wee1-50

Our stock This study Our stock

with 0.1 mg/ml RNase A (SIGMA–ADLRICH, UK) and incubated at 37 °C overnight. Then 2 lg/ml Propidium Iodide (SIGMA-ADLRICH, UK) and 2 lg/ml FITC (F3651, SIGMA–ADLRICH, UK) stains was added. Cells were synchronised by adding 1 lM 1-NmPP1 (Toronto Research Chemicals) to exponential growing cdc2-asM17 cells (MS278) (32 °C, YE4S) for one generation to arrest in G2. Cells were then washed (3) and released into pre-warmed fresh media. Samples were collected during the subsequent synchronous cycle and process for fixation and staining as above. MS278 was derived from published strains [10]. Small cells where generated by growing the temperature sensitive wee1-50 allele in YE4S at the permissive temperature (25 °C) and shifting half of the culture to the restrictive temperature (36 °C) [11]. 3.5 h after the shift, samples were collected from cultures at 25 °C and 36 °C. 2.2. CFC data acquisition Samples were sonicated (20 s) using a sonication probe (JSP Inc., USA) in a volume of 1 ml at a density of 5 106 cells/ml in a 5 ml polystyrene FACS tube (BD Falcon, USA) to reduce the number of aggregates (data not shown). Samples were then acquired on a BD LSRFortessa system (Becton Dickinson, USA). The instrument laser and detector array configuration was as follows: 488 nm with an octagon, 405 nm with an octagon, 561 nm with an octagon, 355 nm with a trigon and 633 nm with a trigon. The forward and side scatter detectors were set at 620 v and 300 v, respectively. PI was excited using a 561 nm laser line and fluorescence was collected using a 610/20 band pass filter using a PMT voltage of 730 v. FITC was excited by a 488 nm laser line and fluorescence collected using a 530/30 filter and a PMT voltage of 500v. All samples were acquired in linear format with the pulse area, width and height selected. The flow rate was set at ‘‘low’’ in order to provide the best measurement resolution and a minimum of 10,000 events was collected. 2.3. IFC-data acquisition Samples were re-suspended in a volume of 500 ll at a density of 107 cells/ml in a 1.5 ml tube. Samples were then sonicated, as described above, prior to loading on a fully ASISST and Cyto-Cal Bead (FC3MV, Thermo Fisher, USA) calibrated ImageStream X (ISx) system (Amnis, Seattle, USA). The system was configured with 405 nm, 488 nm, 561 nm and 642 nm excitation laser lines, the multi-mag option to image up to 60 (0.25 lm/pixel) and two CCD TDI cameras delivering up to 12 imaging channels. For acquisition, bright-field illumination was selected in channels 1 and 9 and the 488 nm and 561 nm lasers (FITC and PI excitation, respectively) laser powers were set so as to avoid saturation in the maximal camera channels (FITC collected in CH2 and PI collected in CH4). Typically the powers used were 100 mW (488 nm) and 100 mW (561 nm). Single stained yeast samples were also acquired with bright-field illumination turned off for the purpose of compensation [12,13]. To ensure that there was no cross contamination of the FITC single stained sample with PI, a 2% bleach-based solution was loaded into the sample line followed by a tube of water. Yeast were imaged at 60 using the low speed, high sensitivity fluidics mode and a total of 200,000 cells

were collected from within a gate set on the area and aspect ratio of the CH1 bright-field mask in order to eliminate any debris or speed beads from the file. 2.4. Analysis of CFC-derived data CFC derived data was analysed using FlowJo X (Treestar Inc., USA) as in Fig. 3. Briefly, single yeast was identified based on the FSC-A versus SSC-A. Doublets and aggregated yeast presented with an increase in both parameters and were present as a clear population in non-sonicated samples (data not shown). The area and width of the PI fluorescence collected in the yellow 610/20 filter was used to subdivide the fission yeast into 3 major phases. Putative G2/M cells were gated based on the having a fluorescence area value of a 2C cell and a pulse width corresponding to a single nuclear focus (i.e. before separation of DNA masses at anaphase onset). Putative G1/M cells were identified and gated on the basis of also having a 2C fluorescence area value, but an increased width suggesting that it contained two spatially distinct nuclei orientated down the long axis (after separation of DNA masses at anaphase onset). Finally, putative S-phase cells were identified and gated based on having the fluorescence area greater than that of a 2C cell in concert with a width of a bi-nucleate. 2.5. Analysis of IFC-derived data The raw image files (.rif) created from the ISx acquisition were analysed using the IDEAS software (Amnis, USA) as outlined previously [5,6]. Briefly, the single stained fluorescent control .rif files for FITC and PI were loaded into the IDEAs compensation wizard to form a merged compensated image file (.cif). The software was instructed as to which channels the peak fluorescence for a given dye should occupy (FITC in CH2 and PI in CH4). IDEAS then calculated the compensation values based on the co-efficient for the slope derived from plots of specific fluorescence (x-axis) versus fluorescence in the cross talks channels (y-axis) of the positively stained population. Each trend line was manually inspected to ensure that no outliers (autofluorescence, debris etc.) were skewing the best-fit line in a way that would affect the compensation value. If outliers were noted, then the compensation population was manually gated so as to eliminate any outliers. A number of images from the FITC and PI single positive populations were randomly selected in order to test the compensation values on a per pixel basis and view fully compensated imagery. The validated compensation matrix is shown in Supplemental Fig. 1 and was applied to the fully stained .rif files to create .cif files and data analysis files (.daf) for analysis. The .daf files served as the interface with which to analyse the yeast cell cycle using a number of inbuilt and custom generated pixel-based features derived from either default or custom channel masks [12]. 2.6. Cell sorting Dual FITC/PI labelled yeast were run on a FACSAria FUSION cell sorter (BD) calibrated using CS&T (BD, USA). The instrument laser and detector array configuration was as follows: 488 nm with an octagon, 405 nm with an octagon, 561 nm with an octagon, and 633 nm with a trigon. An analogous acquisition and gating strategy was set up as described for CFC analysis using the same band pass filters. The major gated populations (G2/M, G1/M, S) and miscellaneous population (putative side-by-side doublets as proposed by Knutsen et al) were sorted into 1.5 ml tubes containing 10 ll of PBS using a 100 lm nozzle, a system pressure of 20 PSI and the 4-way purity sort mode. The gated sort decisions are shown in Fig. 5A. The sorted populations were re-acquired in order to determine sort purity but also to determine the influence of rotation on


the identity of the re-measured sorted populations. Each sorted population, as well as the unsorted parental population, was then acquired on an ISx IFC-system as outlined above to determine the correlation between selecting cells in a given cell cycle stage based on CFC pulse parameters and the image-based, rotation independent appraisal of the same.

3. Results 3.1. A combination of DNA pulse area and width identifies three major populations of fission yeast Fission yeast were fixed and stained for DNA (PI) and protein (FITC) content – as a fluorescent surrogate for cell volume. PI was chosen as the DNA stain of choice over the other commonly used stain Sytox green [14] so that DNA signal could be spectrally resolved from the FITC stain. We then acquired them on a BD LSRFortessa CFC system (see Section 2) to replicate the method described by Knutsen et al (Fig. 3) [4]. First we gated on putative single yeast based on FSC-A/SSC-A (Fig. 3A) and checked the resolution of the DNA prolife as a PI histogram (Fig. 3B). Using a combination of DNA pulse area and width we could detect three major populations corresponding to the G2/M, G1 and S-phases (Fig. 3C) in accordance with Knutsen et al. Moreover the frequencies obtained (G2/M = 84.9%, G1 = 5.39% and S = 6.01%) were in line with the duration of time spent by a cell in each phase [2]. Back-gating analysis revealed that cells in G1 & S (black dots)

77

had an elevated FITC signal (2-fold) compared to those in G2/M (grey dots). Interestingly, FITC intensity positively correlated with the SSC-A signal (Fig. 3D) whereas there was a negative correlation between the FSC-A and FITC signals (Fig. 3E) and between FSC-A and SSC-A (Fig. 3F). These observations suggested that SSC-A may be a better correlate of cell size by CFC compared to FSC-A, as has been described for mammalian cells [15]. However, in line with pervious observations made with mammalian cells [16], fluorescently labelling cellular proteins (FITC in this instance) also provided a strong correlate with putative cells size. As expected, cells in G2 displayed the broadest range of FITC staining given most cell growth occurs in G2 [2]. Collectively, these data recapitulate the methodology of Knutsen et al [4] in potentially dissecting bi-nucleated cells from mono-nucleated cells using CFC as well as demonstrating cell cycle progression correlates with FITC intensity and thus cell size. To overcome potential machine-dependent differences in FSC based size measurements that are likely due to properties of the FSC obscuration bar [17], we would advocate the use of FITC signal as a correlate for cell size in fission yeast by CFC. 3.2. IFC-derived intensiometric and morphometric parameters can subdivide the fission yeast cell cycle Next we acquired the same PI/FITC stained yeast samples on an ISx IFC system. After calculating and applying spectral compensation we used a combination of intensiometric and morphometric features to subdivide the fission yeast cell cycle. Although the

Fig. 3. CFC based assignment of cell cycle stages in fission yeast. (A) Gating on FSC-A and SCC-A to identify single yeast cells. (B) DNA intensity (PI pulse area) histogram of single yeast cells. (C) 2D plot of PI signal pulse area (610_20-Yellow-A) and pulse width (610_20-Yellow-W) to delineate cell cycle stages. (D–F) FITC single pulse area (530_30-Blue-A), side scatter (SSC-A) and forward scatter (FSC-A) are plotted against each for two populations: smaller G2&M and larger cells G1&S.

78


IFC-based method would be robust against axial rotation along the x and y axes, it would still be affected by rotation around the z axis. We therefore began by eliminating any poorly focused events using the gradient RMS feature of the bright-field image in channel 1 (CH1). The term poorly focused encompasses cells that have been imaged wholly outside the optimal focal plane as well as those rotated around the z-axis so that one pole falls outside it. It also eliminated doublets imaged with one cell in front or behind the other (data not shown). Events with an RMS score >55 were gated as being in good focus (Fig. 4A) and we have previously established this as a reproducible cut off value [12]. It should also be noted that the overall RMS distribution tended to vary slightly with different experimental acquisitions due to variations in the flow conditions and cellular morphologies, but the cut-off remained the same. Next we used the area and the aspect ratio of the default CH1 mask (M01). Single yeast could be identified and gated over the full range of possible lengths while eliminating most multiples and debris by gating for negative correlation between area and aspect ratio (Fig. 4B). We also found that cells that had completely rotated on the z-axis so that they were imaged end-on-end were eliminated even if they seemed to be in god focus (see Fig. 4B, region a). Using this approach, we found that the sample contained 5% of putative doublets. However, to further eliminate doublet events that had not been previously excluded due to their close physical proximity we chose the FITC image specifically as it had a higher signal to noise ratio compared to the bright-field channel image and thus allows for a more discerning masking strategy. We

adapted the FITC CH2 default mask (M02) using the threshold criteria to mask the brightest 75% of pixels in CH2 (see Supplemental Fig. 2A). We then instructed IDEAS to calculate the spot count feature from this mask and identify closely associated multiples based on now having a two spot mask after the threshold adaptation (Fig. 4C, x-axis). We still found that if two events were very closely associated even thresholding M02 to 75% could not identify two masked spots. Therefore we used the minor axis intensity feature of the FITC image (Fig. 4C, y-axis) to extract events with two closely associated yeast from within the one spot population and gated them out appropriately. The minor axis intensity is defined as the narrowest dimension of the elliptical best-fit mask and is intensity weighted (IDEAS software, Amnis). This allowed us to restrict our analysis to a population of in-focus, single yeast that were well aligned in the z axis, something not possible to achieve by CFC. Next we constructed a strategy to identify the yeast cell cycle stages based on the properties of the PI image that would be robust against the orientation of the yeast within the hydrodynamic flow. To this end we used integrated PI intensity (within the default channel 4 mask, M04) to measure the total DNA content, analogous to using CFC pulse area in combination with the major axis intensity feature that is derived from an intensity weighted elliptical best fit mask applied to the channel image reported as the longest dimension (IDEAS software, Amnis). Using this morphometric feature we were able to distinguish bi-nucleated cells from mono-nucleated cells independently of image rotation around the x axis (Fig. 4D). This combination of an intensiometric

Fig. 4. IFC based assignment of cell cycle stages in fission yeast. (A) Gradient RMS feature is used to delineate in focus cells. (B) For single yeast cells area and aspect ratio should correlate. As such single yeast have been gated for in a typical band on a bright-field (CH01) area feature and bright-field (CH01) aspect ratio feature plot. (C) FITC ‘‘spot’’ number and signal width were used to further select for single yeast. (D) 2D representation of PI emission (CH04) based on signal length (major axis intensity) and intensity allows cell cycle stage assignment. Representative images shown in the central and right panel.


79

(y-axis) and a morphometric (x-axis) feature allowed us to resolve G2/M (83.2%), G1/M (6.78%) and S-phase (4.27%) populations as well as mitotic cells undergoing nuclear division (2.14%) from those in G1 (4.64%).

we have demonstrated inherent flaws based on the rotation of the cells in the CFC-based method that are overcome by using IFC. However, it should be noted that overall the use of CFC-derived pulse parameters does provide a workable degree of accuracy.

3.3. A combined CFC sorting, IFC analysis pipeline highlights rotational issues inherent in zero resolution cytometry

3.4. IFC increases the accuracy and dynamic range of cell cycle analysis compared to CFC, especially in small cells

Having derived a set of intensiometric and morphometric parameters to subdivide the fission yeast cell cycle, we wanted to directly compare our approach to the CFC-based method. As mentioned previously, G1 cells are resolved from G2 cells because a bi-nucleated G1 cell should generate a broader PI pulse as it travels through the excitation laser compared to a G2 cell with a single foci. The success of this zero resolution measurement relies on yeast cells always orientating through the long axis as they traverse the excitation laser beam with little rotation around the x, or z axes. Any degree of axial rotation along the x and z directions will affect the accuracy of this measurement (see Fig. 2). As such we reasoned that miss-orientated bi-nucleated cells G1 cells could be miss-classified by CFC as mono-nucleated G2 cells. Furthermore, miss-orientated S-phase cells could be missclassified as two cells stuck together side by side if doublet exclusion by FCS/SSC is inefficient (i.e. miscellaneous cells). To test this hypothesis we took PI/FITC stained fission yeast and ran them through a cell sorter. Once again we were able to identify all putative cell cycle stages based on the DNA pulse area and width (Fig. 5A, all panels). We noted that FITC again acted as a good surrogate for cell size based on SSC-A correlation, but that any observed correlation with FSC-A was largely dependent on the cytometer (Fig. S3). We sorted cells from the four discernable populations into collection tubes (Fig. 5A, right panel). We then re-ran a portion of each sorted sample through the sorter to directly assess the observed purity of the sort by CFC as well as measure the potential effect of axial rotation on the observed population distributions. As a measure of the ‘‘ground truth’’, we also analysed the sorted populations by IFC to determine population purity using a method that was not influenced by orientation in the x axis and was able to eliminate events rotated in the z axis (Fig. 5B). CFC-based re-analysis revealed that the sorted G2&M population contained a small number of cells classified by CFC as G1&M cells (4.5%) that were also present by IFC analysis. However, the sorted G1&M population when reanalysed by CFC also had a significant population of cells classified as G2&M that was not present when we analysed the same sorted sample using our orientation-independent IFC-based method. This clearly demonstrated that G1 cells that rotated in the flow cell are liable to be miss-classified as G2 cells when analysed by CFC (Fig. 2). The small number of G1 cells in the sorted G2&M population likely corresponds to G1 cells that rotated during the initial sort and thus were measured as G2&M by pulse width. Given the overall ratio of G1 cells to G2 cells is so low in the parental population, this accounts for the low number of G1 cells in the ‘‘sorted G2&M’’ population in absolute terms. However it will lead to a significant depletion of G1 cells in a CFC analysed population. Secondly, IFC indicated that the majority of cells in ‘‘sorted S-phase’’ and ‘‘sorted miscellaneous’’ were in fact S-phase. However when re-analysed by CFC, there was a significant underestimation of the proportion of S-phase cells. This is likely because bi-nucleated S-phase cells can rotate in the flow cell and be miss-classified as miscellaneous cells. The presence of 25% G2/M cells in the sorted S-phase and miscellaneous populations as measured by CFC and IFC may have been due to the presence of end-on-end G2/M doublets that were misclassified during the sort as S-phase cells or the fact that we used quite broad sort gates. Either way, they have no impact on the conclusions reached. By applying this rigorous sorting-based approach

Despite the workable degree of accuracy demonstrated by CFC analysis for wild-type cells, there are situations in which we predicted CFC would fail to resolve bi-nucleated cells based on DNA pulse width. For example in small bi-nucleated cells, the two nuclei are likely to be too close together to translate into a resolvable increase in pulse width. We tested this by analysing the temperature sensitive allele wee1-50 that is wild type size at the permissive temperature but approximately half the size at the restrictive temperature. At the permissive temperature the width of the PI signal measured by CFC and the major axis intensity of the PI signal measured by IFC were both able to delineate distinct populations of bi-nucleated cells (Fig. 6, left side panels). However at the restrictive temperature CFC was unable to detect any such population (Fig. 6, upper right panel) whereas they could still be resolved by IFC (Fig. 6, lower right panel). This indicates that as well as having greater accuracy the dynamic range for spatial information provided by IFC is greater than that for CFC. We would like to add that it may be possible to improve the resolution of small bi-nucleated yeast mutants on other CFC systems where the excitation laser focus can be optimised, however for all the bench top machines we have tested so far (BD Fortessa and LSRII systems), we have been unable to achieve adequate resolution. 3.5. IFC-derived Imagery can be used to define bi-nucleated cells Having validated the accuracy of IFC compared to CFC in terms of cell cycle subdivision, we wanted to utilise the multi-spectral IFC-derived imagery to design an objective morphometric approach to identify bi-nucleated cells (G1 and S-phase). In parallel to major axis intensity we devised a masking and spot count approach to define bi-nucleated cells. We adapted the default PI channel mask (M04) (Supplemental Fig. 2B) to distinguish bi-nucleated cells as having two individual masked structures, and mono-nucleated cells, as either a single masked structure or none at all. We were then able to calculate the spot count feature of this mask and, in combination with the aspect ratio feature of the same mask, identify cells with a bi-nucleated DNA arrangement (Fig. 7A, upper panel). The aspect ratio parameter was particularly important to eliminate those cells with a single nucleus that still maintained two masking spots. The mask was not based on a thresholding operator and as such overcame issues with de-masking one nuclear pole that simply had a lower intensity than the other due to tipping in the z-axis (data not shown). A back-gating analysis of bi-nucleated cells revealed that this feature set was highly specific (99%) for cells in G1 and S-phase based on PI intensity and axis intensity (Fig. 7A, lower panels). This provides an independent validation of IFC’s ability to define bi-nucleated cells. 3.6. IFC-derived imagery can be used to define septated cells Fission yeast normally form a septum and divide at a length of 14–15 lM [18]. Measuring cells at septation is a routine approach used to estimate whether mutant yeast are advanced (small at septation) or delayed (large at septation) in cell cycle progression. The multi-spectral IFC-derived imagery presented us with the opportunity to rapidly and objectively identify clearly septated cells and measure cell dimensions over an entire population. Identifying and measuring septated cells is not possible by

80


Fig. 5. A CFC sort – IFC – CFC experimental pipeline reveals hydrodynamic focusing errors. (A) Yeast cells were gated for size and PI intensity before 2D PI analysis (as in Fig. 3). 2D PI analysis allowed assignment of cell cycle stages, and yeast were sorted into four populations (miscellaneous, S-Phase, G2 and M, G1 and M). (B) Subsequently, sorted populations were analysed be IFC and re-analysed by CFC. Left panel shows representative images from the IFC analysis of the CFC sorted populations. The Right panel shows the cell cycle assignment of the CFC sorted populations based on CFC or IFC analysis.

CFC and is classically done using manual microscopic approaches that are both labour intensive, subjective and time consuming. We designed a masking and feature strategy to identify cells with a clearly defined septum. In a septated cell, the FITC image was split into two components through an axis defined by the septum position. We had already constructed a new FITC channel mask using the threshold operator to eliminate closely associated yeast, but this was not able to split a single yeast cell into two masked objects. We modified the M02 FITC mask using the morphology operator to better delineate the cell boundary within the FITC image (see Supplemental Fig. 2C) and used the valley operator on the morphology mask to identify the area of the FITC image where there was a drop in overall intensity (i.e. a valley). We found that in septated cells, this was positioned through the middle of the cell but in non-septated cells, we found that it was positioned away from the centre and biased to one pole. We then took the valley mask and used the AND NOT Boolean operator to combine with the morphology mask to generate a new mask that was made up of two components. In non-septated cells, this generated two masking objects of different sizes. In septated cells the two

masking objects were of similar or equal sizes. We then used the range operator to set a masking area threshold that eliminated the smaller of the two masks that arose in non-septated cells. By deriving the spot count feature from this final mask, we found that septated cells had two masking spots, but non-septated only had one or no masking spots (Fig. 7B, upper panel). We back-gated the septated population onto the single, bi-nucleated yeast population (Fig. 7B, lower panel). In wild type fission yeast septation coincides with S-phase and indeed we found that the majority of cells we defined as septated resided within the S-phase compartment (91.2%). Next we wanted to objectively measure the length of fission yeast from different stages, with or without septa. To do this we adapted the CH1 BF mask (M01) using the ‘‘morphology’’ operation within the IDEAS software (see Supplemental Fig. 2D). This created a mask that now better represented the true boundary of the yeast cells. We then calculated the major axis feature of the morphology mask and plotted this as a histogram (Fig. 7C). As expected, septated and bi-nucleated cells (at the very end of the cycle) had a greater major axis median value than the parent. These data demonstrate that we could use IFC-derived imagery


81

Fig. 6. IFC provides the resolution necessary to perform cell cycle analysis in small cells. wee1-50 cells were shifted from 25 °C to 36 °C for a 3.5 h before analysis by IFC and CFC. (A) CFC analysis cannot delineate mono-nucleated (MN) from bi-nucleated (BN) cells in small cells (i.e. after shift up to 36 °C). (B) IFC PI major axis intensity feature provides the resolution necessary to clearly delineate MN from BN cells in small and wild type cells.

to accurately appraise the length of yeast cells across the different cell cycle phases. Our IFC based length analysis also allowed us to validate that total FITC intensity correlates well with cell size (Supplemental Fig. 4), confirming the usefulness of FITC staining for CFC-based size analysis.

3.7. A single platform IFC based analysis pipeline permits the extraction of all important fission yeast cell physiological parameters Having developed an IFC based morphometric and DNA content analysis pipeline, we wanted to compare the ability of IFC and CFC to track changes in multiple parameters on a single platform, overcoming the need to use a laborious combinatorial approach. To this end we arrested cultured cells in G2, released them from the block and collected samples as the population progressed synchronously through mitosis, G1, S-phase and into the subsequent G2. Fig. 8 shows that IFC was able to track nuclear division with greater accuracy in these populations than CFC, when compared to the standard approach of counting DAPI stained cells under the microscope. Furthermore the dye independent metrics for cell size in IFC and CFC presented with a clear drop in cell size when cells divide (after nuclear division), however the distribution of the size data

was far tighter when using IFC indicating a greater degree of accuracy. 4. Discussion Performing a rigorous analysis of the fission yeast cell cycle presents a number of technical challenges. Highly informative measurement such as cell length, nuclear division, cell septation and DNA content require the use of multiple analytical platforms. For example, CFC is routinely used to analyse DNA content, but in fission yeast because cell division occurs after DNA replication bi-nucleated G1 cells cannot be distinguished from mono-nucleated G2 cells. Here we corroborate previous elegant work demonstrating the use of CFC-derived DNA pulse width and area measurements to define G2/M, G1 and S phase cells [4]. We show that this approach is highly valid to the analysis of yeast cell cycle but is limited by the lack of spatial and morphometric information. Specifically, we reasoned that this approach is susceptible to miss-classifying events due to the fact that bi-nucleated cells will only give rise to an increased pulse width if aligned through the long axis as they traverse the excitation laser beam (Fig. 2). As such we have developed an IFC-based method that is able to preserve all the fundamental aspects of CFC analysis but

82


Fig. 7. IFC permits morphometric feature analysis. (A) PI spot counting allows for mathematically independent verification of IFC 2D plot PI based cell cycle assignment. (B) Image based feature analysis of the FITC signal within the bi-nucleated population allows for the assignment of a ‘‘clear septum’’ population. (C) Yeast cell length at division can be measured by analysing cell size within the ‘‘bi-nucleated’’ or ‘‘septated’’ populations. Cell length is generated using the major axis feature of the bright-field mask.

importantly provide the necessary image data to achieve two key advantages over CFC-based approaches. Firstly, the imagery allows for increased accuracy and resolution in the assignment of cells to specific cell cycle stages, allowing us to identify cells undergoing anaphase. We have formally demonstrated that there is a degree of rotational-based error associated with CFC by flow-sorting CFC pulse width/area defined cell cycle populations followed by reacquisition with CFC and IFC under identical conditions. If the yeast cell rotates around the y axis only, then one could argue that the CFC and IFC should still be able to identify bi-nucleated cells. If however there is rotation along the x and/or z axes, then CFC will fail to measure bi-nucleates accurately based on pulse width. By contrast, IFC would still be affected by z-axis rotation if these cells were not eliminated using image based approaches. Any rotation around the x axis would not be an issue at all for an image-based approach. The only explanation for the presence of contaminating populations by CFC re-analysis that were not present by our IFC-based method is the rotation of the cells as they traversed the excitation laser beam (Fig. 2). Specifically the G1/M sorted population was almost 100% pure when re-analysed by IFC but contained a significant proportion of cells classified by CFC as G2/M. Similarly, the sorted S-phase and miscellaneous populations where in majority classified as S-phase cells by IFC but still contained the so-called miscellaneous cells when analysed by CFC. We also

tested whether CFC would be able to define bi-nucleated events in populations of small cell yeast mutants where DNA masses physically cannot be separated over the same distance as in wild type. We show that CFC fails to identify bi-nucleates in small cells whereas IFC is still able to provide excellent resolution of two DNA masses in a bi-nucleated cell. As such it appears this imaging based approach may able to overcome the limitations of analysing cells of non wild type morphology by CFC. Secondly we can extend our analysis of fission yeast to look at morphological changes in combination with cell cycle. Analysing nuclear division, cell septation, DNA content and cell length are all parameters that are routinely used in the fission yeast community to study aspects of cell cycle biology. As such the ability to rapidly acquire population level data in a single assay should prove to be a useful tool for the community. A good example of this is the high-quality performance of our IFC based method in measuring DNA content and cell size in a synchronised culture. Another useful metric that has come from our IFC-based approach is the evaluation of FITC staining in CFC and IFC. The use of IFC allows us to formally prove that FITC intensity does correlate with actual cell size and can be used for this purpose in combination with CFC. FSC has previously been used to approximate cell size [19] however we found that FSC-A was anti-correlated with FITC in our experiments using a BD LSRFortessa system but


83

Fig. 8. IFC analysis of a synchronous culture is more accurate than CFC analysis. DAPI staining of fixed cells and microscopy was used to count bi-nucleation index. IFC estimation of bi-nucleate number during the time course is closer to the values from microscopy than that of CFC. Error bars on cell size measurements show standard deviation. Cell length measurement in IFC is the major axis feature of a bright-field mask in IDEAS. Cell length measurement in CFC is SCC-A. Representative images from IFC are shown in the lower panel.

positively correlated using a BD Aria Fusion. This highlights serious concerns with using FSC-A to infer object size as it is likely machine-dependent. As such we believe that FITC intensity or possibly SSC-A are more reliable metrics for cell size compared to forward scattered light [15]. The ability of IFC to acquire intensiometric and morphometric parameters for bright field, multiple dyes or florescent proteins means a vast array of other parameters can be quantified and analysed in the context of the basic morphological and cell cycle parameters we have described here. The ability to acquire such multi-dimensional data on such a large scale will also lend itself well to computational image analysis platforms. We have provided a framework for analysing fission yeast using IFC and hope this can act as a springboard for the adoption of IFC to address a range of cell biological questions as well as specifically cell cycle biology, in fission yeast. Finally, however we would like to acknowledge the fact that access to IFC systems may not always be possible therefore we strongly advocate the use of CFC-derived pulse parameters to analyse the fission yeast cell cycle under these circumstances.

Acknowledgements MS, JOP are funded by Cancer Research UK (CRUK) and the Wellcome Trust. JOP is also funded by a Boehringer Ingelheim Fonds PhD fellowship. AF is funded by CRUK and also acknowledges support from the ISAC SRL emerging leaders programme.

Please contact the corresponding author for access to the IFC-analysis template. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ymeth.2015.04. 026. References [1] C.H. Golias, A. Charalabopoulos, K. Charalabopoulos, Int. J. Clin. Pract. 58 (2004) 1134–1141. [2] P. Nurse, Nature 256 (1975) 547–551. [3] P. Nurse, Nature 344 (1990) 503–508. [4] J.H. Knutsen, I.D. Rein, C. Rothe, T. Stokke, B. Grallert, E. Boye, PLoS ONE 6 (2011) e17175. [5] D.A. Basiji, W.E. Ortyn, L. Liang, V. Venkatachalam, P. Morrissey, Clin. Lab. Med. 27 (2007) 653–670. viii. [6] A. Filby, E. Perucha, H. Summers, P. Rees, P. Chana, S. Heck, G.M. Lord, D. Davies, Cytometry A 79 (2011) 496–506. [7] E.D. Hawkins, J. Oliaro, A. Kallies, G.T. Belz, A. Filby, T. Hogan, N. Haynes, K.M. Ramsbottom, V. Van Ham, T. Kinwell, B. Seddon, D. Davies, D. Tarlinton, A.M. Lew, P.O. Humbert, S.M. Russell, Nat. Commun. 4 (2013) 1801. [8] O. Thaunat, A.G. Granja, P. Barral, A. Filby, B. Montaner, L. Collinson, N. Martinez-Martin, N.E. Harwood, A. Bruckbauer, F.D. Batista, Science 335 (2012) 475–479. [9] S. Moreno, A. Klar, P. Nurse, Methods Enzymol. 194 (1991) 795–823. [10] Y. Aoi, S.A. Kawashima, V. Simanis, M. Yamamoto, M. Sato, Open Biol. 4 (2014). [11] P. Nurse, P. Thuriaux, Genetics 96 (1980) 627–637. [12] A. Filby, D. Davies, Cytometry A 81 (2012) 637–642. [13] W.E. Ortyn, B.E. Hall, T.C. George, K. Frost, D.A. Basiji, D.J. Perry, C.A. Zimmerman, D. Coder, P.J. Morrissey, Cytometry A 69 (2006) 852–862. [14] S.A. Sabatinos, S.L. Forsburg, Methods Mol. Biol. 521 (2009) 449–461.

84


[15] A. Tzur, J.K. Moore, P. Jorgensen, H.M. Shapiro, M.W. Kirschner, PLoS ONE 6 (2011) e16053. [16] J. Begum, W. Day, C. Henderson, S. Purewal, J. Cerveira, H. Summers, P. Rees, D. Davies, A. Filby, Cytometry A 83 (2013) 1085–1095.

[17] P.F. Mullaney, M.A. Van Dilla, J.R. Coulter, P.N. Dean, Rev. Sci. Instrum. 40 (1969) 1029–1032. [18] E. Wood, P. Nurse, Cell Cycle 12 (2013) 3228–3236. [19] F.J. Navarro, P. Nurse, Genome Biol. 13 (2012) R36.