ABSTRACT
Biological systems are increasingly viewed through a quantitative lens that demands accurate measures of gene expression and local protein concentrations. CRISPR/Cas9 gene tagging has enabled increased use of fluorescence to monitor proteins at or near endogenous levels under native regulatory control. However, owing to typically lower expression levels, experiments using endogenously tagged genes run into limits imposed by autofluorescence (AF). AF is often a particular challenge in wavelengths occupied by commonly used fluorescent proteins (GFP, mNeonGreen). Stimulated by our work in C. elegans, we describe and validate Spectral Autofluorescence Image Correction By Regression (SAIBR), a simple platform-independent protocol and FIJI plug-in to correct for autofluorescence using standard filter sets and illumination conditions. Validated for use in C. elegans embryos, starfish oocytes and fission yeast, SAIBR is ideal for samples with a single dominant AF source; it achieves accurate quantitation of fluorophore signal, and enables reliable detection and quantification of even weakly expressed proteins. Thus, SAIBR provides a highly accessible low-barrier way to incorporate AF correction as standard for researchers working on a broad variety of cell and developmental systems.
INTRODUCTION
Owing to its highly reproducible development and simple geometry, C. elegans has emerged as an ideal system for quantitative analysis of symmetry-breaking (Gross et al., 2019; Lang and Munro, 2017), cell division (Pintard and Bowerman, 2019), cell and tissue mechanics (Zhang et al., 2010), and cellular decision making (Barkoulas et al., 2013). Furthermore, there is a wealth of endogenously tagged genes of interest allowing live quantitative imaging of protein networks operating at native expression levels (Dickinson and Goldstein, 2016). However, despite being transparent, C. elegans exhibits significant intrinsic autofluorescence (AF) produced by a variety of cellular constituents, and AF can be observed across all stages of worm development (Croce and Bottiroli, 2014; Pincus et al., 2016). It is most prominent when using blue and ultraviolet excitation wavelengths, and thus poses problems when using standard GFP illumination conditions (Heppert et al., 2016). Consequently, there is a need for efficient and easily implemented methods for AF correction.
A number of general strategies have been sought to correct for AF. One approach is to experimentally suppress AF. One can use chemical compounds to specifically reduce or quench AF background, or even pre-bleach samples before fluorophore addition, although these methods tend to be restricted to fixed samples (Billinton and Knight, 2001; Cowen et al., 1985; Neumann and Gabel, 2002). In C. elegans, glo mutants exhibit reduced formation of autofluorescent gut granules, although their abnormal physiology may complicate analysis (Hermann et al., 2005).
Another approach has been to optimize the combination of fluorophores, excitation light sources and emission filters to maximize the separation between fluorophore signal and AF (Billinton and Knight, 2001). These strategies usually require specialized imaging setups, and narrow emission bands can often limit the signal being captured. Such approaches may also restrict the choice of fluorophores. In C. elegans, the overlap of the excitation and emission spectra of AF with commonly used green fluorescence proteins, such as GFP or mNeonGreen (mNG), makes this approach difficult to achieve in practice. Nevertheless, some success has been achieved with specialized filter sets (An and Blackwell, 2003; Teuscher and Ewald, 2018) or yellow-shifted excitation, which is compatible with mNG, but avoids the AF excitation peak (Heppert et al., 2016). One can also avoid the AF excitation peak in C. elegans by using red-shifted RFPs such as mCherry and mKate2. However, compared with GFP and mNG, red fluorophores are less optimal due to reduced quantum yield, lower brightness and enhanced photobleaching (Heppert et al., 2016; Shaner et al., 2004; Shcherbo et al., 2009).
Techniques such as fluorescence lifetime and spectral imaging can resolve overlapping fluorescent signals based on their distinct fluorescent lifetimes or spectral characteristics (Billinton and Knight, 2001; Kumar et al., 2009; Mansfield et al., 2005). AF typically exhibits fluorescence characteristics that are distinct from other fluorophores and, therefore, it can often be separated out much as one would an additional fluorophore. Such approaches have been useful in compensating for the high levels of gut autofluorescence in C. elegans (Shi and Grant, 2015). However, such techniques require specialized instruments and analytical tools, to which some may not have access or which may be incompatible with particular experimental workflows.
An alternative, but related approach to spectral unmixing is so-called ‘AF subtraction’ (Alberti et al., 1987; Billinton and Knight, 2001; Steinkamp and Stewart, 1986). First proposed for flow cytometry, this rather simple methodology takes advantage of the fact that AF typically exhibits much broader emission spectra than fluorophores such as GFP. One can therefore quantify AF in a given sample using an AF-reporting channel and, with the appropriate calibration, subtract it on a pixel-by-pixel basis from the signal measured in the fluorophore channel (Pang et al., 2013; Roederer, 2002; Van de Lest et al., 1995). The advantages of this approach are that it is relatively straightforward to implement, it uses commonly available light sources and excitation/emission filters, and it does not require significant prior knowledge about fluorescence spectra beyond identifying a channel that is relatively specific for AF. Variations of this basic technique have since been applied in a variety of contexts, including cell-based monitoring of gene expression and AF correction of fluorescence in situ hybridization samples (Chen et al., 2019; Davis et al., 2010; Lichten et al., 2014; Roederer and Murphy, 1986; Szöllösi et al., 1995).
Here, we demonstrate that AF subtraction is a powerful method for autofluorescence correction during fluorescence imaging of C. elegans embryos. Notably, our implementation achieves results on par with more specialized imaging modalities, correcting both for bulk whole embryo fluorescence as well as for spatial AF variation. It enables reliable optimization of fluorescence signal from even very weakly expressed endogenous GFP fusions, and it is compatible with dual labeled GFP/mCherry samples, bringing substantial improvements to fluorescence signal quantification. At the same time, owing to its use of standard GFP/RFP filters sets and its implementation via an easy-to-use Fiji plug-in, this protocol, which we term Spectral Autofluorescence Image correction By Regression (SAIBR), is readily combined with a variety of imaging platforms and procedures, allowing integration of AF correction as a standard part of imaging workflows. Although developed with C. elegans embryos in mind, the principles behind SAIBR are general and thus should be suitable for a variety of samples and fluorophores. Consistent with this, our SAIBR plug-in is readily applicable to a variety of other experimental systems, and thus should be a useful tool for AF correction for the cell and developmental biology community.
RESULTS
To quantify the potential impact of AF on the specific detection of GFP in C. elegans embryos, we began by comparing the magnitude of AF signal obtained from unlabeled embryos with the signal obtained from embryos expressing GFP fusion proteins when imaged with standard GFP illumination settings (ex488/em535/50, hereafter GFP channel). For this purpose, we selected C. elegans lines expressing GFP fusions to one of three polarity proteins: PAR-6 and PAR-3, which localize to an anterior plasma membrane domain; and LGL-1, which localizes to the posterior plasma membrane. All genes were tagged at the endogenous loci. In these cases, AF accounted for ∼40-90% of the observed signal in the GFP emission band (Fig. 1A). Moreover, when we imaged unlabeled embryos, the AF signal in the GFP channel was intrinsically variable. Not only was there substantial spatial variation in AF (Fig. 1B), but also a nearly twofold variation in the overall magnitude of AF signal between embryos (Fig. 1A,B, embryo i versus iii). Thus, simply subtracting out mean AF signal obtained from unlabeled embryos will fail to account for both sources of variation. In the case of LGL-1::GFP, such mean AF subtraction would clearly lead to apparent negative concentrations in some embryos. Thus, if we wish to accurately quantify the expression and local subcellular concentrations of proteins using GFP fusions, particularly for genes exhibiting low-to-moderate expression, one requires a method for locally correcting AF on a per embryo, per pixel basis.
Implementing a simplified method for AF correction based on dual emission imaging
In principle, the emission signal for GFP and AF will be additive; thus, if one has an independent measure of AF, one can subtract AF from the combined signal to obtain a value for GFP emission. Indeed, if we subtract the fluorescence emission spectrum of autofluorescence measured in unlabeled embryos from the spectrum obtained from embryos expressing PAR-6::GFP, we almost perfectly recover the theoretical spectrum for GFP (Fig. 1C). The challenge is therefore to find a method of quantifying AF directly in embryos also expressing GFP.
One strategy for quantifying AF takes advantage of the distinct spectral properties of AF that allow it to be quantified in an AF-reporting channel distinct from that used for measuring GFP, hereafter AF or ‘predictor’ channel (Alberti et al., 1987; Roederer and Murphy, 1986). One can use AF channel measurements to predict and thus correct for AF in the ‘primary’ GFP channel (Fig. 1D).
In C. elegans embryos, AF peaks in the green-to-yellow wavelengths overlapping GFP, but extending further into the red (Fig. 1C) (Heppert et al., 2016; Pincus et al., 2016). It can therefore be captured on a relatively selective basis through the use of a suitably red-shifted emission filter, such as those typically used for red fluorescent proteins. We therefore specify the AF Channel as ex488/em630/75. In practice, there will be a slight spillover of GFP signal into the AF channel, which could lead to overestimation of AF; however, because spillover is necessarily proportional to GFP amounts, it can be easily accounted for (see Materials and Methods). To establish an AF correction function between channels, we performed a linear regression on fluorescence pixel values obtained from images of unlabeled embryos captured in both the GFP and AF channels, where all signal is attributable to AF. We obtained strong linear correlations (R2>0.8) that were similar between embryos (Fig. 1E). A nearly identical correlation was observed when we plotted the mean intensity values of entire individual embryos (R2=0.952, Fig. 1F), indicating that the same correction function can account for both intra- and inter-embryo AF variation. Thus, we can use AF measurement in the AF channel to accurately infer and subtract AF signal from the GFP channel to obtain an accurate measure of ‘true GFP’ signal (Fig. 1D).
As proof of principle, we captured images of embryos expressing a GFP fusion to CDC-25.3 from the endogenous locus. CDC-25.3 expression is repressed until early embryogenesis, reportedly becoming visible as embryos progress beyond the eight-cell stage (Tsukamoto et al., 2017). Using our AF correction method, we were able to observe clear nuclear localization already at the start of the four-cell stage and accurately track its accumulation and release at NEBD at a time at which AF almost completely masked its expression in uncorrected images (Fig. 1G). We designated this protocol Spectral Autofluorescence Image Correction By Regression (SAIBR). A schematic workflow is provided in Fig. 2, with additional details in the Materials and Methods.
AF correction using SAIBR
We next undertook a detailed quantitative analysis of the effectiveness of SAIBR in both unlabeled and GFP-labeled embryos, using GFP as the primary and AF as the predictor channels. Applying SAIBR to unlabeled embryos effectively reduced observed embryo fluorescence in the GFP channel to background, suggesting we accounted for nearly all AF signal in zygotes (Fig. 3A). We then applied SAIBR to embryos expressing GFP fusions to LGL-1, PAR-3 or PAR-6 from the respective endogenous loci (Fig. 3B-D). For both LGL-1::GFP and PAR-3::GFP, SAIBR revealed a clear peak in signal at the posterior and anterior plasma membranes, respectively, that was obscured by AF in uncorrected images (Fig. 3B,C). Even when averaging cross-sectional membrane profiles across multiple embryos, membrane signal was difficult to discern in uncorrected data (Fig. 3E,F, top). By contrast, SAIBR resolved membrane signal into a clear, well-defined peak (Fig. 3E,F, bottom). Cytoplasmic signal also became substantially more uniform, which was most clearly visible in the suppression of a local fluorescence minimum in the embryo center due to AF exclusion by the pronuclei and mitotic spindle region. Improvements are also visible for PAR-6::GFP-expressing embryos, although the magnitude of improvement is less striking due to the higher ratio of GFP to AF signal (Fig. 3D,G). Similar results were achieved using both spinning disk confocal (Fig. 3) and wide-field microscopy (Fig. S1), confirming that the method is platform independent.
We next turned to quantification of total protein concentrations in embryos and compared SAIBR with a mean AF subtraction protocol (mean AF subtraction). For mean AF subtraction, we establish a mean AF signal in the GFP channel based on fluorescence signal measured across multiple unlabeled embryos not expressing GFP and simply subtracted this value from the GFP channel signal in GFP-expressing embryos. As a test, we used C. elegans lines expressing GFP::PAR proteins described above as well as embryos that are heterozygous for the par-6::gfp fusion and thus only half of the PAR-6 pool is labeled. For PAR-6::GFP, both SAIBR and mean AF subtraction yield the expected 2:1 ratio of GFP signal between homozygous and heterozygous embryos (Fig. 3H). However, by correcting for embryo-to-embryo AF variation, SAIBR substantially reduced the coefficient of variation (c.o.v.). We obtained similar results for PAR-3 and LGL-1 (Fig. 3H). For LGL-1, in which the GFP signal was of the same order as variation in AF, the advantage of SAIBR was particularly striking. Whereas correction by mean AF subtraction resulted in negative values for GFP in embryos, the ability of SAIBR to suppress the effects of embryo-to-embryo variation in AF allowed it to achieve consistent and positive values for GFP signal in all embryos.
Benchmarking against alternative strategies
To benchmark our method with other approaches, we used an alternative AF-minimization strategy in which we use a fluorophore compatible with wavelengths that minimize AF excitation. mNG behaves similarly to GFP under standard GFP illumination settings (488 nm). However, owing to a slight shift in its excitation spectrum, unlike GFP, it can be efficiently excited by a yellow-shifted laser line (514 nm) to substantially reduce AF (hereafter, mNG Channel, ex514/em550/50) (Heppert et al., 2016). Consistent with this observation, the magnitude of AF signal, as measured in unlabeled embryos relative to total signal for embryos expressing mNG::PAR-3, is reduced substantially in the mNG channel relative to the GFP channel (compare Fig. 4A with 4B). We next compared the effectiveness of AF correction in three regimes: a standard regime using the GFP channel and mean AF subtraction, as described in the previous section; an mNG-specific regime using the mNG channel and mean AF subtraction; and a regime using the GFP channel but corrected by SAIBR. By plotting normalized corrected signal, we found that using either the mNG channel or SAIBR regimes showed similar and substantial improvement in the variance of mean embryo fluorescence (Fig. 4C), suppression of spatially varying cytoplasmic AF (Fig. 4D) and accurate quantification of membrane signal (Fig. 4E).
Using LGL-1::GFP as a case study, we also compared SAIBR with several AF suppression tools available as ‘off-the-shelf’ tools on the Zeiss 880, including using an optimized emission band (499-508 nm) and spectral unmixing. Spectral unmixing uses a built-in automatic component extraction algorithm, which can be performed in both reference-free (‘blind’) and reference-calibrated (‘calibrated’) modes. All three methods substantially reduced AF compared with images captured using typical broadband GFP emission band (499-562 nm), which served as a reference (Fig. 4F-I). Unsurprisingly, of the three, calibrated spectral unmixing showed the best performance when judged by either membrane-to-cytoplasm ratios in LGL-1::GFP-expressing animals or by their ability to reduce AF in unlabeled embryos (Fig. 4H,I). However, SAIBR generally outperformed all three methods for our samples, reducing AF effectively to zero in unlabeled embryos and achieving the highest membrane:cytoplasm ratios for LGL-1::GFP. Results were similar for both narrowband and broadband emission, despite the increased AF present in the latter images. Thus, SAIBR is highly competitive with other state-of-the-art AF compensation techniques, providing comparable and, in some cases, better improvement in image quality and signal quantitation without the need for specialized imaging modalities typically required for other techniques.
Extension of SAIBR to dual-labeled samples
Because SAIBR requires measurement of AF in a red-shifted emission channel, the use of dual fluorophore pairs, such as GFP/mNG together with RFP/mCherry/mKate, can introduce complications. Specifically, because red FPs (RFPs) are weakly excited by typical wavelengths used for GFP excitation, they will contribute to apparent AF, which is captured in the AF channel. As long as this contribution is low, i.e. for weak-to-moderate expression levels, it can be safely ignored (e.g. TH209, PAR-2::mCherry, Fig. 5A). However, at higher expression levels, the contribution of RFP signal to the AF channel becomes significant (e.g. NWG0033, mCherry::MEX-5, Fig. 5A). At such levels, this bleedthrough signal induced a deviation in the mapping of AF between the AF and GFP channels compared with N2 in proportion to the degree of RFP expression (Fig. 5B) and leads to overcorrection for AF in the GFP channel if not properly accounted for (e.g. Fig. 5F, left, two-channel SAIBR). In principle, this bleedthrough signal of RFP in the AF channel is proportional to the concentration of the RFP fusion and therefore is proportional to RFP signal captured using standard RFP illumination (RFP channel, ex561/em630/75). We can therefore compensate for bleedthrough by using inputs from both the AF channel and the RFP channel to establish a revised three-channel SAIBR correction function (primary, GFP; predictor 1, AF; predictor 2, RFP). To this end, we captured images of embryos expressing only RFP (no GFP) in the GFP, AF and RFP channels, and performed a multiple linear regression in which AF in the GFP channel is a function of signal in both the AF and RFP channels (Figs 2, 5C,D). Applying three-channel SAIBR to embryos expressing PAR-6::GFP with MEX-5::mCherry eliminates oversubtraction and yielded measurements that were nearly identical to those for embryos expressing PAR-6::GFP alone with no increase in data scatter, validating the expanded use of SAIBR to dual-labeled samples (Fig. 5E,F). As a further test of its use in dual-labeled embryos, we used three-Channel SAIBR to follow the relative localization of endogenously tagged LGL-1::GFP and PAR-6::mCherry in time-lapse recordings of early embryos, which previously relied on multi-copy overexpression of LGL-1::GFP (Beatty et al., 2010; Hoege et al., 2010) (Movie 1).
Eggshell fluorescence is another issue in C. elegans embryos that sometimes arises and can potentially complicate SAIBR. Eggshell fluorescence is usually relatively minor in the GFP and AF channels, and can often be ignored. However, eggshell fluorescence is variable and may occasionally be significant for some methods of sample preparation and/or mounting (see Materials and Methods). Eggshell fluorescence has a distinct spectral profile compared with the AF we have discussed so far. Indeed, in many respects it behaves similarly to an RFP, although with a broader emission spectrum and hence a stronger bleedthrough into the AF channel. Pronounced eggshell fluorescence is particularly problematic when quantifying fluorophore signal at the plasma membrane because eggshell fluorescence will result in oversubtraction of AF in regions where the membrane and eggshell are in contact (Fig. 5G). However, similar to an RFP, it could be compensated for by treating it as a red fluorophore and applying three-channel SAIBR (Fig. 5G). It is important to note that because the emission spectrum of the eggshell is distinct from RFPs, we could not simultaneously correct for both eggshell and RFP in the same samples.
SAIBR in late C. elegans embryos and larvae
To expand the applicability of this method, we extended our analysis to late embryos (post-gastrulation) and larval stages. AF is known to be increasingly problematic in intestinal cells as C. elegans development proceeds and autofluorescent gut granules are formed (Laufer et al., 1980). In unlabeled 1.5-fold stage embryos, the cells of the embryo posterior were visibly brighter due to AF signal in this region (Fig. 6A). This signal was largely eliminated by SAIBR (Fig. 6A). We next examined the localization of LGL-1::GFP (Fig. 6B,C). Some membrane staining of LGL was visible on the plasma membrane in uncorrected images but was often obscured by AF. Thus, the locations of cells in many areas were only visible due to the reduced AF signal in the nucleus, particularly in the developing intestine (arrows). With SAIBR, the membrane localization of LGL-1::GFP was much more clearly resolved and the cytoplasmic signal significantly more uniform. Notably, basolateral membrane localization of LGL-1 in intestinal cells could be clearly seen, juxtaposed to apical PAR-6::mCherry signal (arrowheads) as observed previously in overexpression lines (Fig. 6C) (Beatty et al., 2010; Sallee et al., 2021).
In L1 larva, gut granules are particularly prominent in intestinal cells in addition to the more-diffuse AF signal characteristic of earlier embryos. Hence, we were curious about how SAIBR would perform (Fig. 6D). When applying SAIBR, we found that a substantial fraction of granule fluorescence was over-subtracted, indicating that the fluorescence profile of mature gut granules differs from the more-diffuse AF signal in the cytoplasm (Fig. 6D, top). This emphasizes the problems associated with the presence of multiple, independently varying sources of AF. Nonetheless, despite modest over-subtraction of some granules, SAIBR significantly improved visualization of LGL-1::-GFP in these tissues (Fig. 6D, bottom), revealing the expected basolateral pattern of localization in intestinal cells (Castiglioni et al., 2020). Thus, despite some over-subtraction of gut granule signal, our method can still improve visualization of weakly expressed fluorophores in both late embryo and larval stages.
A GUI-based FIJI SAIBR plug-in enables simple AF correction in diverse systems
Although the calibration and correction steps involved in SAIBR are relatively straightforward, we recognize that the need to implement such a workflow may limit widespread adoption. Therefore, to facilitate its use, we have implemented SAIBR as a simple graphical user interface (GUI)-based Fiji plug-in that allows output of AF-corrected images in a few easy steps (summarized in Fig. 2). A detailed description of the plug-in along with full instructions can be found together with sample datasets at https://github.com/goehringlab/saibr_fiji_plugin.
With the SAIBR plug-in in hand, we solicited samples from a variety of experimental systems to validate its general suitability. To this end, we obtained suitable sets of fluorescence images for two systems that exhibit autofluorescence. In starfish oocytes, bright autofluorescent cortical granules dominate the signal in the GFP channel, in this case in comparison with a relatively dim signal for the mother centriole (Fig. 7A). The SAIBR plug-in significantly suppressed the AF signal of granules, typically leaving the centriole clearly visible relative to the residual AF signal (Fig. 7A,B). We observed a similar reduction in AF originating from vacuoles in the fission yeast S. pombe, here shown relative to a mNG fusion to the ER/nuclear envelope-localized phosphatase component Nem1 (Fig. 7C). These results confirm the potential broad applicability of SAIBR for AF compensation in cell and developmental systems.
DISCUSSION
Here, we describe and validate a simple and easy-to-use implementation for autofluorescence correction, SAIBR, provided as a Fiji plug-in. A simplified form of spectral imaging, we leverage the typically broad AF spectrum to allow accurate estimation and correction of AF signal in the GFP channel. SAIBR is platform independent and relies only on commonly available GFP/RFP illumination sources and filter sets. Yet SAIBR yields similar performance to more specialized methods in both single and dual fluorophore-labeled samples, enabling the visualization and accurate quantification of even weakly expressed proteins that are at the limits of detection above AF.
Our aim was to provide a tool that would enable regular and widespread use of AF correction by the research community as part of day-to-day investigation. Autofluorescence is a particularly common problem when imaging GFP in a number of systems with AF sources, including yolk and cortical granules, extracellular matrix, and lysosomal compartments. Although more complex solutions exist, as we have shown, SAIBR provides a simple and straightforward solution that is likely to work efficiently in many contexts. The ease of use of the SAIBR plug-in and the generic imaging conditions required mean that it costs users little to test on their system of choice, allowing one to quickly determine whether a more-complex approach is required.
The simplicity and platform independence of SAIBR allow it to be integrated into a variety of experimental workflows with minimal extra investment of time and resources. We envision that SAIBR and its potential to drive widespread adoption of routine AF correction should enable new experimental approaches. In the C. elegans embryo, for example, by allowing accurate AF correction on a per embryo basis, not only will this method provide improved measurement of protein concentration and subcellular distribution within cells, but it will also allow us to address questions related to protein dose, such as assessing variability of protein expression and its potential effects on developmental pathways.
In principle, there is no need to restrict oneself to the wavelengths used here, which were chosen to solve the problem of GFP channel autofluorescence. The technique itself requires only the identification of an appropriate ‘predictor’ channel or channels that can be used to infer AF in the desired fluorophore-reporting ‘primary’ channel, and hence should be well separated from the channel used for fluorophore imaging.
At the same time, simplicity and flexibility come with certain trade-offs. First, SAIBR requires identification of an AF predictor channel that is reasonably well isolated from the primary reporter channel for the fluorophore of interest. Second, it relies on the existence of an AF signal that exhibits a consistent correlation between predictor AF and primary fluorophore channels, and hence is not suited to samples with multiple independently varying sources of AF. Third, SAIBR combines pixel noise from multiple channels, effectively amplifying salt-and-pepper noise in images, although one could, in principle, apply computational denoising strategies to reduce this effect (Krull et al., 2019). Finally, as implemented here, SAIBR requires one to capture images in at least two emission channels, effectively doubling sample illumination and minimum time intervals. The time lag between frames may also lead to pixel mismatches between primary and predictor channels for samples exhibiting rapid motion. However, this last limitation can be bypassed with suitable optics that allow simultaneous capture of multiple emission bands.
There are a variety of approaches to correct for autofluorescence and SAIBR will not and is not intended to replace them, as each has its own advantages and disadvantages. Fluorescence lifetime imaging and spectral unmixing are likely to be substantially better for samples with multiple complex AF signals, but often come with the need for specialized imaging platforms (Zimmermann et al., 2003). Spectral unmixing approaches also require a reference spectrum for each fluorophore to achieve optimal performance. This may not be possible for all samples, although there has been some progress in addressing this problem by ‘blind’ unmixing algorithms (McRae et al., 2019). Another approach is to use feature-based algorithms that seek to define or ‘learn’ characteristics of autofluorescent objects to identify and remove them (Baharlou et al., 2021). However, these are unlikely to deal well with AF, which is not characterized by distinct object features or which spatially overlaps with fluorophore signal, both of which are the case for C. elegans embryos. Ultimately, the best choice is likely to be sample dependent, requiring one to quantitatively assess various options as we have here. Minimizing barriers to adoption and testing is key, which is why we have provided our method as a simple, fully open-source and easy-to-use plug-in. In summary, by combining ease of implementation and accurate AF correction with relatively few trade-offs, we hope SAIBR will help facilitate widespread adoption of autofluorescence correction and enable more accurate quantification of the concentration and distribution of fluorescently tagged proteins in cells and tissues.
MATERIALS AND METHODS
C. elegans – strains and maintenance
C. elegans strains were maintained on OP50 bacterial lawns seeded on nematode growth media (NGM) at 20°C under standard laboratory conditions (Stiernagle, 2006). Strains are listed in Table S1. OP50 bacteria were obtained from CGC. Oocytes and zygotes were obtained from hermaphrodites unless otherwise noted. Analysis of zygotes precludes determination of animal sex.
C. elegans – strain construction
Mutation by CRISPR-Cas9 was performed based on the protocol published by Dokshin et al. (2018). Briefly, tracrRNA (IDT DNA, 0.5 µl at 100 µM) and crRNA(s) for the target (IDT DNA, 2.7 µl at 100 µM) with duplex buffer (IDT DNA, 2.8 µl) were annealed together (5 min, 95°C) and then stored at room temperature until required. PCR products containing the insert DNA sequence (GFP in this instance) and an insert with an additional 130 bp homology to the insertion site were generated, column purified (Qiagen, QIAquick PCR purification kit), mixed in equimolar amounts, denatured by heating to 95°C and annealed thorough slow cooling to room temperature to generate a pool of products with long single-stranded DNA overhangs that act as the repair template. An injection mix containing Cas9 (IDT DNA, 0.5 µl at 10 mg/ml), annealed crRNA, tracrRNA and the repair template was incubated at 37°C for 15 min and centrifuged to remove debris (15 min, 14,100 g). Young gravid N2 adults were injected along with a dpy-10 co-CRISPR injection marker (Arribere et al., 2014) and mutants identified by PCR and sequence verified. Resulting lines were backcrossed with N2s twice before use. Sequences are available in Table S1.
C. elegans – dissection and mounting for microscopy
For most experiments, early embryos were dissected from gravid hermaphrodites in 5-6 µl of M9 buffer (22 mM KH2PO4, 42 mM NaHPO4, 86 mM NaCl and 1 mM MgSO4) on a coverslip and mounted under 2% M9 agarose pads (Zipperlen et al., 2001). In some instances (Figs 1B,G, 3A,B,E and 4, and Fig. S1), to minimize eggshell autofluorescence that may be prominent with agarose mounts, embryos were dissected in 8-10 µl of egg buffer [118 mM NaCl, 48 mM KCl, 2 mM CaCl2 2 mM MgCl2 and 25 mM HEPES (pH 7.3)], and mounted with 20 µm polystyrene beads (Polysciences) between a slide and coverslip as described previously (Rodriguez et al., 2017).
To harvest late embryos (Fig. 6A-C), gravid worms were allowed to lay embryos for 4-5 h at 20°C. Embryos were collected and mounted in 8-10 µl of egg buffer supplemented with 18.8 µm polystyrene beads (Polysciences).
L1 larva (Fig. 6D) were collected from plates where gravid adult worms were allowed to lay eggs for 12-13 h at 20°C. Whole larva were then mounted between a 2% M9 agarose pad and coverslip in M9 containing 0.1 µm polystyrene beads (Polysciences) and 10 mM levamisole to induce worm paralysis (Reich et al., 2019).
C. elegans – fluorescence microscopy
Unless specified otherwise, midsection confocal images were captured on a Nikon TiE with a 60×/1.40 NA oil objective, further equipped with a custom X-Light V1 spinning disk system (CrestOptics) with 50 µm slits, Obis 488/561 fiber-coupled diode lasers (Coherent) and an Evolve Delta EMCCD camera (Photometrics). Imaging systems were run using Metamorph (Molecular Devices) and configured by Cairn Research. Filter sets were from Chroma: ZT488/561rpc, ZET405/488/561/640X, ET535/50m and ET630/75m. For late embryo and larval imaging, 1.5× magnification was applied (using TiE intermediate magnification switching).
Images were typically captured sequentially, although GFP and AF channels could alternatively be captured simultaneously with a suitable optical setup to minimize the effects of sample movement and bleaching.
Midsection wide-field fluorescence images were captured on a Nikon TiE with a 60×/1.40 NA oil objective, further equipped with a Spectra-X Light Engine (Lumencor). Imaging systems were run using Metamorph (Molecular Devices) and configured by Cairn Research. Filter sets were from Chroma: ET490/20x, ET525/50m and ET632/60m.
To obtain the emission spectra shown in Fig. 1C, embryos were imaged under 488 nm excitation at consecutive 20 nm wavebands over a range of 510-710 nm (yielding lambda stacks of 10 images per embryo). Wavelength-scans were performed on a Leica TCS SP8 inverted microscope equipped with an Apo CS2 63x/1.40 NA oil objective and a HyD detection system. Imaging was managed with LAS X software (Leica Microsystems), and acquisition was set at a scanning speed of 700 Hz with pinhole aperture set to 2 AU.
For midsection confocal imaging of mNG-expressing embryos and comparison between 488 nm and 514 nm excitation configurations (see Fig. 4), experiments were performed on a Leica SP8 microscope (as above) at the indicated emission filter settings. Acquisition was set at a scanning speed of 600 Hz and pinhole aperture was set to 3 AU.
For linear unmixing and emission band optimization, we captured midplane images on an inverted Zeiss LSM880, equipped with Plan-Apochromat 63×/1.4 Oil DIC M27 using 1.7× zoom (0.155μm/ pixel). Imaging was managed with Zen Black software, with a pixel dwell time of 4.10 μs (10.07 s scan time), line averaging of 4 and a pinhole size of 1 AU. Samples were excited at 488 nm and emission captured for ∼9 nm wavebands spanning ∼410-695 nm yielding 32 images. For calibrated unmixing, unlabeled N2 embryos and myo-2::GFP in the pharynx of adult animals were used to define the reference AF and GFP spectra, respectively. Comparative datasets for SAIBR analysis were obtained by summing intensities for the relevant wavebands (broadband: GFP, em499-562; AF, em579-668; narrowband: GFP, em499-508; AF, em588-597).
C. elegans – image processing and quantification
In some cases (Fig. 1A-F, Fig. 3, Fig. 5, Figs S2 and S3), images of embryos/samples were taken alongside a local background image (with no samples in the field of view), which was subtracted from the image before analysis. This step can usually be omitted without much detriment, as an even and consistent background signal can be factored into the calibration parameters. However, background subtraction may improve images in cases where the background signal is uneven or variable. In some cases, a median filter (diameter 1-2 px) was applied before incorporation in figures.
Whole-embryo fluorescence intensities are defined as the mean pixel intensity within a manually defined region of interest (ROI) encompassing the embryo. Individual cross-membrane profiles were extracted by taking 50-pixel line profiles perpendicular to and centered on the plasma membrane, using bicubic spline interpolation. Profiles were taken at pixel-width intervals around the circumference of the embryo, and averaged over the posterior-most (Fig. 3E) or anterior-most (Fig. 3F,G) 30% of the circumference of the embryo.
SAIBR – autofluorescence correction
In the following discussion, the GFP channel is the primary channel for which we want to apply AF correction, and the AF and RFP channels are predictor channels, which we use to predict autofluorescence in the primary channel. In principle, one can adapt the method to any suitable set of primary plus predictor channels.
Two channels
Three channels
Calculation of inter-channel correction factors
Calibration samples were taken contemporaneously with test samples and imaged under identical conditions, as changes to excitation/emission parameters will necessarily alter the parameters of the correction function.
GFP spillover correction
The long tail of the GFP emission spectrum means that a small fraction of GFP emission will appear in the AF channel. This spillover of GFP signal into the AF channel will artificially inflate AAF and therefore result in oversubtraction upon application of SAIBR. However, because the magnitude of this effect is always proportional to GFP concentration, it simply rescales the magnitude of GGFP and thus can be safely ignored for normalized data or for relative comparisons between different GFP-containing samples. Alternatively, the magnitude of this effect can be measured and a correction applied. See Fig. S3 for additional details.
Additional methods – starfish oocytes
Starfish (P. miniata, also known as A. miniata) oocyte collection and injection were performed as described previously (Borrego-Pinto et al., 2016a; Terasaki, 1994). Briefly, starfish were obtained from the Southern California Sea Urchin Company (Marinus Scientific, South Coast Bio-Marine) or the Monterey Abalone Company and maintained in seawater tanks at 16°C at the European Molecular Biology Laboratory (EMBL) Marine Facility. The mRNA encoding fluorescent mother centriolar Odf2::mEGFP (https://www.ncbi.nlm.nih.gov/nuccore/1040843242) (Borrego-Pinto et al., 2016b) was injected the day before imaging, while Cy3 tubulin (a gift from the Nédélec laboratory, EMBL, Heidelberg, Germany) was injected shortly before imaging. After meiotic maturation with 10 µM 1-methyladenine (Acros Organics), oocytes were imaged on a Leica SP5 confocal microscope, as described previously (Borrego-Pinto et al., 2016b). Sequential scanning was performed: in a first scan, 488 nm excitation was coupled to both mEGFP (primary) and red-shifted (predictor 1) emission channels to record Odf2-mEGFP and autofluorescence, respectively. In a second scan, 561 nm excitation was combined with the same red-shifted emission channel to record Cy3-tubulin fluorescence (predictor 2).
Additional methods – S. pombe
Schizosaccharomyces pombe (S. pombe) cells were grown in YES (yeast extract with supplements) medium overnight at 30°C. Prior to imaging, 1 ml S. pombe cell culture with OD595nm 0.4-0.6 was concentrated to 50 µl after centrifugation at 1500 g for 30 s. 2 µl of cell suspension were loaded under a 22×22 mm glass coverslip (VWR, thickness: 1.5). Spinning disk confocal images of S. pombe were captured with an Eclipse Ti-E inverted microscope fitted with a Yokogawa CSU-X1 spinning disk confocal scanning unit, 600 series SS 488 nm, SS 561 nm lasers, single band filters FF01-525/50-25 and FF01-617/73-25 (Semrock Brightline), Nikon CFI Plan Apo Lambda 100x (NA=1.45) oil objective and an Andor iXon Ultra U3-888-BV monochrome EMCCD camera. Image acquisition was controlled by Andor IQ3 software.
Acknowledgements
We thank Josana Rodriguez, Ruben Schmidt and Donald Bell for comments on the manuscript and/or pilot testing of the SAIBR plug-in. Strains and/or reagents were graciously provided by Dan Dickinson and Ken Kemphues. Additional strains were provided by the Caenorhabditis Genome Center (CGC), which is funded by the National Institutes of Health Office of Research Infrastructure Programs (P40 OD010440). Confocal imaging was performed with assistance of Matt Renshaw in the Crick Advanced Light Microscopy (CALM) STP. Some experiments were performed in the labs of Peter Lénárt (EMBL and MPI-BPC) and Snezhana Oliferenko (The Francis Crick Institute and King's College London).
Footnotes
Author contributions
Conceptualization: N.T.L.R., T.B., N.W.G.; Methodology: N.T.L.R., T.B.; Software: T.B.; Formal analysis: N.T.L.R., T.B., K.N.; Investigation: N.T.L.R., T.B., J.B.-P., K.N., Y.G., S.F.; Resources: N.T.L.R., T.B., N.H., S.F.; Writing - original draft: N.T.L.R., T.B., N.W.G.; Writing - review & editing: N.T.L.R., T.B., J.B.-P., K.N., N.H., Y.G., S.F., N.W.G.; Supervision: N.W.G.; Project administration: N.W.G.; Funding acquisition: N.W.G.
Funding
This work was supported by The Francis Crick Institute, which receives its core funding from Cancer Research UK (FC001086), the UK Medical Research Council (FC001086) and by the Wellcome Trust (FC001086), as well as by a Wellcome Trust Investigator Award in Science (220790/Z/20/Z) and a Biotechnology and Biological Sciences Research Council grant (BB/T000481/1) to Snezhana Oliferenko. Open Access funding provided by The Francis Crick Institute. Deposited in PMC for immediate release.
Data availability
Source code, sample datasets and documentation for the plug-in are available at https://github.com/goehringlab/saibr_fiji_plugin.
Peer review history
The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200545.
References
Competing interests
The authors declare no competing or financial interests.