High fidelity lineage tracing in mouse pre-implantation embryos using primed conversion of photoconvertible proteins

Accurate lineage reconstruction of mammalian pre-implantation development is essential for inferring the earliest cell fate decisions of mammalian development. Lineage tracing using global labeling techniques is complicated by increasing cell density and rapid embryo rotation, impeding automatic alignment and rendering accurate cell tracking of obtained four-dimensional imaging data sets highly challenging. Here, we exploit the advantageous properties of primed convertible fluorescent proteins (pr-pcFPs) to simultaneously visualize the global green and the photoconverted red population to minimize tracking uncertainties over prolonged time windows. Confined primed conversion of H2B-pr-mEosFP labeled nuclei combined with light-sheet imaging greatly facilitates segmentation, classification, and tracking of individual nuclei from the 4-cell stage up to the blastocyst. Using green and red labels as fiducial markers, we computationally correct for rotational and translational drift and accomplish high fidelity lineage tracing combined with a reduced data size – addressing majors concerns in the field of volumetric embryo imaging.


Introduction 29
Accurate lineage tracing and precise tracking of single cells in pre-implantation embryos 30 is essential for a mechanistic understanding of the first cell fate decisions during mammalian 31 development 1 . Selective plane illumination microscopy (SPIM) has the potential to play a major 32 role in achieving comprehensive, non-invasive imaging of mammalian pre-implantation 33 development. During these early steps of development, a major fraction of embryos (n=5/11, 34 45% in this study) exhibit confounding rotational and translational drift (Videos 1 and 2), which 35 often leads researchers to exclude these embryos from analysis, drastically decreasing efficiency, 36 losing valuable data, and potentially biasing downstream results 2,3 . While high imaging rates 37 have helped to overcome these challenges for samples like zebrafish embryos, they demand 38 increased data storage capacities. Higher framerates can further increase photodamage from laser 39 overexposure and are hence less applicable for highly sensitive mouse embryos 2,4 . 40 Sparse labeling strategies using green-to-red photoconvertible fluorescent proteins 41 (pcFPs) merit a great potential for facilitating lineage tracing and trophectoderm (TE) and inner-42 cell-mass (ICM) fate assignments after photoconversion 5 . However, to our knowledge these 43 sparse labels have not been combined with SPIM -presumably because photoconversion has 44 been limited by the need for axially unconfined, potentially photodamaging, intense violet light 6 . 45 Our recent report of a novel photochemical mechanism called "primed conversion" overcomes 46 this long-standing problem by using dual-wavelength illumination with blue 488nm, and far-red 47 as the optimal fluorescent protein variant for in vivo primed conversion followed by long-term 74 imaging. 75 Next, we investigated whether a single round of green-to-red photoconversion at the 4-76 cell stage would create a sufficiently large pool of red-converted protein that could be followed 77 throughout development until the early blastocyst stage. For this purpose, we used confined 78 primed conversion to photoconvert a single nucleus of an H2B-pr-mEosFP expressing embryo at 79 the 4-cell stage 8 , and monitored early embryo development for 60 hours in an inverted SPIM 2 80 realignment with all algorithmic corrections, and iii) after final manual review by calculating the 114 total distance between these lineage trees (see methods for more details) 11 . Notably, the resulting 115 lineage trees required a minimal amount of time for manual corrections (i.e. 1-1.5 hours for the 116 total lineage tree). 117

Summary and conclusion
In summary, the presented approach enables fast, automated, high fidelity lineage tracing 119 of mammalian pre-implantation development combined with reduced illumination time and data 120 volume, key considerations for handling and analyzing data by the biological community 12 . In 121 addition, the ability to correct for both spatial and rotational drift overcomes the need to exclude 122 spinning embryos from the analysis. On a different note, it might enable the experimenter to 123 achieve similar tracing quality with datasets acquired at lower sampling rate. In the future, 124 implementing primed conversion to take place inside the SPIM used for volumetric imaging, will 125 allow for repeated manual or automatic primed conversion of nuclei once the red fluorescence 126 drops below a user-defined threshold. Similar pulse-chase experiments can then be extended 127 even longer, ultimately being only limited by the rate of new green pr-pcFP synthesis. The 128 combination of confined primed conversion of pr-pcFPs with our imaging pipeline will allow 129 researchers to get more accurate insight into the dynamic processes responsible for cell fate 130 decisions in the early mammalian embryo. 131 132

mRNA microinjection of mouse preimplantation embryos and ex utero culture up to 4-cell 142
stage 143 C57/Bl6 wild-type females were superovulated by hormone priming, mated to C57/Bl6 144 males, and mated females were euthanized by CO 2 asphyxiation. Embryos were recovered by 145 flushing oviducts as described previously 8,13 . Embryos were cultured at 37°C and 5% CO2 in 146 KSOM+AA medium covered with mineral oil. mRNA constructs were microinjected into the 147 pro-nucleus at 50 ng/ul or in both cells in two-cell stage embryos, following standard protocols. 148 All these experiments were approved by the veterinary authority of the canton Basel Stadt, 149 Swizerland. 150

Confined primed conversion of single nuclei in mouse embryos 151
Confined primed conversion of single nuclei was performed on mouse embryos at the 4-152 cell stage as previously described in great detail 8 . 153 Volumetric imaging of mouse pre-implantation embryos 154 Right after confined primed conversion was performed, the 4-cell stage embryos were 155 transferred to a pre-equilibrated, inverted SPIM setup and continuously cultured/imaged until 156 they reached blastocyst stage 2 . For each embryo a z-stack consisting of 80 planes, 3 µm apart, 157 was acquired every 7.5 or 15 minutes. 158

Mouse embryo lineage tracing 159
To establish a reference, mouse embryos were lineage traced using the state-of-the-art 160  4. Discard all spurious red spots that do not colocalize with a green spot. Note that due to 181 the equilibrium between protonated and de-protonated chromophore, green to red 182 photoconversion of pcFPs is never exhaustive and will always retain a green population, 183 rendering this quality control step possible. To assess the power of our newly created algorithm, we sought to compare the lineage 208 trees obtained with i) no corrections, ii) after embryo realignment with all algorithmic 209 corrections, and iii) after final manual review. We quantified the effects of the corrections and 210 validations on the quality of the lineage trees by calculating the total distance between the 211 lineage trees using the implementation of the tree Zhang-Shasha edit distance algorithm 11 by Tim 212 Henderson and Steve Johnson 14 . The zss algorithm assigns a (user-defined) cost for each node 213 insertion, removal, and update necessary to transform an ordered tree into another, and gives 214 therefore a quantitative measure of dissimilarity of the two trees. Small tracking differences 215 between corrected and uncorrected trees, however, can result in quite large tree distances if the 216 zss algorithm is applied to the complete trees. A correction that relinks one cell to its mother cell 217 in just one time point causes the whole branch to be flagged as incorrect, and the longer the 218 branch, the higher the distance between the trees. In other words, the earlier the tracking error 219 occurs, the larger the distance; yet, only the first time point in the track is incorrect, and its 220 penalty should be the same whether it happens at the beginning of the time series or the end. 221 To circumvent these issues, we applied the algorithm to a condensed version of the The funding sources had no role in the study design, data collection and interpretation, or the 249 decision to submit the work for publication. 250

Acknowledgements: 251 252
We thank all members of the Pantazis lab and especially M. Haffner for discussion and 253 advice. We thank U. Nienhaus and K. Nienhaus for discussions and advice as well as E. 254 Schreiter, and L. Looger for discussions; W.P. Dempsey for feedback on the manuscript and C. 255 Morkunas for administrative management.