Widespread changes to metabolite pools during early X. laevis development
When the abundant yolk platelets are excluded, X.laevis
eggs have a volume of ~500 nL, equivalent to ~250,000 mammalian tissue culture cells. In preliminary experiments, we found that we were able to measure the abundance of numerous metabolites in single X.laevis
eggs and embryos. (Here-in, once eggs have been fertilized, they are called embryos.) By measuring metabolites in single eggs and embryos, we could avoid the averaging implicit in many biological measurements and determine the extent of variation between individuals. For each measurement, single eggs or embryos were rapidly quenched in a 2
1 acetonitrile:methanol:water mixture at −20°C. Soluble metabolites were extracted from the insoluble cellular material (e.g. protein, DNA, RNA) and subjected to metabolomic analysis with a liquid chromatography- tandem mass spectrometry (LC-MS/MS) method. This platform has been previously demonstrated to provide reproducible and linear measurements of more than 100 common metabolites 
. We were able to detect a core set of 48 metabolites in single eggs or embryos in nearly all of our experiments (). We focused our analyses on this robust set of 48 metabolites, which represents a broad cross-section of central metabolic pathways, including amino acids, nucleotides, and phosphorylated sugars.
Remodeling of the metabolome during early X.laevis development.
We prepared extracts from single eggs and from single embryos in the earliest stages of development. Embryos were collected 2.5, 4.5, 7, 9, and 11 hours after in vitro fertilization. The time course was repeated with three crosses of unique male and female frogs. The stages of development examined include the cleavage cycles, the MBT, and the late blastula period (). Phosphorylation of tyrosine-15 (pY15) in Cdc2 increased 8 hr after fertilization, which marks the onset of the MBT () 
. The time course ends prior to gastrulation (~13 hr post-fertilization).
Widespread changes occur to the metabolomes of frog embryos in the time between egg fertilization and the late blastula stage 11 hours later (). The abundance of at least 15 of the core 48 metabolites changed during these early stages of development. To confirm this surprising result and the validity of the single embryo measurements from the LC-MS/MS system in the Rabinowitz laboratory (Lewis-Sigler Institute for Integrative Genomics, Carl Icahn Laboratory, Princeton University, Princeton, NJ, U.S.A), we performed a second set of time courses, altered many of the details of quenching and extraction, and then quantified metabolites with a different LC-MS/MS system in the Mootha laboratory (Broad Institute of MIT and Harvard, Cambridge, MA, USA). Although each laboratory’s system is calibrated to measure a different set of metabolites, the great majority of overlapping metabolites showed consistent patterns following fertilization (Figure S1
Throughout these early stages of development, the key energetic and reductive currencies ATP, NADPH, and NADH remained nearly constant with little embryo-to-embryo variation in abundance (). But intermediates in the core metabolic pathways that sustain ATP, NADPH, and NADH levels did change in abundance. Most of the components of the glycolytic pathway that we measured showed reproducible increases (, Table S1
), with the phosphoenolpyruvate (PEP) pool showing a particularly sharp ~4–8 fold expansion just following fertilization (). Like the glycolytic intermediates, the sedoheptulose-7-phosphate (S7P) pool expanded greatly in early development (~8–32 fold, ). S7P is the only constituent of the non-oxidative portion of the pentose phosphate pathway (non-ox PPP) that we measured and the non-ox PPP is connected to the glycolytic pathway via reversible reactions. The rise in S7P concentration likely reflects an equilibrium between S7P and the glycolytic intermediate fructose-6-phosphate (F6P), whose concentration also rises during the time course. In contrast, the intermediates in the tricarboxylic acid (TCA) cycle showed more complicated changes. The α-ketoglutarate pool consistently expanded by ~4–16 fold (), but there were no reproducible increases in any of the three other tricarboxylic acid (TCA) cycle components that we measured. Rather, malate, citrate/isocitrate, and fumarate showed less dramatic, coordinated changes and these changes were distinct for each of the three individual time courses (Figure S2
Our core set of 48 metabolites includes 16 of the 20 amino acids that are incorporated into proteins. Of the 20 amino acids, 11 cannot be synthesized de novo by animals and are often called essential (we include cysteine and tyrosine in the essential group because their synthesis requires the breakdown of other essential amino acids) 
. While phenyalanine and to a lesser extent histidine increased during early development, the other 7 measured, essential amino acids did not change in abundance. The 9 amino acids that can be synthesized de novo in animal cells are closely derived from glyolytic and TCA intermediates, some via a single transamination step. Several of these amino acids did show strong, consistent changes with aspartate falling ~3–8 fold and glutamine rising ~4–7 fold by the late blastula stage ().
To confirm and expand on the metabolomic results, we used amino acid analysis (AAA) to measure the abundance of free amino acids during some early and later stages of development. With one exception (phenylalanine), AAA confirmed the patterns observed in our metabolomic analyses and also provided the absolute abundance of each of the 17 measured amino acid pools (Figure S3
). During the early stages of development, polysome content and protein synthesis rates are low but not insubstantial. Amino acid incorporation into proteins has been estimated to increase from ~50 pmol hr−1
just following fertilization to ~150 pmol hr−1
in late blastulae 
. Based on our AAA, these low rates would still be sufficient to drain the pools of most essential amino acids over the 11 hr time course. As most of the 9 essential amino acid pools that we could measure remained constant, it is likely that the net rates of protein degradation and synthesis are fairly evenly matched during these early stages of X.laevis
Frogs lay eggs with distinct metabolite concentrations and display distinct metabolic changes in response to fertilization
A minority of the metabolite changes we observed in early development were clearly consistent in the single unfertilized eggs and embryos derived from the same cross but were not reproduced in clutches derived from different crosses. This inter-cross variability was particularly evident for the TCA cycle components (Figure S2
), as well as for alanine and proline, which showed strong declines in abundance in only one of the three clutches (). By AAA, eggs derived from three different frogs contained three different alanine concentrations (range: ~0.4–2 mM). But, for all three frogs, the alanine concentration in the resulting embryos rapidly decreased after fertilization and converged to a low level (~0.1 mM) in late blastulae (Figure S3
). It appears that for some metabolites, like proline and alanine, different frogs lay eggs with different quantities of that metabolite. The X.laevis
frogs used in this study are outbred and so approximate the level of variation found in wild populations. We conclude that in addition to being able to buffer the genetic variation within a population, the developmental program is also robust to some metabolic variation in the egg.
Alanine is an energy source for early X.laevis embryos
It has been proposed that prior to gastrulation, amino acids are the primary source of energy for X.laevis
. The cognate amino acid(s) was never identified, although very early work had suggested that aspartate levels fall prior to gastrulation in several amphibian species 
. Here, we have definitively identified alanine and aspartate as amino acids that are consumed during the early development of X.laevis
(, Figure S1
To better understand the rapid decline in alanine that accompanies fertilization, we injected alanine containing stable, heavy isotopes of nitrogen (15
N) or of carbon (U-13
C) into 2-cell embryos and tracked the movement of the isotopes into other metabolite pools by mass spectrometry (). The heavy nitrogen from the injected alanine was rapidly transferred into the glutamate and aspartate pools (). The more rapid initial increase in the proportion of labeled glutamate indicates that the dominant enzymatic activity is an alanine-glutamate aminotransferase, with amino groups then being transferred to aspartate by an aspartate-glutamate aminotransferase. The activity of such aminotransferases was further evidenced by reproducible drops in the abundance of α-ketoglutarate, the α-keto acid equivalent of glutamate, and malate, which is likely in equilibrium with oxaloacetate, the α-keto acid form of aspartate (Figure S4
). Nitrogen from alanine fluxes through the aspartate and glutamate pools to other amino acids, including valine, proline, and glutamine (). Although we could not directly measure pyruvate, the alanine-glutamate aminotransferase reaction described above will lead to the formation of pyruvate, the α-keto acid equivalent of alanine. Labeled pyruvate should therefore form after injecting alanine with three heavy carbon atoms (U-13
C-alanine). Consistently, we observed groups of 2 or 3 heavy carbons entering into the pools of TCA cycle intermediates upon injection of U-13
C-alanine (). Malate with 3 heavy atoms appeared, demonstrating the activity of pyruvate carboxylase (PC), a mitochondrial enzyme that catalyzes the reaction of a bicarbonate ion (which will be predominantly 12
C) with pyruvate to form oxaloacetate (). Oxaloacetate is expected to be in equilibrium with malate through the action of malate dehydrogenase. PC activity may help restore the pools of TCA intermediates that were partially drained after the alanine injection (Figure S4
). In contrast, the rapid accumulation of citrate and α-ketoglutarate with 2 heavy carbons demonstrates the sequential action of pyruvate dehydrogenase and citrate synthase, which catalyze the transfer of 2-carbon units from pyruvate into the TCA cycle for the purpose of generating ATP.
Metabolic fluxes that were traced in early embryos.
Alanine consumption generates energy in early embryos.
We used the same approach to track the fate of the aspartate consumed during early development. Nearly immediately upon being injected (<3 min) into the 2-cell embryo, the heavy carbon chain of U-13C-aspartate containing four 13C atoms was equilibrated with malate, suggesting a rapid flux through aspartate-glutamate aminotransferase and malate dehydrogenase (). The heavy carbon chain was then distributed around the TCA cycle as seen in the rise of α-ketoglutarate labeled with three 13C atoms.
The metabolic fluxes that we observed in early embryos are summarized in . Based on our data as a whole, we propose that the decline in aspartate and the rise in glutamine are linked. By AAA, embryos consume ~2 nmol of aspartate by the late blastula stage but produce ~1 nmol of glutamine (Figure S3
). By this stoichiometry, the nitrogen is roughly accounted for and free ammonium levels would not rise, which is known to be the case 
. In this view, the egg is stockpiled with aspartate, which acts as a source of nitrogen for generating glutamine for the embryo. Although we are unsure what the benefit of these changes to the embryo are, as aspartate and glutamine are the two nitrogen providers for nucleotide biosynthesis, this flux may help ensure adequate availability of usable nitrogen for nucleotide biosynthesis.
Declining dATP pool: reconsidering an old model for the MBT
We observed reproducible changes in the abundance of dATP (). dATP levels rose following fertilization and then started falling around the time of the MBT (~8 hrs post-fertilization). There was a ~2-fold change between the peak concentration observed at 7 hours after fertilization and the concentration measured in late blastulae, at 11 hours post fertilization. dTTP and dCTP pools exhibited more variable, clutch-dependent patterns and dGTP could not be detected by our methods (). While not as dramatic as many of the changes we observed, these changes were nevertheless somewhat surprising, as it is thought that cells tightly control dNTP concentrations 
. All four dNTPs are synthesized through the action of ribonucleotide reductase (RNR). A complex set of allosteric feedback loops are thought to allow the four dNTP pools to achieve the desired concentrations and stoichiometry, in large part by controlling the activity of RNR towards each of its substrates. But the dNTPs clearly did not all rise and fall in perfect tandem in our data and it has been previously reported that dNTP pool sizes can change differentially. For instance, in mammalian cells, dATP pools are far more sensitive to inhibition of RNR by the small molecule inhibitor hydroxyurea (HU) than are the other dNTP pools 
Evidence suggesting that dATP limitation triggers the MBT.
We considered potential causes of the decline in dATP around the time of the MBT. We observed no change in the abundance of either ribonucleotide reductase (RNR) subunit (). It also seemed unlikely that the declines in dATP abundance were caused by changing cell cycle distributions in the ~4000–8000 cell embryo at the MBT, since S-phase is when dNTP levels would be expected to be highest if these levels were cell cycle regulated at later stages of development 
, , but the first cell cycle phase to be extended at the X.laevis
MBT is the S-phase 
. In fact, the proportion of each cell cycle devoted to S-phase does not decrease as cells traverse the MBT and cell cycles elongate in late blastulae 
We reasoned that the embryo’s dATP pool might decline around the MBT due to the increasing demands of DNA synthesis. The number of genomes increases exponentially during the cleavage cycles but the length of S-phase remains constant (~20 min) until the MBT 
. Therefore, for the embryo as a whole, the rate of dNTP fixation into DNA also increases exponentially. We used time lapse imaging of the animal poles of intact embryos to re-examine the timing of the MBT. We found that cells at the animal pole of X.laevis
progressed through 13 synchronous cleavage cycles but the 13th
cycle was sometimes, but not always, noticeably longer (). When the 13th
cycle was longer, it was a <20% increase (n
5). The 13th
cycle is one cycle later than was reported in two studies of cells from dissociated embryos 
, but is consistent with a previous report in which intact embryos were imaged 
. Also as previously reported, incubating embryos with HU led to few subsequent divisions after the 12th
cleavage cycle, and embryo death () 
. As HU will prevent the synthesis of additional dNTPs by RNR, this result strongly suggests that embryonic pools of at least one dNTP are functionally exhausted around the time of the MBT, leading to cell cycle arrest in S-phase and cell death. As the embryo contains ~4096 cells after 12 synchronous divisions, these experimental findings are fairly consistent with the finding that unfertilized X.laevis
eggs contain ~10 pmol of each dNTP which would suffice to synthesize ~2500 X.laevis
We consistently observed, however, that imaged embryos went through the cleavage cycles more slowly than non-imaged siblings (Figure S5
). This phototoxicity was present even when light exposures/intensities were kept to the minimum necessary to obtain reasonable images (3s of light every minute) and when all UV wavelengths were filtered out. Despite their slower cleavage cycles, imaged embryos did go on to develop normally (data not shown), perhaps explaining why this phototoxicity had not been previously reported. We were concerned about the effects this mild phototoxicity could have on the relative timing of events. To confirm the coincidence between the timing of the MBT and the time at which cell cycles arrest in response to HU treatment, we measured the abundance of pY15-Cdc2 in embryos that were only exposed to normal room lighting in the presence or absence of HU (). This alternative approach demonstrated that HU does not affect the embryonic cell cycle until the MBT, confirming the functional exhaustion of dNTP pools at the MBT.
The coincidence between the time at which maternal pools of dNTPs are exhausted and the MBT, as well as our novel result that dATP pools are diminished around the time of the MBT, support an old idea that the MBT is triggered by declining dNTP pools 
. In this model, the MBT occurs when dNTP pools and the biosynthetic capacity of RNR are outpaced by the exponentially increasing demands for dNTPs at replication forks. Although dATP does not decrease dramatically in abundance (~2-fold, ), replication forks in yeast arrest when dNTP concentrations are only 20% lower than normal 
. One significant problem with the limiting dNTP model is how difficult it has been to definitively test. Newport and Dasso were unable to repeat the results of Lovtrup and colleagues, who claimed that injecting dNTPs could increase the number of sychronous cleavage cycles in X.laevis 
. When we attempted this experiment, we consistently found that injecting dNTPs to 100–300 µM, or ~2.5–7.5 fold their endogenous concentration, slowed the cleavage cycles relative to control siblings. The unknown toxic effects of dNTP injection made further interpretation treacherous.
To our knowledge, the limiting dNTP model is consistent with every fact that is known about the MBT in X.laevis
. The MBT occurs when the DNA:cytoplasm ratio increases above a critical threshold 
. In the limiting dNTP model, the amount of DNA is represented by the number of replication forks that are fixing dNTPs into polymer, while the cytoplasm is represented by dNTP pools and RNR enzyme. As frog eggs have been shown to quickly form nuclei around injected plasmids 
, and frog extracts as well as frog eggs replicate injected plasmid DNA 
, the limiting dNTP model is compatible with experiments showing that injected plasmids can cause a precocious MBT 
. It has been demonstrated that the ATR-Chk1 replication checkpoint is transiently activated during the MBT 
. In the limiting dNTP model, depleted dNTP pools would cause replication forks to stall after ~12–13 replication cycles, activating the ATR-Chk1 checkpoint pathway to retard origin firing and extending S-phase. ATR-Chk1 is known to be the primary signaling route for stalled replication forks in animal cells and, in X.laevis
egg extracts, ATR signaling has been shown to suppress origin firing 
As the orthologs of Chk1 and ATR are required to slow the rapid replication cycles of syncytial Drosophila
embryos and to coordinate cell size and cell cycle time in early C.elegans
, limiting dNTPs could be a conserved feature of early animal development, possibly including mammals 
. An attractive feature of the limiting dNTP model is that it provides a mechanistic explanation for how the MBT occurs so reproducibly from embryo-to-embryo. That is, rather than requiring the female frog to accurately load each egg with a certain amount of a limiting protein, embryo-to-embryo variation in dNTP pool sizes would be intrinsically low due to allosteric control over RNR.
It is intriguing what this model, if true, would say about relations between cell size and the cell cycle. The average embryonic frog cell after 13 cleavage divisions has ~60 pL of non-yolk volume and is much too large to be affected by the cell size requirement for the G1/S transition that has been surmised to exist in typical somatic cells (e.g. a HeLa cell is ~2 pL) 
. The limiting dNTP model suggests that diminishing cell size first limits the speed of the cell cycle by decreasing the absolute rate of the metabolic fluxes required for DNA synthesis. Perhaps this would not be so surprising. In most cells, dNTPs need to be synthesized in real-time during DNA replication. Even for the largest dNTP pool in mouse fibroblasts (dTTP) 
, the amount of dTTP that needs to be polymerized to duplicate the genome is greater than 40 times the pool size.