Species-specific oscillation periods of human and mouse segmentation clocks are due to cell autonomous differences in biochemical reaction parameters

While the mechanisms of embryonic development are similar between mouse and human, the tempo is in general slower in human. The cause of interspecies differences in developmental time remains a mystery partly due to lack of an appropriate model system1. Since murine and human embryos differ in their sizes, geometries, and nutrients, we use in vitro differentiation of pluripotent stem cells (PSCs) to compare the same type of cells between the species in similar culture conditions. As an example of well-defined developmental time, we focus on the segmentation clock, oscillatory gene expression that regulates the timing of sequential formation of body segments2–4. In this way we recapitulate the murine and human segmentation clocks in vitro, showing that the species-specific oscillation periods are derived from cell autonomous differences in the speeds of biochemical reactions. Presomitic mesoderm (PSM)-like cells induced from murine and human PSCs displayed the oscillatory expression of HES7, the core gene of the segmentation clock5,6, with oscillation periods of 2-3 hours (mouse PSM) and 5-6 hours (human PSM). Swapping HES7 loci between murine and human genomes did not change the oscillation periods dramatically, denying the possibility that interspecies differences in the sequences of HES7 loci might be the cause of the observed period difference. Instead, we found that the biochemical reactions that determine the oscillation period, such as the degradation of HES7 and delays in its expression, are slower in human PSM compared with those in mouse PSM. With the measured biochemical parameters, our mathematical model successfully accounted for the 2-3-fold period difference between mouse and human. We further demonstrate that the concept of slower biochemical reactions in human cells is generalizable to several other genes, as well as to another cell type. These results collectively indicate that differences in the speeds of biochemical reactions between murine and human cells give rise to the interspecies period difference of the segmentation clock and may contribute to other interspecies differences in developmental time.

reactions in human cells is generalizable to several other genes, as well as to another cell 23 type. These results collectively indicate that differences in the speeds of biochemical 24 reactions between murine and human cells give rise to the interspecies period difference 25 of the segmentation clock and may contribute to other interspecies differences in 26 developmental time. To compare murine and human segmentation clocks in vitro, we induced PSM-like cells 30 from mouse embryonic stem cells (ESCs) and human induced pluripotent stem cells 31 (iPSCs) (Fig. 1a), as other groups have recently reported 7-12 . In essence, our PSM 32 induction protocol is based on activation of WNT and FGF signaling as well as inhibition 33 of TGFβ and BMP signaling 9,12 . Prior to the PSM induction, mouse ESCs, which are in 34 the naïve pluripotent state, were pretreated with ACTIVIN A and bFGF and converted to 35 mouse epiblast-like cells (EpiLCs) that possess primed pluripotency as human iPSCs do. 36 HES7 oscillations have been proposed to arise from a delayed autoregulatory 1 negative feedback loop: HES7, a transcription repressor, directly binds to and inhibits its 2 own promoter with time delays, resulting in an oscillatory expression of HES7 3 ( Supplementary Fig. 1a) 6,14,29,30 . Knocking out other HES family members, such as HES1 4 and HES5, does not disrupt segmentation in mouse embryos 31 . Since HES7 itself is 5 considered the most critical gene for HES7 oscillation, we first hypothesized that 6 differences in the sequences of HES7 loci between murine and human genomes might 7 lead to the observed oscillation period difference. To test this hypothesis, we swapped 8 HES7 loci between mouse and human ( Fig. 2a): the human HES7 locus, which was 9 defined as the sequence including a promoter, exons, introns, and UTRs of HES7, was between mouse and human, so these results suggest that human HES7 locus in mouse 28 PSM gives rise to an essentially mouse-like oscillation period. 29 One potential defect in our experimental design of interspecies genome 30 swapping is, however, that the swapped HES7 region might not be long enough, and that 31 a crucial sequence for the oscillation period might exist upstream of the HES7 promoter 32 we defined, for instance. To rule out this possibility, we performed 'knock-out and rescue' 33 assays (Fig. 2i): The endogenous mouse HES7 gene was first knocked out in mouse ESCs, 34 leading to disruption of the HES7 oscillation in the induced PSM (Fig. 2j). Then the 35 disrupted oscillation was rescued by introducing an exogenous construct containing a 36 5 promoter, exons, introns, and UTRs of murine or human HES7 locus ( Fig. 2k; 1 Supplementary Fig. 4c). Note that these exogenous constructs were integrated into 2 random positions of the genome by transposon vectors, implying that the HES7 regions 3 used for the constructs should be sufficiently long to restore the oscillations. Importantly, 4 both murine and human HES7 constructs restored mouse-like oscillation periods in the 5 mouse PSM (Fig. 2l). We further attempted a 'complementary' experiment: we knocked 6 out the endogenous human HES7 gene and rescued the disrupted oscillation with the 7 murine or human HES7 construct in human PSM (Fig. 2m, n). Again, murine and human 8 HES7 constructs were indistinguishable in terms of the restored oscillation period (Fig.   9 2o). These results collectively indicate that the 2-3-fold period difference between murine 10 and human segmentation clocks is not caused by the sequence differences between 11 murine and human HES7 loci. 12 We then hypothesized that differences not in the sequences but in the 13 biochemical reaction speeds of HES7 between murine and human cells might lead to the 14 observed oscillation period difference. Since the most important biochemical parameters 15 that affect the oscillation period of HES7 are the degradation rates of HES7 (i.e., δm and 16 δp in Fig. 3a) and the delays in the feedback loop of HES7 (τTx, τIn, τTl, and τRp in Fig.   17 3a) 14,20,29,30,32 , we measured those parameters in both murine and human PSM, exploring 18 which parameter(s) are different between the species. To measure the degradation rate of 19 HES7 protein (δp), we overexpressed either the murine or human HES7 sequence and 20 then halted its expression (Fig. 3b). Interestingly, both murine and human HES7 proteins 21 were degraded more slowly in human PSM as compared with mouse PSM (Fig. 3b, c; 22 Supplementary Fig. 6a), meaning that the changes in the degradation rate depend on the 23 differences not in the HES7 sequences but in the cellular environments (i.e., whether 24 HES7 is hosted in a murine or human cell). The half-life of HES7 protein in mouse was 25 previously reported to be 22 min 29 , consistent with our measurements where half-lives in 26 murine and human PSM were estimated to be 21 ± 0.8 min and 40 ± 4 min, respectively.

27
To measure the delay caused by the transcription and translation of HES7, we 6b). The transcription and translation delay (τTxTl) of HES7 was estimated to be longer in 32 human PSM (30 ± 1 min) as compared with mouse PSM (17 ± 2 min) (Fig. 3f, top). The 33 fitting also allowed us to estimate the degradation rate of HES7 mRNA (δm), showing 34 slower degradation in human PSM (half-life in mouse: 10 ± 0.3 min; in human: 16 ± 0.3 35 min) (Fig. 3f, bottom). Note that the HES7 gene used in these measurements did not 36 6 include the introns (see Fig. 3b, d). Since introns affect mRNA splicing and therefore 1 serve as another source of delays 14,30,32 , we measured the delay caused by HES7 introns 2 by creating HES7 promoter-luciferase reporters with (w/) and without (w/o) HES7 introns 3 (Fig. 3g, h) and estimating the phase difference between the oscillations of the two 4 reporters ( Fig. 3g; Supplementary Fig. 7). Again, the HES7 intron delay (τIn) was longer 5 in human PSM (37 ± 3 min) compared with mouse PSM (13 ± 3 min) (Fig. 3i). Roughly 6 consistent with our measurements, the intron delay or splicing delay in mouse embryos 7 was previously reported to be 12-19 min 14,32 . Finally, to measure the delay for HES7 to 8 start repressing its own promoter, we induced the expression of HES7 and estimated the 9 onset of decline in the HES7 promoter activity ( Fig. 3j; Supplementary Fig. 8). Fitting  To confirm that the degradation rates and delays measured in both murine and 14 human PSM can indeed explain the interspecies period difference in the segmentation 15 clock, we built a simple mathematical model of HES7 oscillation 20 based directly on the 16 following parameters: δp, δm, τTxTl, τIn, and τRp ( Fig. 3k; see Methods). Note that our 17 mathematical analyses of the model showed that the oscillation period depends on these 18 measured parameters (i.e., degradation rates and total delays), and that other parameters, Text 1), varying this parameter within a realistic range did not change the period 23 dramatically (Fig. 3l). Remarkably, our simulation of oscillations with the murine 24 parameters showed periods of ~150 min whereas that with human parameters showed 25 ~300 min periods (Fig. 3l), reproducing the 2-3-fold period difference between actual 26 murine and human PSM (see Fig. 1e). These results mean that the slower biochemical 27 reactions of HES7 (i.e., slower degradations and longer delays) in human PSM as 28 compared with those in mouse PSM are sufficient to quantitatively account for the longer 29 oscillation period of the human segmentation clock. 30 Next, we explored how universal our finding of slower biochemical reactions in 31 human cells is. To test whether it is specific to the HES7 gene or generalizable to other 32 genes, we measured the degradation rates of six other genes, transcription factors 33 expressed at the PSM stage 7 (Fig. 4a, b; Supplementary Fig. 9). GBX2, MSGN1, and 34 TBX6 proteins showed slower degradation rates in human PSM than in mouse PSM, 35 whereas CDX2, EVX1, and Brachyury T did not (Fig. 4c). We also measured the 36 transcription and intron delays (τTx, τIn) (Fig. 4d, e; Supplementary Fig. 10). TBX6, GBX2, 1 and MSGN1 showed longer delays in human PSM than in mouse PSM whereas EVX1 2 did not show a significant interspecies difference (Fig. 4f). These results suggest that the 3 slower biochemical reactions in human PSM with respect to mouse PSM can extend to 4 several other genes, but not to all genes.

5
Finally, to test whether the slower biochemical reactions in human cells are 6 specific to PSM or generalizable to other cell types, we induced neural progenitor cells 7 (NPCs) from mouse ESCs and human iPSCs ( Fig. 4g) Supplementary Fig. 11). All the four proteins tested, OTX2, FOXG1, PAX6, and SOX1, 10 showed slower degradation rates in human NPCs as compared with mouse NPCs (Fig.   11 4i). We also measured the transcription and intron delays of OTX2, FOXG1, and SOX1 12 ( Fig. 4j; Supplementary Fig. 12), demonstrating slightly longer delays in human NPCs 13 for all three genes (Fig. 4k). These results suggest that slower biochemical reactions in 14 human cells can be applicable not only to the PSM fate but also to other cell types, even 15 though more systematic measurements will be necessary in the future. We propose that            To simulate the oscillation of HES7, previously proposed delay differential equations of 25 HES feedback loop were used 20 .
where m and p are the concentrations of mRNA and protein, respectively. δm and δp are 29 the degradation rates of mRNA and protein, α and β are the translation and transcription 30 rates, K is the repression threshold, and n is the repression Hill coefficient. τm and τp are 31 the mRNA and protein delays, and they have the following relationships with the 32 experimentally measured delays: where τRp, τTx, τIn, and τTl are the repression, transcription, intron, and translation delays,

29
Expression delay model: where βT is the transcription rate of the TetOne promoter. 1 The solution of this is where τ = τ + τ , and = β /δ .

5
Degradation model: where βrT is the transcription rate of the rTetOne promoter.

10
The solution of this is 14 As for δp, the value estimated in the degradation assay was used. τTxTl and δm, together Expression delay model: The solution of this is Repression delay model: where f and F are the mRNA and protein concentrations of FLuc, respectively. τTxf and 22 τTlF are the transcription and translation delays of FLuc (τTxTlF = τTxf + τTlF), and αF is the 23 translation rate of FLuc. The numerical calculation was performed with Python, and the 24 resulting F(t) was multiplied by C(t) to incorporate the effect of cell population growth.
where C0 is the initial cell density, γ is the growth rate, and fnorm is the scaling factor for 27 luminescence. As for δp, δm, δF, δf, τTxTl, and τTxTlF, measured values were used. The data of 28 repression delay assay were fitted to F(t)×C(t) manually. The fitting was good when τRp 29 = 0 with both murine and human parameters.      Fig. 3d (Ex2, Ex3). Fitting of Ex1 is shown in Fig. 3e. The same 29 data of degradation assay as Fig. 3e was used for fitting.    Time-lapse imaging of HES7 reporter activity in murine (left) and human (right) PSM.            The mathematical model of the HES7 system that we use to describe the behavior of the segmentation clock is [1]: where m and p are the concentrations of HES7 mRNA and protein, respectively, α and β are the translation and transcription rates, K is the repression threshold, n is the repression Hill coefficient, δ m and δ p are the degradation rates of the mRNA and protein, which we have measured experimentally as explained in the main text. The mRNA delay τ m is composed by the repression delay τ Rp , the transcription delay τ Tx , and the intron delay τ In , all of which we have measured experimentally as well. The protein delay τ p , in turn, corresponds to the translation delay τ Tl , which was also quantified using our experimental observations.
This system has a fixed point (m * , p * ) for which the two derivatives above are zero, which obeys: The stability of this fixed point can be analyzed by assuming the following temporal response to a small perturbation (a, b): m(t) = m * + a exp(λt) (4) p(t) = p * + b exp(λt) Introducing expressions (4)-(5) into Eqs. (1)-(2), linearizing around a = b = 0, and imposing that a solution of the form (4)-(5) exists with nonzero a and b, leads to the following transcendental characteristic equation for the eigenvalues λ: (λ + δ m )(λ + δ p ) + n(δ m δ p ) 2 αβ p * n+1 K n exp(−λ(τ m + τ p )) = 0 , In general the eigenvalues are complex numbers λ = σ + iω. The eigenvalue with highest real part determines the stability of the fixed point, with σ < 0 corresponding to an unstable fixed point, and σ > 0 to a stable one. The corresponding imaginary part establishes the frequency at which the system oscillates towards the fixed point (if σ < 0) or away from it (if σ > 0). In the case of an unstable fixed point with ω = 0, the system usually falls on a limit cycle whose period can be expected t o be close t o 2π/ω. Taking t hese considerations into account, Eq. ( 6) s hows t hat the period of the H ES7 oscillations does not depend on the mRNA and protein delays separately, but only on the total delay τ m + τ p . 1 For the parameters that we consider in this paper, p * K, as can be seen in Fig. 1, which represents graphically the solution of Eq. (3) as the crossing point between its left-hand side (blue line) and right-hand side (orange line). In the limit p * K, Eq.
(3) has the following approximate solution: Inserting expression (7) into the characteristic equation (6) leads to: (λ + δ m )(λ + δ p ) + nδ m δ p exp(−λ(τ m + τ p )) = 0 , Considering again that the imaginary part of the eigenvalue with the highest real part gives us an estimate of the oscillation period, we can observe from Eq.  Finally, if we focus on the bifurcation point (σ = 0), we can obtain in closed form the period of the oscillations at that point by computing ω. To that end, we write the real and imaginary parts of Eq. (8) for λ = iω and divide one by the other, to reach the following transcendental equation: We can thus see that at the bifurcation point, the period of the oscillations does not depend on n, but only on the degradation rates of the mRNA and the protein, and on the total delay. These observations are reproduced by our numerical simulations.