Avalilação PREreview de A fast 2C-induction method reveals a barrier role of SP2 for totipotency

This is an interesting and potentially useful study describing a seven-compound chemical cocktail (“2C medium”) that induces MERVL-tdTomato+ 2C-like cells from mESCs within 36 h. The rapid induction is potentially valuable, and the time-resolved single-cell dataset provides an interesting view of the transcriptional changes accompanying the transition. The authors also investigate transcription-factor motifs associated with different trajectory branches and nominate SP2 as a potential regulator of the non-reprogrammed states.

I found the study particularly interesting because it addresses an important question in the field: how efficiently and reproducibly can ESCs be pushed toward a 2C-like state, and what molecular barriers limit this transition?

At the same time, I think several of the central interpretations would benefit from additional evidence or more cautious framing. In particular, the current data do not yet clearly distinguish population-level enrichment of 2C-like features from functional totipotency, and the evidence that SP2 acts as a barrier to reprogramming is currently based primarily on gain-of-function experiments. I outline the main points below.

Major issues

• 1. Evidence for SP2 as a “barrier” would be substantially strengthened by loss-of-function experiments

  The authors propose that SP2 acts as a barrier to the acquisition of the 2C-like state. The current evidence is primarily based on SP2 overexpression, which reduces the tdT+ fraction.

  A barrier model would be more directly supported if depletion of SP2 increased reprogramming efficiency or accelerated the kinetics of tdT induction. I therefore think SP2 KO, KD, degron, or another loss-of-function approach would be important to establish the proposed direction of causality. Since SP1, SP2, SP3, and SP4 share the GC-box motif and may have partially redundant functions, it may also be informative to consider combined perturbation of SP-family factors.

  Several additional points would help interpret the existing gain-of-function experiment:

  • The reduction in Fig. 4f appears to be approximately 47% to 29%, corresponding to an approximately 40% relative decrease, rather than the “approximately 50%” stated in line 349. Fig. 4g appears to show a reduction from approximately 60% to 36%.

  • A Dox-only control in the reporter line would help exclude nonspecific effects of Dox treatment.

  • RcisTarget enrichment in Fig. 4a cannot distinguish SP1, SP2, SP3, SP4, or PATZ1 because of their shared motif. Thus, the motif analysis alone does not specifically nominate SP2.

  • SP2 also does not appear to be the most specifically changed member of the family in every comparison shown in Fig. 4b–c.

  • The ChIP-seq experiment uses an anti-FLAG antibody and therefore measures binding of overexpressed SP2 rather than endogenous SP2. Since overexpressed GC-box-binding factors may occupy GC-rich promoters broadly, the approximately 70% versus approximately 40% promoter-binding comparison in Extended Data Fig. 4f would benefit from an appropriate GC-content-matched background. Endogenous SP2 binding, ideally using a knock-in tagged allele, would provide stronger evidence.

  • Finally, it would be useful to know whether the remaining tdT+ cells under SP2 overexpression resemble normal tdT+ cells. Sorted tdT+ populations or scRNA-seq of SP2-overexpressing cultures could distinguish between fewer cells entering the 2C-like state and cells entering the state incompletely.

  The conclusion that SP2 “activates” lineage-associated genes such as Vegfc, Sema4b, and Sox4 is also currently correlative. ChIP occupancy together with expression differences in bulk populations does not by itself establish direct transcriptional regulation. In particular, reducing the proportion of 2C-like cells would itself be expected to alter bulk expression of lineage-associated genes.

  2. Several key analyses are performed on mixed populations

  The RNA-seq, ATAC-seq, SP2 RNA-seq, and chimera experiments appear to use 36 h cultures containing both reprogrammed and non-reprogrammed cells. This makes it difficult to assign the observed molecular features to the induced 2C-like population itself.

  For example, a strong 2C-like signature in bulk RNA-seq could originate from a minority population, while the opposing pluripotency- or differentiation-associated signature could originate from the remaining cells. This is particularly important when interpreting the effects of SP2 overexpression.

  Sorted tdT+ versus tdT− transcriptomes and chromatin profiles would therefore substantially strengthen the study. If the scRNA-seq was performed using the reporter line, incorporating tdT expression directly into the analysis could also help establish how the inferred transcriptional states correspond to the experimentally measured reporter.

  The chimera experiments would similarly be more informative using sorted tdT+ cells. As currently presented, the experiments demonstrate population-level contribution but cannot determine whether the same induced cells contributed to both embryonic and extraembryonic lineages.

  3. The evidence for functional totipotency should be interpreted cautiously

  The chimera data are interesting because the 2C condition shows considerably more TE-only contribution than the 2i control. However, the frequency of dual ICM+TE contribution appears very low in both conditions.

  Extended Data Fig. 2h shows approximately:

  • 2i: 96% ICM-only, 4% ICM+TE (2/49)

  • 2C: 57% ICM-only, 41% TE-only, 2% ICM+TE (1/46)

  Thus, the frequency of dual ICM+TE contribution is essentially unchanged. The major difference appears to be an increase in TE-only contribution.

  This is compatible with a heterogeneous population containing pluripotent-like and TE-biased cells, but does not by itself demonstrate that individual induced cells have dual embryonic and extraembryonic potential. The data do support enhanced extraembryonic contribution, which is an interesting phenotype, but I think the distinction between this observation and functional totipotency should be made more explicitly.

  Additional lineage-marker staining (e.g., Cdx2, OCT4, and Gata6), appropriate statistical analysis, and ideally clonal or single-cell lineage tracing would help resolve this issue. A post-implantation assay could provide additional functional evidence.

  There also appears to be a methodological discrepancy that should be clarified: the Methods describe aggregation, whereas the figure legend describes injection. Please also check the scale bar in Extended Data Fig. 2g.

  I would recommend using “totipotent-like” rather than “totipotent cells” throughout unless stronger evidence for individual-cell dual lineage potential is provided.

  4. The trajectory analysis does not by itself demonstrate “failed” reprogramming

  The time-resolved scRNA-seq dataset is one of the strengths of the study, but some of the fate interpretations appear stronger than what the analysis can establish.

  Palantir was provided with manually selected terminal states (Methods, lines 575–578). Consequently, the inferred branch structure is at least partly constrained by the predefined terminal states. In particular, a population observed only at 12 h will tend to appear as a terminal state if it is not detected at later time points. Its disappearance therefore does not necessarily demonstrate failed reprogramming.

  The cluster proportions in Extended Data Fig. 3c also raise an important alternative explanation. A relatively small Nlrp12+ population at 12 h appears to account for a large proportion of the cells at 24 h. This could reflect differential proliferation, survival, or both. The manuscript discusses DUX-associated apoptosis, but does not provide direct measurements of viability, apoptosis, cell-cycle state, or total cell number.

  It would therefore be useful to determine whether the increase in the 36 h tdT-high fraction results from expansion of this population, selective survival, or loss of other populations. Likewise, the fate of cells assigned to the “failed” branches at 36 h would be informative.

  There also appear to be some inconsistencies between the text and Extended Data Fig. 3. For example, Foxp1+ cells are described as belonging to the 12 h failed branch (lines 246–247), whereas Extended Data Fig. 3g appears to connect this population with H2-Q10+ and Hoxb1+ cells in the tdT-low path. Lines 229–231 also repeat Arl4d/Tmem72 where H2-Q10/Hoxb1 may have been intended. These annotations should be carefully reconciled.

  5. The reported efficiency should be reconciled across figures and experimental conditions

  The headline efficiency is difficult to reconcile with the range of efficiencies shown throughout the manuscript.

  The reported tdT+ fractions include approximately 75% (Fig. 1b), 38% for “All” (Fig. 1d), 64% (Extended Data Fig. 1a), 60% (Fig. 4g), 47% (Fig. 4f), and approximately 53% in C57BL/6 cells (Extended Data Fig. 1c). The Abstract states “>70%,” whereas the Discussion describes an average efficiency exceeding 50%.

  It would be helpful to clearly define which cell line and experimental condition generated the >70% value and to provide the mean ± variability across independent biological replicates. This would make the performance of the protocol much easier to assess.

  The genetic-background dependence is also potentially important: approximately 53% efficiency is reported for C57BL/6 cells compared with approximately 29% for 129 cells. This suggests that the efficiency may depend substantially on genetic background and that terms such as “robust” should be used with some caution.

  The dropout experiment also deserves clarification. Fig. 1d appears inconsistent with the statement that “the loss of any single component reduced the efficiency” (line 118). Removal of Vc or JNKi appears to give efficiencies equal to or higher than the complete cocktail, while the effect of BML277 removal appears to fall within the experimental variation. If these observations are reproducible, some components may be dispensable under the tested conditions. This would be important for both the mechanistic interpretation and the rationale for retaining all seven compounds.

  6. Comparisons with other totipotent-like cell states require stronger control for technical differences

  The comparison with ciTotiSCs, TPSCs, TLSCs, and TBLCs is potentially informative, but the datasets appear to have been generated using different experimental platforms and protocols, including bulk RNA-seq and scRNA-seq.

  Without explicit batch correction or integration, it is difficult to determine how much of the observed clustering reflects biological differences versus technical differences between datasets. The use of a relatively small curated gene set may further influence the result.

  In particular, the clustering of the 2CLC sample with mESCs in Fig. 2c is unexpected. I suggest verifying the sample identity and annotation, including whether the dataset represents sorted MERVL+ cells, unsorted cells, or Dux-induced cells. It would also be useful to discuss whether this clustering is consistent with the original report from ref. 5. Fig. 2d suggests that the 2CLC sample nevertheless retains some 2C-associated signal.

  A head-to-head comparison using samples generated and profiled in the same experiment would be the most direct solution. Alternatively, an explicitly described and validated batch-integration strategy would help establish the robustness of the comparison.

Minor issues

• 1. The use of “screen” may overstate the experimental design

  Approximately eight compounds, most apparently selected from the authors’ previous work, were tested in a limited number of combinations. No compound library, predefined selection criteria, readout thresholds, or systematic combinatorial design are described.

  I think “literature-guided optimization” or similar terminology would more accurately describe this experiment. The rationale for including BML277 and JNK inhibitor VIII would also be useful.

  2. Clarify the language around small molecules

  The statement that the small molecules themselves “play crucial roles” would be more precise if it referred to the biological pathways or cellular processes that these compounds modulate.

  3. Morphological changes should be interpreted in context

  The cells lose their compact, dome-shaped morphology and become more spread and spiky. Some morphological change would be expected when cells are transferred from 2i/LIF into serum/KSR-containing conditions, and 2C-like cells do not necessarily need to retain a dome-shaped morphology.

  Nevertheless, viability measurements and lineage-marker analysis would help determine whether stress or differentiation contributes substantially to the observed phenotype.

  4. The interpretation of chromatin architectural genes is currently indirect

  Extended Data Fig. 3i describes “more pronounced downregulation” of chromatin architectural genes in the tdT-high branch, but this is not quantitatively clear from the current presentation. The subsequent interpretation in terms of a “relaxed 3D genome” is also indirect.

  A chromosome-conformation assay such as Hi-C would be needed to directly establish changes in 3D genome organization. If such experiments are beyond the scope of the study, the wording could instead be moderated to describe the transcriptional changes without inferring a specific 3D chromatin state.

  5. Data and methodological details

  The study would be easier to evaluate if the authors provided:

  • the biological replicate structure for each experiment;

  • the criteria used to define “tdT-high” and “tdT-low”;

  • analysis code where possible;

  • additional information on the validation of the SP2 ChIP antibody.

  If conclusions about endogenous SP2 binding are intended, validation of the antibody for endogenous SP2, or an orthogonal approach such as a tagged endogenous allele, would be particularly useful.

  6. Terminology and presentation

  A final round of language and terminology editing would improve clarity. Examples include “blastocytes,” “expending,” “maker genes,” and “Interest of Conflict.”

  More broadly, I suggest consistently distinguishing among “2C-like,” “TLC,” and “totipotent” rather than using these terms interchangeably.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.