# PEPS: A suggestion/draft for B1

#### Introduction / Rationale

In most animal organs, stem cell reservoirs are crucial to replenish cells lost to injury or simply to "wear and tear". The more complex an organ's structure the more difficult it is to ensure that the new cells are correctly integrated into the existing structure. Nevertheless, even the vertebrate brain, arguably the most complex structure in life, has stem cell niches. To which extent these manage to contribute to maintainance of brain function through life is a topic of active research, with obvious translational implications: can the potential or activity of stem cell niches be boosted to combat neurodegenerative disorders?

Any stem cell niche has to strike a balance between supplying the organ's needs and maintaining enough of its own capacity to last through life time -- the latter by safeguarding some stem cells by keeping them in deeper quiescence and by replenishing them via self-renewing decision. It is becoming apparent that it is not merely individual stem cell's function (cell-autonomous programs) that ensure this balance but rather that this "perpetuation of stemness" is an emergent property, arising from a complex system of interactions between the cells in the niche: on the one hand, feed-backs between stem cells and their progeny, on the other hand, influence by other niche cells.

*Add*: Earlier studies have only distinguished between coarse states (quiescent and active stem cells, amplifying progenitors), but scRNASeq now shows much more details, reveiling find differences (heterogeneity) whose meaning is not elucidated yet.

#### Background and prior work

The zebrafish pallium is a most suitable model of a stem cell niche in the brain: the LBC lab has establishhed a method to repeatedly image it in vivo, over a period of weeks, while tracking individual cells -- thus providing information on stem cells progression through activation, amplification, and differentiation. These studies have revealed, inter alies, inhibitory feed-back from an activated stem-cell towards still quiescent stem cells in its close neighborhood.

While 4D imaging interaction can reveal interactions between cells across space and time, complementary detailed information about cells' internal state can be reveiled by single-cell (sc) RNA-Seq and other sc-omics methods. The AMV lab has pioneered these methods to study the ventricular-subventricular zone (vSVZ) of the adult mouse brain.

In-vivo 4D imaging allows us to track cells in time and see what other cells they interact with but offers only limited information on the cells' internal state. Omics methods with single-cell resolution, on the other side, give detailed information about cell states and their abundances, but cannot tell about the cells' interaction partners nor track their fate over time in vivo.

*Add*: Now introduce roles for SA and AMC

#### New methods for new insights -> Objective and project design

In the proposed project, we suggest to develop several ground-breaking new methods, both new experimental techniques as well as new mathematical approaches, to connect these very different assay modalities and thus derive a comprehensive model of the dynamics of neural stem cell niches. Comparing between two species, mouse and zebrafish, will show which of the proprties of the dynamic characteristics are specific adaptions and which are conserved principles. Furthermore, the methodological innovation will be of wide-ranging use for essential all research aiming to elucidate the dynamics of systems of interacting cells.

*Add*: Necessity to bridge between computational fields.

#### The "single-cell time machine"

Is there something in a resting stem cell's internal state that determines whether it will become activated or not? If we could look into and compare sampled cells' pasts, we could tell how far back we can go to still find small differences in the past that are correlated with big differences at sampling time. Expanding on recent results of our collaborator, J. Gribneau, we will establish the following method: A fusiongene, a fusion of a bacterial methyltransferase to the endogenous RNA polymerase II, is expressed transiently via a doxycycline pulse. Whenever this modified polymerase transcribes a gene, the gene body's DNA will be marked with methyl groups. Days or even weeks later, when the animal is sacrificed, we use combined methylome and transcrptome sequencing (scM&Tseq) and thus reconstruct, from the methylome, each cell's past transcriptome at the time of the pulse and connect it to the cell's transcriptome at sampling time.

So far, inferring temporal dynamics from scRNASeq data can only relies on cell similarity measures, inferring temporal evolutions from connecting similar cells, which hopefully but not necessarily lead teh inferrence along all intermediate states. This error-prone "breadcumb following" can now be replaced with reliable information on long-range connetcions.


We can now, for example, compare cells that have taken different paths (C and D in the Figure) and ask whether their past states (A and B in the figure) already showed differences, perhaps in signalling states. State-of-the-art methods, in contrast, would not allow to see that A evolves to C and B to D rather than vice versa. Furthermore, by having a selecting of several chase times, we can determine whether the path from A/B to C/B must go through cell cycles or might also sometimes take the direct route -- an information so far only available from cell-tracking 4D imaging as discussed above. Finally, the ability to calibrate such scRNASeq-inferred trejectories through feature space to real time allows to connect this data to 4D imaging data.

#### Bridging from scRNASeq to 4D imaging

In order to follow up findings from sequencing-based approaches with imaging-based ones, we need reporters that allow us to recognize subtle differences seen in the scRNASeq data (such as between A and B) also in imaging data. Therefore, by leveraging transcriptional signatures found for these differences, we will construct reporter lines that can indicate whether a cell is in, say, state A rather than B. Specifically, we will construct "quiescent cell indicator" (QuCI) lines that report the different sub states of quiescent and activated NSCs that we find from the scTM data (somewhat similar to the well known FUCCI cell-cycle state reporter system). Such iteration between sequencing- and imaging-based assays, combined with perturbation experiments, will provide much more comprehensive information about the system than either technique could on its own. The synergy between for labs will be essential here, combining expertise on in-vivo 4D imaging (LBC), advanced single-cell multi-omics systems (AMV), development of biostatistical data analysis methods for multimodal data (SA), modelling of complex dynamical systems in continuous state spaces (AMC), while working with both zebrafish (LBC) and mouse (AMV) stem cell niches.

#### Studying interaction between cells

However, so far we are still studying cells in isolation. Yet, we assume that it is interactions and feedback between cells that govern the dynamics. Therefore, the other core aspect of our project comprises methodological advances to study these. Imaging and spatial transcriptomics are the current methods of choice to see which cells are in close proximity and hence might influence each other. The state reporters mentioned above will allow us to assign states to imaged cells with sufficient precision to notice patterns of enrichments of certain neighborhood compositions and, from this, form hypotheses about feedbacks (e.g., finding deeply quiescent cells close to cells in adavnced activation might suggest that activated stem cells drive neighboring quiescent cells into deeper quiescence). Spatial transcriptomics such as the Visium system complements imaging by offering more precise cell state determination, at the cost of less information on temporal evolution.

#### Barcode-transfer sequencing

We will add a third powerful option by establishing barcode-transfer scRNASeq methods: Cells can be engineered to exchange barcode oligos with their interaction partner, either by using rabies virus which take the barcode with them while hopping from cell to cell (RABID-Seq, recently presented by XX et al.) or by using adeno-associated virus that induce expression of oligos and flourescent proteins with an exosome-targeting signal to be transferred from infected cells to their neighbours via exosome trafficking. This latter idea of ours is ideal for prolonged in-vivo studies because it does not exhibit the toxicity inherent with rabies infection, because we can target AAV infections reliably, and because exosome transfer connects across different cell types, thus providing a very comprehensive view. With both methods, subsequent scRNASeq sequencing allows us to connect the assayed cells in two ways: First, the transcritional state positions the cell in feature space, and hence in the network of possible cell-state trajectory, putting cells with differen states next to each other; and second, finding the same barcodes in several cells shows that these have been neighbors in physical space and have been interacting.

#### Interpreting connectome data

Finding a mathematical framework to integrate feature space and physical space data will be a crucial challenge, discussed below. Having solved this, however, we can then mine the transcriptional data of interacting cells specifically for signs of signalling exchanges: For example, consider again the cells in regions A and B in Figure 0, whose subtle difference in transcriptional state might give rise to different fates (C and D). If we were to find that these subtle state differences correlated with a specific signalling ligand being expressed at a higher rate by the neighbors of the A cells than by the neighbours of the B cells, we may hypothesize that this specific signal, sent by these specific niche neighbours, plays an important role in regulating the cells' later decision between C and D. To verify such a hypothesis, we might then construct reporters to indicate the ligands production in the neighbor cells, trace the stem cell's fates with 4D imaging and compare the observed fate with the reporter state of its past neighbours. Furthermore, we will, of course, perturb the observed interaction. Again, it is the synergy between the four PI's complementary expertise that enables such iterations between sequencing, imaging, data interpretation, modelling, and experimental validations.

#### A better mathematical framework to represent our data

As already alluded to, the data obtained in these assays is of a rich but complex nature. So far, there is huge gap between description and modelling, and we now discuss these two sides in turn, to show how to fill the gap. The state of the art to describe dynamic systems from scRNASeq and other sc-omics systems is to represent each cell as a point in the "feature space" spanned by the observed features (i.e., typically, the expression strengths of the genes), then connect cells with similar transcriptional state to infer "trejectories", i.e., branched curves through feature space that map out the cell-state progressions observable in the system. Position along this space is measure with an artificial quantitiy called the "pseudotime" -- but our scTM method will allow to calibrate to real time.

Wherever cell states change quickly and too few intermediate cells are available as "breadcrumbs" to map the trajectory, the trajectory network will have gaps -- which, again, we will fill with scTM data. Similarly, we will finally be able to resolve cycles, a very limiting weak point of state-of-the-art approaches.

#### From trajectory networks to measure spaces

However, it is far from clear how currently used trajectory inference methods can be adapted to make use of scTM data. Furthermore, trajectories are quasi-one-dimensional: they cannot represent the heterogeneity seen among cells with the same position in the trajectory net. (In Figure 0, the cells in A and B would be represented by the same point on the trajectory line passing between them.) We need to augment this quasi-1D structure to capture the distribution of cell states orthogonal to the trajectory. The mathematical concept of measure space seems suitable: we consider the distrubution of cells in feature space as a probability density or as a measure space, which we then factor into a cartesian product of distribution along the trajectory network and a space suitable to represent distribution of cells with the same trajectory position. Temporal evolution can then be described as a stochastic transport process, transporting measures between slices corresponding to different trajectory positions. Here, AMC's expertise in [...] on Radon measures (a topic on which she recently has authored a textbook), will be crucial, together with SA's experience in statistical inference for omics data, needed to connect the framework to the data. The result will be a mathematical framework that describes not only the cell states observed at sampling time, along with their connections via trajectories, but also all available stochastic information on connections across longer temporal ranges.

Put into simpler terms: scRNASeq data allows to infer state evolution only from immediate neighborhood in feature space (similarity of transcriptional state), thus covering time scales of mere hours, while scTM works on the scale of days and weeks. Our envisioned framework based on transport of measures will allow to interpolate and thus bridge between these separated time scales, and so provide the continuous picture required for dynamic modelling.

#### Modelling continuous and heterogeneous state transitions

So far, such modelling was done using compartmentalized or individual-based models, i.e. cells were assigned to a small number of discrete states. However, if one lumps together, say, all quiescent stem cells into just one or two compartments, one cannot study the meaning of subtle differences, such as between A and B in the figure. Consequently, the high resolution in feature space provided by omics data is entirely lost when dynamics are modelled using state-of-the art methods.

The ground-breaking promise of the continuous framework just outlined is thus that it allows to retain during modelling the insight that state changes are gradual and continuous, and that there is heterogeneity within any step along the quasi-1D progression that may influence the dynamic. Methods to model dynamics in continuous state spaces have been studied in mathematics [Anna: here we need a suitable example] and starting from there, we develop an approach suitable for modelling cellular dynamics, that is able to leverage for parameter estimation the rich continuous data from scRNA-Seq, the scale-separation-bridging data from scTM, and the detailled information on interactions from barcode-transfer scRNASeq (locallized in state space and physical space). Studying this model will then give true understanding how the feedbacks we have found are necessary to ensure the stem cell niche's continued ability to produce cells as needed while maintining its own potential.

#### Conclusion

We will explore what governs the ability to perpetuate stemness, and how its limitation explains, among other things the niche's deterioration with age in mice and its long-term robustness in zebrafish -- or, more generally, how the brain's stem cell niches contribute to the brain's function.

On the biology side, these results will be of great value to neurobiology and to stem cell research. Furthermore, and not at all merely as a bonus, our methodological innovations will, for the first time, demonstrate how to study the dynamics of a complex cellular system on a truly comprehensive scale:
- understanding the whole network of state progressions and decisions,
- bridging the gaps between all three time scales (minutes to hours in scRNASeq, days to weeks in scTM, months to years in comparison between young and old animals),
- mapping out not only a cell's journey through its own state progression (and its journey through time and space), but also keeping track of its interaction with a dynamically changing neighborhood, combining both sequencing and imaging-based approaches to assess the connectome,
- integrating all these data into a comprehensive mathematical framework
- establishing a dynamical model that leverages the continuous nature and high resolution of the data
- gaining insights into constraints, bottlenecks, rate limiting components, to understand the necessity behind the observed interactions
- and thus gaining a truly deep understanding of a complex system, viz., the stem cell niche.