Because of their capacity to divide and differentiate into specialized cells, stem cells offer a potential treatment for diseases such as diabetes and heart disease Figure 3. Cell-based therapy refers to treatment in which stem cells induced to differentiate in a growth dish are injected into a patient to repair damaged or destroyed cells or tissues. Many obstacles must be overcome for the application of cell-based therapy. Also, the destruction of embryos to isolate embryonic stem cells raises considerable ethical and legal questions.
In contrast, adult stem cells isolated from a patient are not seen as foreign by the body, but they have a limited range of differentiation. Some individuals bank the cord blood or deciduous teeth of their child, storing away those sources of stem cells for future use, should their child need it. Induced pluripotent stem cells are considered a promising advance in the field because using them avoids the legal, ethical, and immunological pitfalls of embryonic stem cells. One of the major areas of research in biology is that of how cells specialize to assume their unique structures and functions, since all cells essentially originate from a single fertilized egg.
Cell differentiation is the process of cells becoming specialized as they body develops. While all somatic cells contain the exact same genome, different cell types only express some of those genes at any given time. The primary mechanism that determines which genes will be expressed and which ones will not is through the use of different transcription factor proteins, which bind to DNA and promote or hinder the transcription of different genes. Through the action of these transcription factors, cells specialize into one of hundreds of different cell types in the human body.
Arrange the following terms in order of increasing specialization: oligopotency, pleuripotency, unipotency, multipotency. Explain how a transcription factor ultimately determines whether or not a protein will be present in a given cell? Skip to content Increase Font Size.
Chapter 3. The Cellular Level of Organization.
Learning Objectives By the end of this section, you will be able to: Discuss how the generalized cells of a developing embryo or the stem cells of an adult organism become differentiated into specialized cells Distinguish between the categories of stem cells. Review Questions 1. Which type of stem cell gives rise to red and white blood cells? What multipotent stem cells from children sometimes banked by parents? Critical Thinking Questions 1. Discuss two reasons why the therapeutic use of embryonic stem cells can present a problem. Each trace is colored by its relative end-point Nanog immunofluorescence intensity level.
Inset bar shows mean as well as upper white and lower quartiles of the transition durations of cells. The inferred dynamics of differentiation can therefore be visualized in a low-dimensional subspace of gene expression, showing that differentiation occurs through a sequence of discrete cell state transitions. Inspection of the genes that make up the local transition and marker gene classes Figure 3C ; Figure 3—source data 1 allowed us to match clusters to embryonic cell types found in vivo that show similar gene expression.
Clusters C 2 and C 3 , which branch out from C 1 , show differential expression of pluripotency genes relative to C 1 ; Bptf and Cbx1 are downregulated in both C 2 and C 3 , Oct4 , Etv5 and Dnmt3a are downregulated in cluster C 3 but maintained in C 2 , and Sox2 , Otx2 and Dnmt3b are downregulated in cluster C 2 but maintained in cluster C 3. Cluster C 2 is further characterized by a high expression level of primitive streak markers Mixl1 and T Hart et al. Together, these patterns strongly suggest that clusters C 2 and C 3 represent mesendoderm and bi-potent ectoderm progenitor cell types, respectively.
The bi-potent ectoderm progenitor-like cluster C 3 is then followed by a lineage split into clusters C 5 and C 6. While Stmn4 is downregulated in both C 5 and C 6 compared to C 3 , Sez6 is downregulated in only C 5 , and Stmn3 as well as neural progenitor marker Pax6 are downregulated in C 6 but maintained in C 5. Cluster C 5 is further characterized by Smarce1 and Zic2 , and cluster C 6 by Slug and Msx2 , suggesting that C 5 and C 6 may be related to neural progenitor and neural crest cells, respectively Brown and Brown, ; Nicole and Chaya, ; Vogel-Ciernia and Wood, Cluster C 4 , although similar in its expression level of Mixl1 and T to cluster C 2 , shows higher expression of other primitive streak genes such as FoxA2 and Tcf3 Merrill et al.
Cluster C 4 is then followed by a bifurcation between clusters C 7 and C 8. Cluster C 7 shows high expression levels of Gata4 and Snai1 , indicative of its relation to mesoderm, and cluster C 8 is characterized by high FoxA2 compared to clusters C 4 and C 8 , suggestive of its relation to definitive endoderm Kim and Ong, ; Rojas et al. We predict that cluster C 4 represents a primed bi-potent mesendoderm cell type relative to cluster C 2 Nakanishi et al. Together, these results suggest that the cell clusters and sets of transitions computationally inferred from single-cell transcriptomics data correspond to known in vivo cell types and their lineage relationships.
The fact that gene expression in each cell cluster does not vary significantly — as measured by the relative sizes of the largest eigenvalues of the PC components of the gene expression data or percent variance explained thereby versus that of the same data randomly shuffled Figure 3B ; Figure 3—figure supplement 1B — allows for genes to be sorted into a few gene classes that show highly correlated expression patterns across clusters Figure 3C.
This suggests that one can validate the inferred sequence of cell state transitions and its gene expression dynamics by measuring the expression of one gene from each class in differentiating cells over time. In order to confirm the gene expression dynamics over the inferred sequence of cell state transitions, we assessed populations of cells for their expression levels of key transition and marker genes each taken from a different gene class via immunostaining and flow cytometry.
Although T is not assigned to a specific gene class, it is highly expressed in the mesendoderm-like states C 2 and C 4 , and it thus allows us to distinguish C 2 from the earlier epiblast-like state C 1.
The flow cytometry density contour plots shown Figure 3D are characterized by high-density peaks which are separated from one another by regions of low density, mirroring the discreteness of the cell states inferred from single-cell transcriptomics data. The relative locations of these high-density peaks and the time at which they appear and disappear recapitulate the inferred gene expression dynamics of the cell state transitions of the lineage tree.
During the first two days of differentiation, all cell populations downregulated Klf4 and upregulated Otx2, as shown in the first row of density contour plots in Figure 3D. On day three of differentiation third column of plots in Figure 3D , Sox2 and Oct4 are asymmetrically downregulated relative to the preceding population, as is seen in mesendoderm-like state C 2 and bi-potent ectoderm-like state C 3 relative to the epiblast-like state C 1.
Sox2-high, Oct4-low cells on day three are either high for Pax6 or for Slug, consistent with comparisons between the neural ectoderm-like state C 5 and neural crest-like C 6.
Oct4-high, Sox2-low cells on day three of differentiation are high for T, but show two discrete levels of FoxA2, mirroring the difference between the two mesendoderm-like states C 2 FoxA2-low and C 4 FoxA2-high. Further, we found that Etv5, a gene whose expression dynamics had hitherto not been implicated with early mesendodermal differentiation in mammals, was significantly downregulated from C 2 to C 4 , as predicted from the single-cell gene expression data Figure 3—figure supplement 1C. Finally, at days four and five, we observe FoxA2-high, Gata4-low and FoxA2-low, Gata4-high cell populations, which correspond to the primed mesendoderm and definitive endoderm-like states C 4 and C 8 and the mesoderm-like state C 7 , respectively.
We thus confirmed that differentiating cell populations recapitulate the gene expression dynamics of cell state transitions inferred from single-cell data Figure 3A.
The observation that the majority of randomly sampled cells are found to belong to one of nine discrete cell states both transcriptionally and at the protein level suggests that cell state transitions occur within a relatively short timeframe compared to the amount of time cells spend within each state. In agreement with our hypothesis, we observed that Otx2 levels, at the end of 24 hr of differentiation, show a bimodal distribution Figure 3F , top , and cells tend to occupy either an Otx2-low state corresponding to ES state C 0 or an Otx2-high state corresponding to epiblast-like state C 1.
We find that cells transition from an Otx2-low to an Otx2-high state well within the duration of a single cell cycle mean transition duration of 4. In contrast, cells tend to stay in either Otx2-low or -high states for up to multiple cell cycles, with a large amount of cell-to-cell variability in the residence duration Figure 3F , bottom.
Together with our results from the analysis of single-cell transcriptomics data, these observations show that cells reside in discrete states in gene expression space and correspondingly undergo abrupt state transitions. Our analysis of single-cell gene expression data suggested a lineage tree composed of discrete cell states, and identified genes associated with individual cell states and transitions between them. While we predict the existence of discrete cell states based on their gene expression pattern, finding unique physiological properties that can define and distinguish their existence functionally would lend even greater support to this prediction.
We therefore next sought to find properties of cell states that distinguished them functionally from one another. In order to do so, we built a predictive and testable quantitative model of the underlying gene regulatory network based on the expression patterns of the marker and transition genes. From the genes that were categorized as either marker or transition genes for all the high probability triplets, we first chose genes involved only in the triplets that fall directly along the inferred lineage tree.
For instance, we did not consider the genes categorized as marker or transition genes only in the triplet C 0, C 1 and C 5 , because C 3 is skipped between C 1 and C 5. Since some transition genes inferred from our Bayesian analysis are re-used to infer multiple local state transitions Figure 3C ,e. Further, because our goal was to test whether different cell states were functionally distinct i.
These signaling factor genes constituting each of these modules were selected based on GO categories, leading to a total of 29 gene modules Materials and methods. We denote each gene module by a representative gene in square brackets; for example, the gene module that uniquely characterizes the ES state C 0 is denoted as [ Klf4 ] Figure 4—source data 1 and 2. Owing to the large number of gene modules, and consequently even larger number of potential interactions between these modules, even the simplest mathematical model would consist of hundreds of parameters.
However, for most of these parameters, direct experimental measurements are not available. In order to overcome this challenge, we exploited recent developments based on renormalization group approaches to determine which parameters are relevant for the observed data Machta et al. We adapted the seminal model of artificial neural networks, known as the Hopfield model Fard et al. By construction, we required that this mathematical model produces the nine cell states seen in Figure 3A.
We considered a network that contains direct interactions, in which each module j exerts a drive on module i , which is equal to an interaction strength J i j positive or negative multiplied by the concentration of module j. The total drive on module i is the sum of the drives from the different modules. For simplicity, we assumed that the expression of every gene module exhibits a non-linear, step-function response, when subjected to the same drive; thereby reducing the number of parameters of the model.
Indeed there are numerous genes that manifest sigmoidal-like response in expression, in the presence of internal and external stimuli Lebrecht et al. Thus the effective dynamics of expression levels m i of each module i are given by the non-linear equation:. We determined the set of interactions J i j that are consistent with the observed cell states C 0 -C 8 , Figure 3A being stable fixed points of the network. Thus, for each stable state, we have 29 constraints on the possible values of J i j , one for each module.
The problem is therefore underdetermined even for our simplified model of the underlying network, and there are an infinite number of solutions that would allow for the observed cell states to be stable. By using a linear programming method to obtain an ensemble of 10, sets of J i j interactions Materials and methods , each satisfying the constraint that all nine cell states are stable fixed points, we estimated the probability distribution for the parameters of the model Figure 4A and Figure 4—figure supplement 2 , giving us a probabilistic model of the underlying network.
We further assumed that all the possible 10, sets of J i j interactions that reproduced the nine stable cell states were equally likely, since we did not have any experimental evidence to distinguish between them. Each circle represents a gene module. Mean positive and negative interactions between the modules are shown in red and green, respectively, and their thickness and transparency are proportional to the absolute magnitude of the mean and the coefficient of variation c.
The colored circles represent the gene modules expressed uniquely in only one of the cell states color code matched with Figure 3A for each state. As cells transition from C 0 to C 1 , expression of [ Klf4 ], [ Apex1 ], [ Ets2 ], [ Atf2 ] modules is downregulated shown in gray while [ Hes6 ] and [ Otx2 ] modules are upregulated, leading to changes in the effective interactions between gene modules that are common to both C 0 and C 1 states, such as [ Sox2 ] and [ Oct4 ].
D [ Sox2 ] overexpression x-axis plotted against the probability of [ Oct4 ] downregulation y-axis computed over 10, models Materials and methods. In order to obtain the error bars for this and subsequent predictions, we randomly sampled three subsets of from the 10, models. For each set we computed the mean and standard error of the proportion of models that show downregulation of Oct4 in response to Sox2 overexpression.
E, F Subsets of the model consisting of gene modules that are expressed in the epiblast-like C 1 E and mesendoderm-like C 2 F states, and their interactions with [ Snai1 ], which is not normally expressed in C 1 or C 2. As cells transition from the C 1 to C 2 state, [ Hes6 ], [ Sox2 ], [ Otx2 ], [ Churc1 ] are downregulated shown in gray , while [ Hmga1 ], [ T ], [ Atf2 ], [ Hes1 ], [ Ets2 ], [ Apex1 ], [ Brd7 ], [ Hmgn2 ] and [ Smarce1 ] are upregulated, leading to changes in the effective interactions between [ Snai1 ] and modules that are common to both C 1 and C 2 , such as [ Oct4 ].
G The probability of [ Oct4 ] being downregulated y-axis as a function of [ Snai1 ] overexpression x-axis. I Cells in different states are predicted to respond differently to morphogens. We used this probabilistic model to make testable predictions as to how different cell states respond to perturbations: to see if different cell states are defined not only by their distinct transcriptional profiles, but also functionally distinct in their phenotypic responses to the same perturbations.
There are a vast number of testable predictions that one could extract from our gene regulatory network model. However, given the low throughput nature of perturbation experiments, we selected three distinct probabilistic predictions, each probing different aspects of the model gene regulatory network. First, we considered changes in the effective interaction between two gene modules as a function of cell state i. To this end, we looked at two classes of gene module pairs: i gene modules that are co-expressed in two mother-daughter cell states and ii gene modules that are never co-expressed in any cell state.
Gene modules [ Sox2 ] and [ Oct4 ] are highly expressed in both the ES cluster C 0 and the epiblast-like C 1 cluster, after which they are asymmetrically downregulated in the mesendoderm-like C 2 and ectoderm-like C 3. We find that for Although both [ Sox2 ] and [ Oct4 ] are present together in the C 0 and C 1 states, their effective interactions are altered in different ways in each cell state by the presence of other gene modules. As cells transition from state C 0 to C 1 , they downregulate gene modules [ Klf4 ], [ Atf2 ], [ Apex1 ] and [ Ets2 ], and upregulate [ Hes6 ] and [ Otx2 ], among others Figure 4B and C , leading to changes in the effective interaction strength between [ Sox2 ] and [ Oct4 ].
By incrementally increasing [ Sox2 ] levels relative to its base value and assessing the fraction of models that show [ Oct4 ] downregulation, we found that [ Oct4 ] levels are predicted to be more stable to [ Sox2 ] overexpression in state C 0 than in C 1 Figure 4D , thus distinguishing C 0 and C 1 functionally Geula et al. On the other hand, [ Snai1 ] and [ Oct4 ] are not expressed together in any of the nine cell states.
We investigated the predicted effects of [ Snai1 ] overexpression on [ Oct4 ] in the epiblast-like state C 1 and mesendoderm-like state C 2 , both of which normally express [ Oct4 ] but not [ Snai1 ]. Although [ Snai1 ] has a negative interaction with [ Oct4 ] in This leads to the prediction that [ Oct4 ] is less sensitive to [ Snai1 ] overexpression in state C 1 compared to C 2 Figure 4G. We next considered the effect of morphogen signals in different states. We grouped signaling genes by their respective pathways defined by GO categories and assigned each group to a module based on its average expression pattern across the nine cell states.
Because WNT and FGF modules show no changes in expression across all cell states most likely due to the large number of genes that fall into the relevant GO categories , we focused on investigating the effects of LIF and BMP signaling on cells in the epiblast-like C 1 and in the bi-potent ectoderm-like state C 3 Figure 4H. Given an initial state C 1 or C 3 , we calculated the probabilities that cells either remain in the same state or move to a different state in response to [LIF] and [BMP] Materials and methods.
However, in response to the same perturbation, the vast majority of cells in the C 3 state either transitioned to the neural crest-like state C 6 Importantly, we further noted that the model predictions were robust to changes in the probability cutoff for the genes we considered: although the number of gene modules changed 27 modules for a cut off of 0. Thus, by categorizing genes into different modules by their expression patterns across the observed cell states, these modules provide a starting point for modeling the gene regulatory network responsible for cell fate decisions, allowing us to make predictions for how the network gives rise to distinct phenotypic responses to the same perturbation across different cell states.
We transiently transfected cells with a plasmid containing a Tet-inducible bi-directional promoter, flanked by the open reading frames of Sox2 and mCerulean, which we used as a fluorescent reporter of induction Figure 5—figure supplement 1A. We induced overexpression in cells either in the undifferentiated C 0 state or the epiblast-like C 1 state, which correspond to Day 0 and Day 2 of differentiation, respectively Figure 3D , Figure 5—figure supplement 1D. As a control, we used identical populations that were transfected with a plasmid containing only mCerulean under the inducible promoter.
In such experiments, we typically saw mCerulean fluorescence appear approximately three hours into induction and persist for about three to four days after transfection. We therefore induced overexpression for 24 hr to minimize the effect of plasmid loss but still allow for several cell cycles to occur during induction. Following induction, we fixed and immunostained the cells for Oct4, and analyzed the results via flow cytometry. Plots showing mCerulean marker -only overexpression in C 0 or C 1 are indistinguishable from Sox2 overexpression in C 0 Figure 5—figure supplement 1C.
C Comparison of the effects of Snai1 and mCerulean-only left overexpression on Oct4 levels x-axis in the epiblast-like C 1 and mesendoderm-like C 2 states y-axis; T-low and —high, respectively shows downregulation of Oct4 in response to Snai1 overexpression in the C 1 state but not in C 2. D Fraction of Oct4-high cells in Snai1 overexpressing cells, normalized by this fraction in mCerulean overexpressing control cells y-axis , plotted against binned Snai1 overexpression level x-axis confirms the prediction Figure 4G that Snai1 overexpression leads to greater downregulation of Oct4 in C 2 compared to C 1.
F Time series x-axis traces of single-cell Oct4 expression y-axis taken every 15 min from live cells. Each trace is colored by its relative end-point Msx2 immunofluorescence intensity level. G The initial Oct4 reporter mCitrine intensity y-axis and final Msx2 immunofluorescence x-axis are negatively correlated.
Each dot represents a single cell.
Other levels of cell-type-specific regulation of segmentation genes may be imposed on the translation and further metabolism of their protein products; however, studies of the proteins involved have not progressed as far as the RNA studies have. Because WNT and FGF modules show no changes in expression across all cell states most likely due to the large number of genes that fall into the relevant GO categories , we focused on investigating the effects of LIF and BMP signaling on cells in the epiblast-like C 1 and in the bi-potent ectoderm-like state C 3 Figure 4H. Embryonic stem cells Adult stem cells Cancer stem cells Induced pluripotent stem cells Induced stem cells. Article activity alert. The control of direction of cellulose synthesis is achieved by means of cytoplasmic microtubules that lie just inside the plasma membrane of the cells.
Based on this histogram, we defined a range of threshold values for determining Oct4-high and —low shown in overlapping region of orange and green along y-axis. We then tested the effects of Snai1 overexpression on Oct4 in the epiblast-like state C 1 and mesendoderm-like state C 2 , using the same experimental framework as described above.
On day three of differentiation, cell populations either contain a mixture of C 1 , C 2 and minimally C 4 cell states, or a combination of C 1 , C 3 and C 5 or C 6 , depending on the signaling conditions Figure 3D. Using the signaling conditions that yield the former set of cell states C 1 , C 2 and C 4 , we transfected cells at 2. After 24 hr of Snai1 overexpression and further differentiation, we fixed and immunostained the cells for T to distinguish cells in C 1 T-low and C 2 T-high states.
We also immunostained the cells for Oct4 to distinguish the C 1 state from other T-low states that arise during the last 24 hr of differentiation following the initiation of induction.
On the other hand, the fraction of C 2 cells within the transfected population and their Oct4 levels were maintained relative to control, in agreement with our predictions Figure 4G ; Figure 5C and D. We found that 2. We therefore utilized an Oct4 - mCitrine mES cell line that we had previously engineered Thomson et al. By using learned sparse patterns of gene expression from established experimental systems Furchtgott et al. This method naturally identifies a small set of transcription factors whose expression profiles are multimodal across neighboring cell states.
Given that transcription factors are key orchestrators of gene expression and therefore cell fate decisions Spitz and Furlong, , multimodal distributions of the expression levels of even a small set of transcription factors can define cell states in a population of cells. While cell states can be characterized by the gene expression patterns of key sets of genes, these states can only be fully validated by demonstrating distinct physiological properties.
To discover distinct properties of the cell states in early mES cell differentiation, we built probabilistic models of the underlying network. Thus, the cell states we discovered can be functionally defined by their responses to perturbation. Our experimental tests show, as predicted by the model network, that Oct4 is either downregulated or unaffected by overexpression of Sox2 or Snai1 , depending on the cell state.
Previous studies have already shown that Sox2 and Oct4 , along with Klf4 , constitute part of a positive feedback loop that stabilizes the pluripotent ground state Kim et al. It is also known that in undifferentiated cells, Snai1 overexpression leads to downregulation of Oct4 expression and, subsequently, to exit of pluripotency Galvagni et al.
However, our results demonstrate that these interactions are state-dependent by showing that the effective positive interactions between Sox2 and Oct4 become destabilized as Klf4 levels drop and cells transition to a primed, epiblast-like pluripotent state. Similarly, the negative interaction exerted by Snai1 on Oct4 becomes attenuated in the presence of early primitive streak genes such as T.
These results are further supported by the fact that both LIF and BMP signaling pathways can be used to keep cells in the pluripotent cell state Chambers, ; Tam et al. Comprehensive interrogation of gene expression through RNA sequencing is impossible without the termination of cells, providing only static snapshots of gene expression during differentiation. Despite this and the complexity of the underlying network, we discover that both cell states and the sequence of cell state transitions can be accurately determined by monitoring the levels of just a few transition or marker genes.
Monitoring the expression dynamics of these key genes in live cells using microscopy will allow us in the future to continuously track the cell-fate decisions of individual cells. Our inferred cell clusters show mixing of cells from different time points Figure 1—source data 1 , Figure 2—source data 1 , suggesting that the observed states themselves do not change over time and that at the population level, differentiation occurs as a change in the proportions of cells in various cell states rather than through changes in the cell states themselves Figure 3D. Since cells interpret perturbations differently even in consecutive states Figure 5 , this suggests that heterogeneity arising from timing variability is further amplified in response to signal addition or fluctuations in gene expression level.
These findings emphasize the importance of understanding how the timing of cell state transitions is controlled during development. Clustering was performed using Seurat Satija et al. For the initial seed clustering, we applied Seurat to the gene expression of all transcription factors for the single cells. This reduced set contained between and genes at each of the reclustering steps Figure 2—figure supplement 2A.
Seurat performs spectral t-SNE on the statistically significant principal components PCs of the gene expression dataset, and it determines the significance of each PC score using a randomization approach developed by Chung and Storey Chung and Storey, Our initial seed clustering was performed using the first 10 PCs; subsequent re-clusterings used the first 8 PCs.
In order to test that our results were robust to the choice of seed clusters, we further used k-means clustering, a standard clustering method, which has previously been applied to identify different cell types using single-cell transcriptomics data Buettner et al. The number of clusters was determined using the gap statistic Tibshirani et al. In the next iteration, the number of clusters went down to 9, and so on. By iteratively determining the most likely sets of transitions, the corresponding most likely marker and transition genes and re-clustering the cells within the subspace of these genes, our algorithm converged upon the most likely set of cell clusters Figure 2A.
We found that the eventual clustering configurations obtained using k-means clustering and Seurat are the same, confirming that the seed clusters do not affect the final outcome Figure 2—figure supplement 2A, B and C. Classifying genes based on their patterns of expression along the inferred lineage tree rather than by gene-gene correlations allowed us to identify gene modules which included the transition and marker genes we inferred as well as signaling genes: BMP, WNT, LIF, see Tables S4 and S5 with similar expression patterns in successive cell-fate decisions.
We obtained transcription factors from the triplets along the tree and classify them based on their pattern across the triplets. In order to explain the discretization procedure let consider the example of Otx2, which is a transition gene for the triplet involving C 1 , C 2 and C 3 clusters, where C 1 is the intermediate cluster. Since Otx2 is expressed at high levels in cluster C 1 and C 3 and is downregulated in cluster C 2 , we assigned it a value of 1 in clusters C 1 and C 3 respectively and 0 in cluster C 2.
We then repeated this local binarization process across all triplets along the lineage tree. We grouped all the genes that showed the same locally binarized expression pattern as Otx2 and obtained their average expression level across all the other clusters. Some genes, such as Oct4 and Etv5 are re-used at multiple branching points i.
Certain genes that are re-used exhibit three distinct levels of expression. For instance, Sox2 comes up as a marker gene for C 0 cluster, when we consider the triplets involving clusters C 0 , C 1 and C 2 and C 0 , C 1 and C 3 clusters respectively. However, it also acts as a transition gene for the triplet involving C 1 , C 2 and C 3 clusters, where Sox2 is downregulated in C 2.
We have previously shown that DHA acts in a very narrow concentration range i. For immunocytochemical analysis of the cultures, the cells were sequentially fixed with paraformaldehyde, permeated with Triton X for 15 minutes, incubated for 1 hour with primary antibodies and then with either biotinylated or fluorescence-conjugated secondary antibodies, as previously described. The number of amacrine and photoreceptor neurons, the two major cell types in the cultures, was determined by immunocytochemistry with the monoclonal antibodies HPC-1 and Rho4D2, respectively, and according to morphologic criteria previously reported.
Identification of Undifferentiated Cells and Multipotent Neuroblasts. Undifferentiated cells usually bear numerous processes and have either a round or rather irregular morphology. In addition, pluripotent neuroblasts express the neuroectoderm marker nestin, a kDa intermediate filament, whereas they fail to express the characteristic markers of mature retinal neurons. Therefore, the identity of pluripotent neuroblasts in vitro was assessed by several parameters: immunocytochemical determination of nestin, occurrence of mitotic figures, absence of markers of specific neuronal types, such as Rho4D2 and HPC-1, and an undifferentiated morphologic appearance in microscopy.
Most photoreceptor progenitors complete their last mitotic division during the first postnatal days, 40 41 the highest number of photoreceptors being generated at PN0. Several parameters were used to determine cell division of photoreceptor precursors in vitro. Mitotic figures were detected and quantified by fluorescence microscopy by sequentially permeating the cells with 0. Cell labeling with [ 3 H]-thymidine was determined by autoradiographic methods in cultures fixed with glutaraldehyde, as previously described. The number of cells expressing PCNA p36 antigen, a G 1 -S cyclin used as a marker for proliferating cells, was detected by immunocytochemical methods, with a mouse monoclonal anti-PCNA antibody.
The number of cells in each condition was then determined. To evaluate the percentage of progenitor cells committed to exit from the cell cycle, the amount of cells expressing the Cdk inhibitors p27 Kip1 and p57 Kip2 was assessed by immunocytochemistry. Double labeling of 1- and 2-day cultures with DAPI and p27 Kip1 or p57 Kip2 was also performed, to determine whether the expression of these inhibitors occurs during mitosis and whether this expression is equally distributed in both daughter cells or asymmetrically in only one of the cells in the postmitotic pair.
Whenever pairs of progenitor cells undergoing the mitotic telophase or completing cytokinesis were observed, the distribution of labeling in these pairs was also assessed. Unless specifically indicated, each experiment was performed in triplicate. The average number of cells thus obtained was multiplied by a factor to calculate the total number of cells per mm diameter dish. All dishes were counted blindly. At day 1 in vitro, two populations of cells, differing in the size of cell body, were present in the cultures.
These cells had either a round or rather irregular morphology, generally presenting a dark appearance and one or two processes that finally retracted, giving way to the short and usually unique axon characteristic of mature photoreceptor cells Figs. At day 1 in vitro, very few small cells expressed the photoreceptor antigen Rho4-D2 Figs. This expression increased gradually with development Fig. Large cells had a completely different fate, undergoing rapid differentiation. This percentage increased very rapidly until almost all large cells differentiated as amacrine neurons by days 4 to 5.
To further investigate whether undifferentiated retinal cells had the potential to continue in the cell cycle or turn into different cell types, the changes in the expression of nestin, Rho4D2, and HPC-1 during development in vitro were determined. Consistent with their lack of differentiation, more than one third of the cells expressed nestin at day 1 Figs. Nestin localization also changed during development. Initially, it was usually localized in the immature processes, which frequently showed their endings more labeled than their origins, but just before starting differentiation, nestin lost its filamentous structure and tended to concentrate, with a punctate pattern, in the cell body not shown.
Expression of nestin strongly corresponded with an undifferentiated morphology. Generally, cells showing morphologic or immunochemical evidence of differentiation failed to express this protein Fig. Large undifferentiated cells had scarce nestin expression at day 1, which was rapidly replaced by expression of HPC-1, suggesting that they had differentiated as amacrine neurons not shown.
Most nestin-positive cells were small cells. As a rule, the decrease in nestin expression in these cells was parallel to an increase in opsin expression Fig. By day 5, when almost all cells had differentiated, nestin expression was negligible. The pattern of nestin and opsin expression during neuronal development in vitro was similar to their in vivo counterpart.
At PN0 nestin was noticeably expressed in undifferentiated newborn rat retinas, showing a radial pattern of distribution in all layers Figs. This expression gradually decreased during development, although more slowly than in vitro; by PN15 nestin was still detectable with a faint radial distribution compare Fig.
As observed in vitro, the decrease in expression of nestin was parallel to an increase in expression of opsin. Opsin was undetectable at day 0, but was dramatically augmented by PN15 Figs. Therefore, similar changes in protein expression accompanied development in vivo and in vitro, although culture conditions speeded up neuronal differentiation.
GDNF favored proliferation, markedly increasing the percentage of undifferentiated, nestin-positive neuroblasts in comparison with those present in control cultures at days 1 and 2 Fig. On the contrary, DHA promoted cell cycle exit, lowering nestin expression, and simultaneously and significantly increasing opsin expression, compared with controls Fig. Of note, almost no nestin-positive cells were found in any culture condition by day 5.
Several parameters confirmed the existence of an active cell cycle in undifferentiated small cells. In addition, BrdU labeling and [ 3 H]-thymidine incorporation showed that the number of cells in the S phase of the cycle was 27, and 24, cells per dish, respectively Fig. On the contrary, a negligible proportion of large cells underwent cell division at this time of development not shown. To investigate whether GDNF and DHA regulates cell cycle progression in small cells, both trophic factors were separately added to the cultures immediately after the cells were seeded.
GDNF noticeably increased all parameters characteristic of cell cycle progression Fig. Thus, the number of neuroblasts undergoing mitosis almost doubled to 37, cells per dish Fig. DHA supplementation had the opposite effect, reducing by half to 10, cells per dish, the number of neuroblasts undergoing mitosis. Cells labeled with either [ 3 H]-thymidine or BrdU or expressing PCNA decreased similarly, to 13,, 14,, and 7, cells per dish, respectively Fig. Because p57 Kip2 and p27 Kip1 are essential for cells to exit the cell cycle, their expression was determined at different times of development.
No p57 Kip2 expression was detected in photoreceptor progenitors. Expression of p27 Kip1 was more ubiquitous. At day 7, This suggests that p27 Kip1 expression was necessary to arrest the cell cycle and start differentiation in both neuronal precursors. To assess whether GDNF and DHA control the timing of cell cycle withdrawal in multipotent progenitor cells by regulating the levels of p57 Kip2 and p27 Kip1 , their expression was analyzed in cultures separately treated with each trophic factor.
The percentage of amacrine cell progenitors expressing p57 Kip2 or p27 Kip1 was unaffected by the addition of these factors and remained similar in GDNF- or DHA-supplemented cultures to that determined in the respective controls Table 1. Hence, DHA upregulated p27 Kip1 expression, thus prompting photoreceptor progenitors to exit the cell cycle and initiate their differentiation earlier than in control conditions.
During mitosis, the dividing daughter cells may distribute those proteins affecting cell cycle progression—that is, Cdk-cyclin complex inhibitors, in either a symmetrical or an asymmetrical manner. In the later case, the fate of each daughter cell would be quite different, one of them remaining and the other exiting the cell cycle. Hence, the percentage of daughter cell pairs that were completing or had just completed mitosis and expressed p27 Kip1 either symmetrically or asymmetrically was assessed.
A significant amount of the large amacrine progenitor cells symmetrically expressed HPC-1 Fig. Hence, although many small daughter cells were exiting the cell cycle, these results confirmed that a significant fraction of these cells was still in the cell cycle at day 1. The effects of DHA and GDNF on the symmetric or asymmetric distribution of p27 Kip1 expression in small and large daughter cell pairs was consistent with the just described effects of these trophic factors on cell cycle progression Fig. For both neuronal progenitors in every condition studied, cell pairs symmetrically expressing p27 Kip1 were the major fraction.
However, in photoreceptor progenitors, this fraction was significantly reduced in GDNF-treated cultures, when compared with DHA-treated cultures, with a simultaneous increase in cell pairs symmetrically negative for p27 Kip1 expression Fig. This suggests that a fraction of the cells still cycling in GDNF-supplemented cultures were induced to exit the cell cycle by addition of DHA.
Expression of p27 Kip1 during Mitosis. To investigate whether the cell fate of each daughter cell was predetermined before mitosis, mitotic cells were double labeled with DAPI and p27 Kip1. As shown in Figure 8 , at day 1,