|Home | About | Journals | Submit | Contact Us | Français|
Determining the molecular identities of adult stem cells requires novel technologies for sensitive transcript detection in tissues. In mouse intestinal crypts, lineage-tracing studies suggested that different genes uniquely mark spatially distinct stem-cell populations, residing either at crypt bases or at position +4, but a detailed analysis of their spatial co-expression has not been feasible. Here we apply three-color single molecule fluorescence in-situ hybridization to study a comprehensive panel of intestinal stem-cell markers during homeostasis, aging and regeneration. We find that the expression of all markers overlap at crypt-base-cells. This co-expression includes Lgr5, Bmi1 and mTert, genes previously suggested to mark distinct stem cells. Strikingly, Dcamkl-1 tuft cells, distributed throughout the crypt axis, co-express Lgr5 and other stem cell markers that are otherwise confined to crypt bases. We also detect significant changes in the expression of some of the markers following irradiation, suggesting their potential role in the regeneration process. Our approach can enable the sensitive detection of putative stem cells in other tissues and in tumours, guiding complementary functional studies to evaluate their stem-cell properties.
Characterizing the physical locations and molecular identities of stem cells during tissue homeostasis and repair has been impeded by the lack of experimental tools for monitoring individual cells in intact tissue. The mouse small intestine is a prime example in which, despite decades of research, the molecular identities and precise locations of stem cells remain debatable1, 2. The epithelium in the mouse small intestine forms invaginations called crypts that protrude into the underlying connective tissue. Stem cells that reside in the lower parts of the crypts divide to give rise to transit amplifying cells, which rapidly migrate along the crypt axis while dividing a few more times. When the transit amplifying cells reach the upper crypt regions they become post-mitotic and differentiate into either enterocytes – nutrient absorbing cells that form the bulk of the tissue, or several types of secretory cells, including goblet cells, enteroendocine cells and tuft cells3, 4. The differentiated cells continue to migrate up, exiting the crypts towards larger invaginations into the lumen called villi. They are finally extruded from the tops of the villi about 5 days after their birth from stem cells. Paneth cells are longer–lived secretory progenies that migrate down towards the crypt bottoms where they are thought to play a role in crypt defense and stem cell maintanence5.
While it is widely accepted that the intestinal stem cells that give rise to all epithelial lineages reside in the lower portions of crypts, different identities in terms of numbers, exact locations and genetic signatures have been proposed for these stem cells, that appear mutually exclusive1, 2. The “+4 hypothesis”, originally proposed by Potten6 posits that stem cells reside in cell position +4, just above the Paneth cells. This is based on unique characteristics of cells at these positions, including their high susceptibility to apoptosis, their non-random DNA strand segregation and suggested specific expression of genes such as Bmi17, 8, mTert9 and Dcamkl-110. Alternatively, the stem cell zone hypothesis originally formulated by Leblond11, 12 posits that crypt base-columnar-cells (CBC) residing at the very bottom of the crypts are the actual stem cells. While independent lineage tracing studies using Lgr512, Sox913 and Prominin-114 have demonstrated stable labeling of the progenies of CBC cells, and a single Lgr5-high stem cell has been shown to reconstitute a long-lived and complete, self-renewing small intestinal organoid in vitro15, lineage tracing with Bmi17, 8 and mTert9 has implied the +4 cell as the stem cell of the small intestine. These results pose the question of whether two or more distinct stem cell populations uniquely marked by these genes co-exist in mouse intestinal crypts1, 2.
Lineage tracing experiments provide functional proof that a gene of interest is expressed in stem cells, but are limited in detecting the precise location of the expressing cells and the expression pattern of other genes in these cells. Genes that are broadly expressed throughout the tissue in both stem cells and in their differentiated offspring would yield stable labeling of progenies, but would not be informative as to the location of stem cells and could not, on their own, be considered stem-cell markers. Thus detecting stem cell genes in mammalian tissues requires complementing lineage-tracing studies with sensitive methods to measure the precise location where candidate markers are expressed and to determine their co-expression patterns.
Previous attempts to characterize this co-expression program were based on methods such as qPCR or microarray analysis of GFP-sorted cell populations15, 16 or laser-capture microdissected tissue17. While yielding important insights, these methods have several disadvantages, such as the use of knock-in mice, standardization issues related to the qPCR process, insufficient sensitivity for the analysis of single cells and most importantly the loss of spatial information18. Immunohistochemistry and classic RNA in-situ hybridizations19 preserve tissue morphology, but sensitivity and specificity problems limit the generic use of these methods in yielding quantifiable co-expression data of several genes at the single cell level. To overcome these limitations, several studies used multiply labeled fluorescent probes to detect single mRNA in fixed yeast and mammalian cells20–23 as well as nuclear transcription sites in paraffin-embedded tissue24. However detection of single mRNA in adult mammalian tissue, where single-cell resolution is crucial for identifying the distinct roles of individual cells, has not yet been demonstrated.
We have previously developed a sensitive method of transcript counting based on singly labeled fluorescent probes25, enabling simultaneous detection of three different endogenous transcripts in individual cells. This technique was successfully applied to study expression in mammalian cells, as well as in Drosophila25 and C. elegans embryos26. Here we apply this method to mouse intestinal frozen sections, to obtain, for the first time, a quantitative comprehensive in-situ description of the spatial patterns and combinatorial expression of stem cell markers at the single transcript level.
We designed a panel of 15 libraries of fluorescently labeled probes, each composed of 48 20-bp oligos complementary to the coding sequences of previously suggested stem cell markers (Fig. 1). These included the R-spondin receptor Lgr512, 27, 28, the WNT targets Ascl216, CD4429, Sox430, Sox913, Mmp7, EphB2 and EphB331, the RNA binding protein Musashi-132, 33, Olfactomedin-4 (Olfm4)16, Prominin-1 (CD133)14, Dcamkl-110, 17, Bmpr1a34, mouse telomarse reverse transcriptase (mTert)9, 35 as well as the polycomb gene Bmi17. Hybridization of 6-micron cryo-sections of small intestinal tissue with these libraries yielded bright diffraction limited dots, representing single transcripts (Fig. 1). These were automatically counted using custom image processing software (Fig. S1a–d). To study the co-expression of these genes at the single cell level we used three different fluorophores to simultaneously probe the expression of Lgr5, Bmi1 and other genes from the panel and assigned their numbers to individual cells manually segmented based on E-cadherin lateral membrane staining.
We first assessed whether our transcript counting method correlates with the expression patterns in reporter mice. To this end we examined both fluorescence and transcript levels in the Lgr5-EGFP reporter mouse model12. We detected cells with intense GFP signal, as well as EGFP transcripts at crypt bottoms in only one out of ten crypts on average, consistent with the pronounced variegated expression pattern previously reported36 (Fig. 2a). Unlike the patchy expression of the transgene we uniformly detected the endogenous Lgr5 transcripts in every crypt throughout the tissue. Importantly the expression level of both Lgr5 and EGFP transcripts, as well as GFP levels were highly correlated in the crypts that were positive for both (Fig. 2a,b, Spearman correlation R = 0.68, p <10−68). Thus our method is highly correlated to the transgene transcript and protein levels, but facilitates a much more comprehensive analysis of the tissue. To further test the specificity of our method we analyzed the expression of the intestinal differentiation markers Gob5, Creb3l3 and Lysozyme and the proliferation marker Ki67. This yielded highly localized expression at the respective goblet, enterocyte, Paneth, and transit-amplifying cells, demonstrating the specificity of the technique (Fig. 2c,d).
To facilitate our analysis of the expression patterns of the putative stem cell marker genes along the intestinal crypt, we created a spatial profile for each gene by projecting the single cell transcript counts on a vertical axis originating at the crypt apex. We found that the spatial expression profiles are remarkably invariant between crypts within the same mouse and almost identical between 4 months and 11 months old mice (Fig. 3a). The genes clustered into two groups (Fig. 3a, Fig. S2c) - the expression of Lgr5, Musashi-1, Ascl2, Sox4, Sox9, CD44, Olfm4 and EphB3 was concentrated at crypt bottoms, leveling off towards the upper crypt positions. In contrast, Bmi1, Prominin-1, Bmpr1a and mTert exhibited a broad expression pattern that was nearly constant throughout the crypt axis (Fig. 3a, Fig. S2c). Notably, all genes for which stable progeny labeling using lineage tracing has been demonstrated were broadly co-expressed in CBC cells at lower crypt positions. More than 75% of Lgr5-positive cells contained Bmi1 transcripts and almost half contained transcripts of mTert (Fig. 3b–e). This co-expression indicates that Lgr5, Bmi1 and mTert do not mark distinct stem cell populations coexisting within a crypt. Our measurements can therefore explain the seemingly contradictory previously published results that demonstrated stable lineage tracing of progenies of cells expressing either of these genes7, 9, 12.
To infer the regulatory connections between the studied markers and to detect whether they are expressed in mutually exclusive cells, as has been suggested for Lgr5 and Bmi11, 2, and Lgr5 and mTert9 we calculated the single-cell correlation coefficients of pairs of genes (Fig. 4). Gene pairs that are highly correlated could be regulated by a common upstream gene or directly regulate each other, whereas pairs that are not correlated are predicted to belong to different regulatory modules. Significant negative correlation of genes would indicate that they tend to be expressed in mutually exclusive cells.
We found that some gene pairs such as Ascl2 and Musashi-1 were highly correlated (R = 0.7, p < 10−16, Fig. 4a) whereas others, such as Bmi1 and Ascl2 were expressed in a non-coordinated fashion (R = −0.05, p = 0.74, Fig. 4b). We next measured our panel in mutants for the two main regulators among the studied genes – a knock-out mouse of the polycomb gene Bmi1 (Fig. 4c), and a conditional knock-out of the transcription factor Ascl216 (Fig. 4d). The duodenum in Bmi1 knock-out mice was histologically similar to that in controls, as previously reported16. We found that the higher the single-cell correlations between pairs of genes in the wild-type mice, the higher the expression reduction in the respective mutants (Fig. 4e–g, R = 0.76, p = 0.0045). Thus positive single-cell correlations between pairs of genes are indicative of regulatory connection between them. Lgr5 and Bmi1 did not exhibit significant correlation regardless of the cell position along the crypt axis from which cells were sampled (Fig. 3b R = −0.025, p = 0.9) and they exhibited significant positive correlations with a mutually exclusive subset of markers (Fig. 4h). Thus our analysis indicates that Lgr5 and Bmi1 are broadly co-expressed in CBC cells but that they do not affect each other's expression and belong to different regulatory modules. mTert was also broadly co-expressed with Lgr5 and these markers exhibited a slight positive correlation (Fig. 3c,e Fig. 4h, R=0.13, p=0.002).
A unique expression pattern was exhibited by Dcamkl-1. Unlike the broad expression patterns of the other stem-cell markers studied, we found that Dcamkl-1 transcripts were strongly concentrated in isolated cells appearing once every 5 crypt sections (Fig. 5). These cells were widely distributed from lower crypt positions to villi (Fig. S4a) and specifically co-expressed the tuft cell marker Cox14 (Fig. 5a,b). Strikingly, Dcamkl-1 cells at all crypt positions significantly co-expressed stem cell markers that were otherwise confined to crypt bottoms. These included Lgr5 (median expression ratio with neighboring cells of 4.99, p < 10−16, Fig. 5a,c, Fig. S4b) and Sox9 (median ratio of 4.9, p < 10−16, Fig. 5c). Other genes that were significantly expressed in Dcamkl-1 cells were Musashi-1 (Fig. S4c) EphB2 and EphB3 (Fig. 5c, Fig. S4f). While only a relatively small fraction of Dcamkl-1 cells at the transit amplifying compartment exhibited Lgr5 expression comparable to the Lgr5 expression in CBC cells (12%), the appearance of Lgr5 above the crypt base was confined to Dcamkl-1 cells (Fig. S4d).
We next asked if the enrichment of stem cell genes in Dcamkl-1 cells represents residual transcripts or rather a regulated expression signature (Fig. 5d). If Dcamkl-1 cells are quiescent and migrate very rapidly transcript decay would be slower in these cells. Indeed, we found that Dcamkl-1 were depleted in Pcna and Ki67, suggesting quiescence37 (Fig. 5c. Fig. S6a). However these cells were enriched for EphrinB1, EphB2 and EphB3 (Fig. 5c, S4e,f), the expression of which has been shown to correlate with slower rather than faster migration rates38. In addition, only a subset of stem cell genes were enriched in Dcamkl-1 cells, whereas others, such as Olfm4 and CD44 were not (Fig. 5c, Fig. S4g). Unlike Lgr5, Olfm4 and CD44 transcripts did not spatially decay more slowly in Dcamkl-1 cells (Fig. 5e–f). Taken together, these findings indicate that some Dcamkl-1 cells, exhibit a regulated expression signature that includes stem-cell markers, which are otherwise confined to crypt bottoms.
Enrichment of Lgr5 and other stem cell markers in Dcamkl-1 cells could potentially implicate Dcamkl-1 cells as reserve stem cells. To address this possibility we repeated our single molecule transcript counting measurements on intestines of mice at different time points following whole body irradiation with 1 Gy, 6 Gy and 12 Gy (Fig. S5). These perturbations have been shown to cause a massive cell death followed by regeneration in intestinal crypts39. Indeed, irradiation with 12 Gy yielded a massive reduction in the number of crypts and their sizes, a phenomenon most prominent 48 hours after irradiation (Fig. S5b). 7 days after irradiation, we observed an increase in crypt sizes and extensive crypt fissions (Fig. S5c).
We found that Dcamkl-1 cells did not enter cell-cycle following irradiation, as apparent by their low Ki67 expression (Fig. S6a,b). The dynamics of Dcamkl-1 cell numbers closely followed that of the short-lived differentiated goblet cells, exhibiting a decrease up to 48 hours after irradiation, followed by an increase at 7 days (Fig. S6c). In addition Dcamkl-1 cells did not exhibit increased death rates 6 hours and 24 hours after 1 Gy irradiation, as detected morphologically, regardless of whether they had Lgr5 transcripts (Fig. S6d). Taken together these results do not support the possibility that Dcamkl-1 cells serve as reserve stem cells.
To obtain a comprehensive view of the expression changes that occur following irradiation, and to detect genes among our panel that could be functionally important for the tissue repair process we also measured the entire panel at different time points after 12 Gy whole body irradiation. We found striking differences in the spatial expression patterns of some stem-cell markers 48 hours and 7 days after 12 Gy whole body irradiation relative to non-irradiated controls. These differences included a considerable expansion in both the spatial range of expression and levels of some stem cell markers (Fig. 6, Fig. S7). Most notable among these markers are CD44 (Fig. 6a,d,e), Musashi-1 (Fig. 6c,d,e) and Ascl2 (Fig. 6c,d,e). Interestingly, both the levels and spatial range of Olfm4 first decreased after 48 hours, and then significantly increased at 7 days (Fig. 6b,d,e). While the average spatial ranges of Lgr5 and Bmi1 slightly expanded following irradiation (Fig. 6d, Fig. S7a,d), their transcript levels did not change significantly (Fig. 6e).
Revealing the molecular identity of stem cells in the mouse intestine has been impeded by lack of sensitive in-situ expression measurements. Here we applied single molecule transcript counting to establish a comprehensive database of expression patterns in the mouse intestine and demonstrated that these measurements can shed light on stem-cell identities during homeostasis, aging and repair.
Our study revealed broad spatial expression profiles for three of the five genes for which stable lineage tracing of progenies has been demonstrated in the mouse intestine – Bmi17, Prominin-114 and mTert9. These were expressed throughout the crypt axis at almost constant levels, and contrasted with Lgr512 and to a slightly lesser extent Sox913, the expression of which were concentrated at lower crypt positions. Importantly, all five genes were co-expressed in crypt base-columnar-cells11. Thus Bmi1, Prominin-1 and mTert, while clearly expressed in stem cells, do not on their own specifically mark intestinal stem cells. These results emphasize the importance of sensitive in-situ transcript detection in mammalian tissue as a complementary approach to lineage tracing in determining the precise location in which candidate stem cell markers are expressed. While previous studies showed co-expression of Lgr5 and Bmi1, as well as mTert by comparing expression between fractions of dissociated low and high Lgr5-GFP cells16, 35, our measurements assess these co-expressions in a symmetric manner at the single cell level in WT mice and indicate the precise location of the cells co-expressing these stem cell markers (Fig. S3a). It should be stressed however that our analysis does not imply that all crypt cells that express both Bmi1 and Lgr5 have equal stem cell potential.
We detected a unique expression signature for Dcamkl-1 cells, which includes significant co-expression with Lgr5. Dcamkl-1 has recently been shown to be a marker of tuft cells, a rare quiescent epithelial lineage of unknown function4, 40. We found that regardless of their Lgr5 expression, Dcamkl-1 cells do not exhibit increased death rates following low dosage of gamma irradiation, as previously suggested for putative stem cells at higher crypt positions6, 39. Following high dosage of gamma irradiation these cells did not enter cell cycle at any time point and were depleted in proportion to goblet cells, a short-lived differentiated secretory cell type. Most importantly, all Dcamkl-1 cells, both positive and negative for Lgr5, exhibited intense expression of the Cox1 gene, a tuft cell differentiation marker4. While Lineage-tracing utilizing a Dcamkl-1-locus driven Cre transgene would definitely resolve the possibility that some tuft cells could posses potential stem cell function, our analysis suggests that such function is unlikely.
The appearance of transcripts at higher crypt positions should not necessarily imply that active proteins are present and may simply represent residual transcripts decaying slower than the rates at which cells migrate. We found however that the expression profile of genes such as Ki67 and Creb3l3 exhibited a dramatic change in levels over one vertical cell position at the crypt-villus borders (Fig. 2d), suggesting that transcript decay rates in intestinal crypts are faster than cell migration rates. Transcript levels detected by our method were also highly correlated with protein levels detected using GFP (Fig. 2a).
Our analysis indicates that during homeostasis the expression patterns of stem cell markers are remarkably invariant between crypts within the same mouse and with aging, with several markers such as Lgr5,Olfm4,CD44,Ascl2 and Musashi-1 exhibiting spatially overlapping expression patterns and high single-cell correlations. The expression program of these genes is however markedly different when the tissue is perturbed. This is evident from the dramatic expansion in range and numbers of Ascl2,Musashi-1 and CD44 transcripts following irradiation, which contrasts with the almost constant levels of Lgr5 and Bmi1, and the more intricate behavior of Olfm4 expression pattern, which first retracts and then expands. These varying responses observed following perturbation are indicative of potential functional differences among the stem cell markers in damage repair.
Our transcript-counting method should be considered as a complementary approach to protein-expression assays as well as to functional techniques such as lineage tracing7, 12, cell ablation8 and ex-vivo cultures15. Our method can be combined with these functional methods in two ways. One would be to use lineage tracing or ex vivo cultures to first detect potential stem-cell markers. Our method can then be applied to characterize in detail the spatial co-expression patterns of these markers in wild-type tissue. Alternatively, unbiased gene-expression measurements using a panel of single-molecule FISH probes could detect potentially interesting gene-expression signatures in terms of spatial distribution in a tissue or an unusual co-expression pattern of a few genes in isolated cells. These genes could then be followed up with other techniques to assess the functional importance of these gene-expression signatures.
The homeostasis of epithelial tissues is based on a complex expression program, controlled by niche-dependent signals, as well as intracellular transcriptional and signaling networks. Here we have shown that single molecule transcript counting combined with computational approaches can yield a detailed characterization of the spatial expression profiles and the single cell co-expression patterns of key genes, as well as the changes during aging and tissue regeneration. Applying this technique to other tissues maintained by stem cells can provide important insights into the architecture of multi-cellular organisms, while similar studies in tumors can facilitate the detection of stem cell like signatures in cancer.
All animal studies were reviewed and approved by the Massachusetts Institute of Technology (MIT) Committee on Animal Care. Duodenum tissue was harvested from C57bl6 female mice at ages 4 months and 11 months. Each age group included between two and five mice. Bmi1 knock-out experiments were performed on 11 week old female Bmi1−/− and a Bmi+/+ littermate control41. These mice were APCfloxed/+, but since the mice had no Cre recombinase both were essentially wild-type for APC. The Ascl2 mutant was an 18 weeks old male Ah-Cre/Ascl2floxed/floxed, 5 days after induction of the Cre enzyme by intraperitoneal injections of 200 ml b-napthoflavone, as described in16. A non-induced Ah-Cre/Ascl2floxed/floxed littermate control was used for comparison. Whole body gamma-irradiation dosages of 1Gy, 6Gy and 12Gy were applied to 4 months old C57bl6 mice as described in42. Mice were sacrificed after 6 hr, 24 hr, 48 hr and 7 days. Two mice per irradiation dosage and sacrification time were used. The Lgr5-EGFP mouse used (Fig. 2a,b) was described in12. All mice were fed ad libitum and sacrificed in the morning. For all mice duodenum was removed, fixed in 4% Formaldehyde, then incubated overnight with 30% sucrose in 4% formaldehyde and then embedded in OCT. 6-micron cryo-sections were used for hybridizations.
Probe libraries were designed and constructed as described in Raj et al.25. Most libraries consisted of 48 probes of length 20bps, complementary to the CDS of each gene (Supplementary table). Lgr5 and Ki67 libraries consisted of 96 probes. Hybridizations were done overnight with three differentially labeled probes using Cy5, Alexa594 and TMR fluorophores. An additional FIT-C conjugated antibody for E-cadherin (BD Biosciences) was added to the hybridization mix and used for protein immunefluorescence. DAPI dye for nuclear staining was added during the washes. Images were taken with a Nikon Ti-E inverted fluorescence microscope equipped with a 100× oil-immersion objective and a Photometrics Pixis 1024 CCD camera using MetaMorph software (Molecular Devices, Downington, PA). The image-plane pixel dimension was 0.13 microns. Quantification was done on stacks of 6‒12 optical sections with Z-spacing of 0.3 microns, in which not more than a single cell was observed.
Dots were automatically detected using a custom Matlab program, implementing algorithms described in Raj et al25. Briefly, the dot stack images were first filtered with a 3-dimensional Laplacian of Gaussian filter of size 15 pixels and standard deviation of 1.5 pixels. The number of connected components in binary thresholded images was then recorded for a uniform range of intensity thresholds and the threshold for which the number of components was least sensitive to threshold selection was used for dot detection (Fig. S1a–d). Automatic threshold selection was manually verified and corrected for errors (<5% of crypts). Background dots were detected according to size and by automatically identifying dots that appear in more than one channel (typically <1% of dots) and were removed. Such dots occasionally appeared in the surrounding mesenchymal cells but were rare in the epithelial cells. Bleed through of transcript signal between channels was minimal (Fig. S1e–g). Cell segmentation was performed manually on a maximal projection of the FIT-C channel. Cells at the lower 15 positions of the crypt were typically segmented. Transcript concentrations were obtained by dividing the number of transcripts per cell by the cell volume estimated as the product of the segmented area and the number of vertical stacks times a voxel size of 0.13um *0.13um *0.3um. Crypt apex and outline were manually marked and used to determine cell position along the crypt axis. For four genes – Olfm4,Dcamkl-1,Gob5 and Lysozyme, transcript abundance was too high in some of the cells to facilitate reliable dot counting. In these cells the cytoplasm was often uniformly fluorescent (Fig. 2c Fig. S4g). Thus for these genes our dot counting algorithm underestimated the number of transcripts per cell.
Spatial profiles were symmetrized by averaging identical cell positions on both sides of the crypt apex, smoothed over 3 crypt positions and normalized to a maximum of 1. Co-expression analysis was performed on pooled cells from all crypts. Transcript concentrations were first normalized by the mean for each crypt, to correct for possible variations in hybridization or imaging conditions. Cells with no transcripts were assigned the lowest transcript concentration measured in the mouse. Spearman correlation coefficients of pairs of genes were compared to those obtained in randomized crypts in which the values of one of the genes was shuffled among cells, and p-values reported were computed by transforming the Z-score of the correlations compared to those in randomized crypts using the normal distribution.
Kolmogorov-Smirnov tests were used to generate p-values for the comparison of expression distributions. To generate the Dcamkl-1 single cell expression signature we computed for each Dcamkl-1high cell the ratio between the transcript concentration in the cell and its immediate neighbors that were not Dcamkl-1high. Dcamkl-1high cells were defined as cells with more than 5 Dcamkl-1 transcripts; similar results were obtained for other thresholds. P-values for the median ratios were computed by creating randomized datasets in which the transcript concentrations of the gene of interest were arbitrarily swapped between Dcamkl-1high cells and one of the neighboring cells and ratios were recalculated. Z-scores for the median ratios were transformed to p-values based on a normal distribution. When estimating the fraction of Dcamkl-1high cells that were positive for Lgr5 (Lgr5+) we applied a threshold equal to the median of the Lgr5 expression at cells at or below cell position 5 that had at least one Lgr5 transcript. Transcript spatial decay rate was computed by linear regression of the logarithm of the transcript concentration vs. crypt position.
The authors would like to thank H. Youk, S. Semrau, S. Klemm and K. Hilgendorf for comments on the manuscript, and X. Wu, Z. Peng Fan and A. Yang for help with the cell segmentation software. This work was supported by the National Institutes of Health (NIH)/National Cancer Institute Physical Sciences Oncology Center at MIT (U54CA143874) and an NIH Pioneer award (1DP1OD003936) to A.v.O., and in part by Cancer Center Support (core) grant P30-CA14051 from the National Cancer Institute. S.I. acknowledges support from a European Molecular Biology Organization postdoctoral fellowship, the International Human Frontiers Science Program Organization and the Machiah Foundation. I.C.B. acknowledges support from the Howard Hughes Medical Institute Gilliam fellowship. T.J. is the D. H. Koch Professor of Biology and a D. K. Ludwig Scholar.
AUTHOR CONTRIBUTION S.I. and A.vO. conceived the project. S.I., I.B. and A.L. carried out most of the experiments. S.I. analyzed the data. M.M., J.L., J.vE., T.J. and H.C. provided mice and assisted with experiments. S.I. and A.vO. wrote the paper.
COMPETING FINANCIAL INTERESTS STATEMENT The authors declare that they have no competing financial interests.