Chromatin Reprogramming via Contact Guidance-Induced Nuclear Deformation Promotes Stem Cell Differentiation

Ecient manipulation of cell fate is important for regenerative engineering applications. Lineage-specic differentiation of stem cells is particularly challenging due to their inherent plasticity. Engineered topographies may alter cellular plasticity through contact guidance. However, the ability to rationally design topographies to regulate phenotypic outcomes has been hindered in part by the lack of tools to quantify nanoscale chromatin structure reorganization in live cells. Herein we use micropillars, molecular, and nanostructural quantication tools to investigate how nuclear morphology in human mesenchymal stem cells (hMSCs) affects chromatin conformation and osteogenic differentiation. We show that micropillar-induced contact guidance is transduced via the cytoskeleton and impacts nuclear architecture, lamin A/C multimerization, histone modications, and the 3-D conformation of chromatin within packing domains, a key regulator of transcriptional responsiveness. Micropillars repressed expression of genes associated with developmental processes and enhanced lineage-specic responsiveness, thereby decreasing cell plasticity and off-target differentiation, and facilitating osteogenic differentiation of hMSCs. Altogether, these ndings reveal that chromatin reprogramming through contact guidance-induced nuclear deformation can be an ecient way to manipulate cell fate. was mixed with the homogenized to generate a uorescent signal due to its cleavage by ALP. The uorescence intensity was read by a cytation 5 imaging reader (BioTek) at (Ex/Em = 360/440 nm). The enzymatic activity was calculated based on the standard curve and normalized to total DNA content tested by the Quant ‐ iT PicoGreen dsDNA assay (Invitrogen). Four biological replicates were used to calculate statistics. The expression of ALP and IBSP of hMSCs on at and micropillar substrates was evaluated using quantitative real-time reverse transcription-polymerase chain reaction (RT-qPCR) after 7d induction. The total RNA of the cells was extracted using the Aurum Total RNA Mini Kit (bio-rad) according to the protocol. The concentration and purity of the extracted RNA were tested using the citation 5 imaging reader. The RT-qPCR was carried out using iTaq Universal Sybr Green One-step Kit following the vendor’s protocol. The designed primers for ALP are: Forward, 5’-GACCCTTGACCCCCACAAT-3’; Reverse, 5’-GCTCGTACTGCATGTCCCCT-3’. The designed primers for IBSP are: Forward, 5’-TGCCTTGAGCCTGCTTCC-3’; Reverse, 5’-GCAAAATTAAAGCAGTCTTCATTTTG-3’. We employed GAPDH as the house-keeping gene. The GAPDH primers are: Forward, 5’-GTGGACCTGACCTGCCGTCT-3’; Reverse, 5’-GGAGGAGTGGGTGTCGCTGT-3’. The data was analyzed using the 2 -ΔΔCt method. The expression of target genes was rstly normalized to that of GAPDH, and then to the average values on at substrates. Three biological replicates were used to calculate statistics. Calcium deposition were stained with Alizarin Red S on both surfaces after 3 weeks of induction. The positively stained area in the bright led images was analyzed using ImageJ software.


Background
Nuclear morphology is regulated by nuclear structure components such as lamins and chromatin, as well as cytoskeletal proteins. 1 Although not fully understood, studies have revealed that mammalian cells can modulate their nuclear morphology to adapt and acclimate to their microenvironments through the mechanotransduction process. [2][3][4][5] Usually, the nucleus is considered to be of spherical or ovoid shape, which is true for many types of cells. However, severe changes in nuclear morphology are also observed in various physiological processes such as malignant cell invasion, 6 smooth muscle cell contraction, 7 stem cell homing, 8 and embryo development. 9 As a cellular mechanosensor, changes in nuclear morphology are considered to directly affect chromatin reprogramming and genome functions that determine cell fate. 1 Contact guidance-induced nuclear deformation, similar to what occurs in vivo, can be reproduced in cells cultured on micro-and nano-pillar substrates. 10,11 The resultant nuclear deformation has been shown to affect phenotypic outcomes in stem cells such as proliferation and differentiation. [12][13][14] Maximizing cell differentiation in cell culture and at tissue-implant interfaces is a major goal in many elds including tissue and regenerative engineering, surgery, and biological sciences. 15 However, the relationship between the nuclear deformation induced by such topographies and phenotypic outcomes remains elusive. Additionally, direct evidence for contact guidance-induced chromatin reprogramming in live cells during such processes is limited. 16 In this study, nuclear morphology changes were induced via contact guidance on micropillars to investigate their in uence on osteogenic differentiation of human mesenchymal stem cells (hMSCs). We hypothesized that mechanical constriction of the nucleus will lead to chromatin reprogramming, and as a consequence, modulate the transcriptional plasticity in stem cells to improve the e ciency of lineage-speci c differentiation. Previously, we have demonstrated that the physical structure of chromatin packing regulates genome-wide transcriptional patterns by altering the kinetics of transcriptional reaction through macromolecular crowding-mediated effects exerted by chromatin density that is inter-related with the local chromatin nanoenvironment. 17 We uncovered that chromatin exhibits length-scale invariant chromatin packing scaling behavior within chromatin packing domains, with sizes on the order of 100s of nm. 18 Speci cally, we have identi ed the chromatin packing density scaling of packing domains, D, as an important statistical descriptor of chromatin behavior and transcriptional plasticity.
From a polymer physics de nition, D de nes the power-law scaling relationship between the genomic length of a polymer and the space it occupies in three-dimensional space. Furthermore, combining the molecular and physical regulators of transcription, the chromatin packing macromolecular crowding (CPMC) model predicts the effect of average packing domain D on global patterns of gene transcription. The model shows that an increase in D increases both the accessible surface area of chromatin, which determines the probability of genes being accessible to transcription factors, and the heterogeneity of crowding conditions within a given transcriptional interaction volume. 17 Therefore, given that D is one of the major predictors of global gene expression, we hypothesized that the differentiation outcomes in hMSCs can be modulated by potentially altering this physical property of chromatin. To address this hypothesis, we integrated surface topography engineering of biomaterials, nanoscale imaging, cell, and molecular biology to investigate the in uence of severe nuclear deformation in hMSCs on chromatin reprogramming and transcription, as well as osteogenic differentiation.

Results
Micropillars Manipulate Nuclear Morphology. In order to study the implications of mechanical constriction of the nucleus on stem cell fate, micropillar structures were fabricated using methacrylated poly (1,8-octanediol-cocitrate) (mPOC) via contact printing (Fig. S1A). A variety of parameters including pillar size, shape, and space were controlled to investigate their effects on nuclear morphology (Fig. S1B). All the pillar structures had a height of 8 μm which is su cient to cause deformation in the nucleus. 13 Nuclear shape index (NSI) was analyzed to quantify the effect of micropillars on nuclear morphology. 19 We found that spacing of micropillars had the most obvious in uence on nuclear morphology as decreasing the spacing resulted in more severe deformation of the nucleus (Fig. S1C, D). The shape of micropillars also had a signi cant in uence on nuclear morphology due to the differences in the curvature of the microfeatures. Herein, we found that micro-square pillars with a size and spacing of 5 x 5 μm had the most signi cant effect on the deformation of hMSC nuclei, which was used for the following studies ( Fig. 1A-D).
Cytoskeleton Mediate Nuclear Deformation. In addition to nuclear morphology changes, cell morphology was also altered by surface topography. hMSCs on the at surface showed normal broblast-like spindle shape and formed well-organized cytoskeletal structures; whereas those cultured on micropillars showed elongated cell morphology with a weak assembly of the cytoskeleton ( Fig. 1E and Fig. S2A). To investigate the in uence of the cytoskeleton on nuclear deformation, hMSCs were treated with 1 μM Latrunculin A, 4 mM acrylamide, and 1 μM colchicine to interfere with the assembly of F-actin, intermediate lament, and microtubule, respectively. After 6 hours of treatment with latrunculin A and acrylamide, a signi cant increase in NSI was observed, while there was no signi cant change in NSI after treatment with colchicine indicating the involvement of F-actin and intermediate laments in nuclear morphology regulation (Fig. S2B, C). 20 Additionally, cytoskeleton mediated mechanotransduction processes were also affected by micropillars (Fig. S2D, E).
Micropillars Manipulate 3D Nuclear Architecture and Mechanical Structure. To further determine the in uence of micropillars on nuclear architecture, we investigated the 3D structural changes in the cell nucleus. Cell nuclei on at surfaces exhibited a atter, 'pancake' shape while nuclei on micropillars were squeezed in-between the pillars with a deformed irregular morphology via contact guidance (Fig. 1F). The projected area of the nucleus on micropillars was signi cantly smaller, but the height was larger (Fig. S3A, B). According to the 3D reconstruction of nuclei, the nuclear volume and surface area of the deformed nucleus were signi cantly smaller than the at condition which was attributed to the constraints imposed by the micropillars (Fig. 1G, H). Additionally, the ratio of surface area to volume remains unchanged indicating a more compact shape (Fig. S3C).
Nuclear lamins, especially lamin A and C have been reported to play critical roles in regulating nuclear structure and mechanics. 21 Recent studies revealed that cell culture substrates affect the polymerization of lamin A/C, which in turn in uences the polarization of nuclear architecture and epigenetically regulates cell functions. [22][23][24] Lamin A/C staining on at and micropillar surfaces showed a distinct pattern of nuclear lamins in both horizontal and vertical directions. Horizontally, the lamin A/C was isotropically stained on at surfaces (Fig. 1I, J and Fig.  S3D). However, it was anisotropically distributed in the cell nuclei on micropillars with high intensity of lamins observed at nuclear protrusions and low intensity observed at the nuclear envelope (NE) around micropillars. In addition, distinct lamin A/C wrinkles were observed surrounding micropillars which might be caused by the severe curvature of the cell nucleus at these regions. Vertically, the lamin A/C was anisotropically stained between apical and basal NE in the cell nucleus on a at surface but was isotropically distributed on micropillars (Fig. 1K). To take into account the in uence of absolute intensity variations between samples, the ratios of relative uorescence intensities between the regions around micropillar and nuclear protrusions, as well as basal versus apical NE were quanti ed from xy-plane and xz cross-sections, respectively (Fig. 1L, M). This quanti cation con rmed the distinct distribution pattern of lamin A/C on at surfaces and micropillars, which might be due to the different cell adhesion geometry and cytoskeleton assembly regulated via contact guidance (Fig. 1N). 22 Despite the different distribution patterns, the lamin A/C showed similar expression levels on both at and micropillar substrates as  On at surfaces, actin-caps that formed above the nucleus compress it and the cell adhere to the substrate beneath which leads to burial of the epitope in polymerized lamin A/C at basal NE. On micropillars, the pillar structures prevented the expansion of the nucleus and led to its compression, and provided adhesion with cells that induced multimerization of lamin A/C around micropillars and induced horizontal polarization of NE. (***p<0.001).
Micropillars Affect Histone Modi cations. Although the mechanism remains elusive, growing evidence indicates that biophysical cues can affect the epigenetic state of the cell via the regulation of histone modi cations. [25][26][27] Therefore, we investigated whether nuclear deformation could lead to alterations in histone acetylation and methylation that are the major types of modi cations in uencing transcription. After 3 days of culture in maintenance medium, we rstly stained a variety of histone acetylation changes including acetylation of H3 at lysine 9 (H3K9ac), 14 (H3K14ac), 18 (H3K18ac), and 27 (H3K27ac) (Fig. S4A). Most nuclei showed decreased uorescence intensity upon nuclear deformation indicating decreased whole-nuclear histone acetylation. A decrease in histone acetylation was further con rmed by detection of global histone H3 acetylation (H3Ac) that was previously reported to affect differentiation of hMSCs ( Fig. 2A-C). 28,29 We then probed for changes in markers of active gene transcription, including methylation of H3 at lysine 4 (H3K4me2) and 36 (H3K36me2 and H3K36me3) and repressive markers including methylation of H3 at lysine 9 (H3K9me3) and 27 (H3K27me3) (Fig.  S4B). The majority of the markers remain unchanged between control and deformed nuclei except for H3K27me3, which presented with higher uorescence intensity and protein expression in cell nuclei on micropillars compared to those on at surfaces ( Fig. 2A-C). Altogether, the global increase in H3K27me3 (repressive) and a decrease in H3Ac (active) in hMSCs cultured on micropillars suggests that such contact guidance may cause transcriptional repression, a response similar to cells undergoing mechanical loading. 30,31 To elucidate the upstream pathways that regulated histone modi cation, we rst studied the total protein levels and cellular localization of histone deacetylases (HDAC) including HDAC1, 2, and 3, which have been reported to affect both mechanotransduction and differentiation of hMSCs (Fig. 2D, E and Fig. S4C-E). 2, 27 We identi ed that the HDAC3 was accumulated into deformed nuclei on micropillars compared to those on the at surface, although the total cellular levels of HDAC3 remained unchanged. Since enhancer of zeste homolog 2 (EZH2) is currently the only reported histone methyltransferase that catalyzes H3K27me3, 32 we then investigated its localization on at compared to micropillar surfaces (Fig. 2F). According to the staining images, EZH2 is located in both the nucleus and the cytosol on hMSCs cultured on at surface, but mainly stained in the nucleus of hMSCs on micropillars. A signi cant increase of EZH2 was also observed on micropillars from WB ( Fig. 2G and Fig. 4F). The increased EZH2 we observed in deformed nuclei could thus contribute to the enhanced H3K27me3 repressive histone marker. Immunostaining images of HDAC 3 in cells on at and micropillar surfaces. White and yellow arrows indicate staining signal in the nucleus and cytosol, respectively. E. Intensity ratio of nuclear HDAC3 to cytoplasmic HDAC3 uorescence intensity per area of cells on at (n=232 cells) and micropillar (n=238 cells) surfaces. N=3 experiments. F. Immunostaining images of EZH2 in cells on at and micropillar surfaces. White and yellow arrows indicate staining signal in the nucleus and cytosol, respectively. G. Relative change of EZH2 expression compared to total H3 expression in cells. The relative expression level on at surface was normalized to be 1 (N=3 experiments). (*p<0.05, *p<0.05, ***p<0.001).
Nuclear Deformation Causes Signi cant Downregulation of Processes Associated with Development. Since histone modi cations are closely related to transcription pro le, 33 we investigated the in uence of nuclear deformation on gene expression in hMSCs. We performed RNA-seq analysis on differentially expressed genes, to evaluate the effect of the micropillars on the early transcriptional changes in hMSCs. We observed that the majority of identi ed differentially expressed genes are downregulated in cells on micropillars when compared to on a at surface in hMSCs (Fig. 3A, B). Our previous survey of histone modi cations indicated a signi cant increase in H3K27me3 in hMSCs cultured on micropillars. EZH2 catalyzes H3K27me3 and is also a part of the PRC2 complex which is involved in the repression of developmental processes in stem cells. 32,34 Therefore, we performed gene ontology (GO) analysis on the differentially expressed genes (Table S1) to determine whether similar processes were associated with the cells cultured on micropillars (Fig. 3C). Since there was more downregulation of genes in hMSCs cultured on micropillars compared to at surface, we also speci cally performed the GO analysis on signi cantly down-regulated genes (Fig. S5). The enriched processes associated with the identi ed differentially expressed genes include those associated with cell cycle, DNA conformation, and development. Next, to identify sets of genes with coordinated enrichment or depletion for hMSCs on micropillar compared to at surfaces, we performed Gene Set Enrichment Analysis (GSEA) for both Canonical Pathways and Gene Ontology Biological Processes gene sets. 35 At a False Discovery Rate (FDR) <25%, 273 gene sets were found to be depleted, and 50 gene sets were enriched in micropillars in the Canonical Pathways gene sets. Using the top 20 signi cantly depleted terms in the pillar conditions, we found signi cant depletion in gene sets for cell cycle associated pathways (Fig. 3D). Similarly, using the top 20 signi cantly enriched terms in the pillar conditions, we found that most of the upregulated gene sets were associated with translation and various key signaling pathways involved in MSC differentiation, such as those related to PDGF, interleukins, and Rho GTPases (Fig. 3E). Furthermore, we noticed enrichment in gene set associated with chromatin-modifying enzymes (related to histone acetyltransferases, histone deacetylases, histone lysine methyltransferases, and histone lysine demethylase, Fig.  S6). This nding con rmed the in uence of micropillars on the epigenome as a consequence of nuclear deformation. Similarly, we also identi ed the top 20 signi cantly enriched and depleted Gene Ontology gene sets for both at and pillar cell culture conditions which were consistent with results from Canonical Pathways gene sets (Table S3). Altogether, our results show that nuclear deformation impacts the transcriptional pro le in hMSCs to alter developmental processes such as differentiation and proliferation that are inherently linked with each other.
Nuclear Deformation Alters Chromatin Conformation within Packing Domains in hMSCs. Chromatin packing scaling is a key physical property that's tightly associated with transcription regulation and is also a crucial regulator of phenotypic plasticity. 17 To directly investigate the in uence of nuclear deformation on chromatin reprogramming, we measured chromatin packing scaling from the level of packing domains in the mechanically constrained nucleus using partial-wave spectroscopic (PWS) microscopy that is capable of measuring chromatin packing scaling with sensitivity to length scales as small as 20 nm. Additionally, PWS enables the label-free sensing of nanoscale variations in supranucleosomal chromatin structure in both living and xed cells. 36 Speci cally, the variations in the chromatin packing density are measured using PWS in the form of a spectral interference signal originating from internal scattering within the cell nucleus. The shape of the autocorrelation function (ACF) of the chromatin density variations or interference signal is then evaluated to determine the average nuclear chromatin packing scaling, D. 18,37 As shown in Fig. 4A-C, hMSCs seeded on micropillar surface compared to those on the at surface showed a decrease in whole-nuclear chromatin packing scaling of about 8.01 ± 0.74% (SEM) in maintenance medium within 24 hours.
Compared to the effect of other external cues, such as treatment with various pharmacological agents for similar durations, such a change in D using morphological cues is indicative of a drastic change in chromatin conformation. 17,38 In addition, we investigated how micropillar induced changes in chromatin packing scaling compared with those during osteogenic differentiation of hMSCs. We observed a signi cant decrease in chromatin packing scaling after Day 1 of osteogenic induction. Such a decrease in chromatin packing scaling was also maintained in osteogenic differentiating cells at Day 4 and Day 14 of induction (Fig. S7). Chromatin packing scaling of hMSCs was higher compared to osteogenically induced progenitor and differentiated cells, consistent with our previous observations that higher D is associated with phenotypic plasticity. 17 Furthermore, by decreasing D, micropillars, may offer an e cient way to modulate this plasticity associated with hMSCs in order to enhance differentiation e ciency towards a target lineage.
In order to elucidate the effects of nuclear deformation on chromatin conformation below 20 nm, we employed ChromTEM to image chromatin in uninduced hMSCs grown on micropillars versus at surfaces (Fig. 4D). Since the nuclear deformation resulted in a signi cant impact on the lamin A/C organization, we hypothesized that the lamina-associated heterochromatin would be altered in micropillars. Previous electron microscopy studies have revealed that the nuclear periphery is enriched in condensed heterochromatin. 39 Additionally, the nuclear envelope and chromatin organized into compacted domains may contribute to the response of the cell nucleus to mechanical forces. 40 Firstly, we observed that the mass density ratio of the peripheral chromatin to non-peripheral chromatin signi cantly decreases in the deformed nucleus compared to the at surface (Fig. S8). Therefore, we speci cally segmented the peripheral chromatin region and analyzed changes in D within this region. We then evaluated the average ACF of chromatin mass density for both the whole nucleus, and peripheral chromatin of hMSCs on micropillars compared to at surface within 50-200 nm (Fig. 4E). Next, to determine chromatin packing scaling of these different regions, we t a linear regression to the chromatin density ACF within the whole nucleus, and the peripheral domains for each cell from both groups (hMSCs on at and micropillars). We measured a 10.82 ± 3.1% (SEM) decrease in chromatin packing scaling for the whole nucleus in micropillars compared to at surface was measured, which was comparable to the changes in nuclear D obtained using PWS. In addition, we observed an 11.02 ± 2.2% (SEM) increase in chromatin packing scaling for peripheral chromatin in micropillars compared to at surface was obtained (Fig. 4F). These ndings indicate that upon nuclear deformation the anchoring of the chromatin to the periphery of the nucleus may be altered resulting in an increased heterogeneity in the peripheral chromatin domains. Additionally, this increased heterogeneity in the peripheral chromatin might be a consequence of the anisotropic distribution of lamin A/C on micropillars. 41 N=2 experiments.
Nuclear Deformation Increases Transcriptional Responsiveness of hMSCs to Osteogenic Induction with enhanced Wnt/β-catenin signaling. If transcriptional plasticity of stem cells is altered, this could enhance their ability to respond to a given induction cue and in turn, increase their differentiation e ciency towards a speci c lineage. 17 Therefore, we evaluated the lineage-speci c responsiveness coe cient, R LS , de ned as the average transcriptional response to an external differentiation stimulus of stem cells on a micropillar surface compared to a at surface, Here E denotes the expression rate of a speci c gene, subscripts "induced" and "stem" refer to osteogenic differentiation induction and control conditions of stem cells, respectively. Using our bulk RNA-Seq data, genes were grouped based on initial pre-stimulated expression and their change in average expression in response to an osteogenic differentiation stimulus was quanti ed in at and pillar populations (Fig. 5A). An increase in R LS during hMSC differentiation in our case would indicate an increase in lineage-speci c-transcriptional response in micropillar compared to a at surface. First, we employed our previously developed CPMC model, inputting our experimentally determined chromatin packing scaling for the micropillar versus at surfaces, to predict the lineage responsiveness coe cient for genes that are upregulated and downregulated by differentiation induction. For initially lowly expressed genes that are downregulated in the stem cell state, there is further downregulation in the lower D cells on micropillars compared to the higher D stem cells on the at surface, which is shown by an increase in R LS upon stimulation with osteogenic differentiation cues (Fig. 5B, C, orange curve). A similar trend was determined for the initially highly expressed genes, although the magnitude of the change was much smaller (Fig.  5B, C, purple curve). Altogether, the model predicted that, for lower D cells compared to higher D cells, genes with higher initial expression in stem cell condition did not show as much downregulation of genes associated with the stem state as the genes with initial lower expression. Next, we checked if such changes are also observed experimentally by analyzing our bulk RNA-seq data. In agreement with our model predictions, we observed that the lineage-speci c transcriptional response to osteogenic differentiation induction increased in low D cells on pillars compared to high D cells cultured on at surfaces in response to differentiation induction as R LS was >1 for the majority of the group of genes grouped by initial control expression (Fig. 5B, C). Notably, genes with initially low expression in the control stem cell population exhibited a greater change in their global transcriptional pro le on induction compared to initially highly expressed genes as predicted by the model. Additionally, we also observed increased downregulation compared to upregulation of genes in induced cells compared to stem cells on Day 1 after differentiation. These genes associated with the stem cell state had stronger further downregulation in pillars compared to a at surface upon differentiation induction as predicted by the model. Furthermore, the genes associated with differentiation were similarly upregulated in low D cells on pillar versus high D cells on at surfaces. Therefore, the ability of hMSCs cultured on micropillars to differentiate more e ciently is due to increased downregulation of stem cell-associated genes instead of their ability to increase the expression of genes associated with differentiation.
To identify the transcriptional processes that guide changes in transcriptional responsiveness in cells on micropillars, differential gene expression combined with gene ontology analysis was employed to determine the upregulated and downregulated processes in induced cells compared to control cells on at surfaces. The processes identi ed in Fig. 5D demonstrate a large cluster annotated to development-speci c processes such as urogenital system development, blood vessel morphogenesis, epithelial cell differentiation, muscle structure development, etc.. Of these, there is an even smaller cluster of bone development processes which included ossi cation, connective tissue, and skeletal system development. Additionally, we analyzed the effect of genes in these processes by evaluating the lineage-speci c responsiveness coe cient, R LS , for stem cells on micropillars compared to at surfaces (Fig. 5E). We notice that these development-speci c genes follow a similar trend as previously observed, although less drastic as the identi ed differentially expressed genes in Fig. 5B, C. Altogether, these results indicate that micropillars increase the overall response of lineage-speci c genes with initial low expression in differentiating hMSCs, which may contribute to their increased osteogenic differentiation e ciency.
Several potential signal transduction pathways regulate osteogenic differentiation of hMSCs upon induction such as Wnt, BMP, Hedgehog, Notch signaling. 42 Based on our RNA-seq results, we noticed that Dickkopf-related protein 1 (DKK1), an inhibitor of canonical Wnt signaling, was the most signi cantly altered gene between hMSCs on at and micropillar surface after induction ( Fig. 5F and Fig. S9). Interestingly, several Wnt signaling-related genes including Dishevelled (DVL3, DVL1, wnt conductor), and Transducin-like enhancer protein 4 (TLE4, Wnt corepressor) were also found to be signi cantly altered in cells seeded on micropillars compared to at surfaces. Therefore, we hypothesized that Wnt signaling might be promoted on micropillars as the responsiveness of hMSCs to osteogenic induction increases. To test this hypothesis, we stained β-catenin and RUNX2 on the same substrate containing both at and micropillar surfaces ( Fig. 5G and Fig. S10A). β-catenin and RUNX2 showed obvious nuclear accumulation on both surfaces. However, the stronger uorescence intensity of both β-catenin and RUNX2 was observed on micropillars. Quanti cation of absolute nuclear intensity and relative intensity ratio of nuclear/cytoplasmic con rmed the enhanced nuclear accumulation of β-catenin and RUNX2 on micropillars (Fig.  5H, I and Fig. S10B, C). These results suggest that nuclear deformation on micropillars promotes the activation of canonical Wnt signaling upon induction to facilitate the osteogenic differentiation of hMSCs.
To investigate the effects of nuclear deformation on cellular phenotype, we then evaluated the in uence of micropillars on osteogenic differentiation of hMSCs upon induction. Alkaline phosphatase (ALP) quanti cation demonstrated that micropillars promoted osteogenic differentiation of hMSCs (Fig. 5J). Both early (ALP) and late (integrin-binding sialoprotein, IBSP) osteogenic related genes showed increased expression on micropillars ( Fig   5K). Additionally, calcium deposition in cells was also enhanced on micropillars (Fig. S10D, E). Interestingly, the differentiated hMSCs maintained and even further decreased NSI which may be due to enhanced cytoskeletal tension after osteogenic differentiation (Fig. S10F, G). 43 These results are in accordance with a previous study that reported an increase of osteogenic differentiation of mouse hMSCs on micropillars. 13

Discussion
This work demonstrates that contact guidance-induced nuclear morphology changes modulated histone modi cations and reprogrammed chromatin, which in turn in uenced cell plasticity and transcriptional responsiveness of stem cells to external cues. These effects ultimately modulate cell phenotype as summarized in Fig. 6. Deformed nuclei showed a 3D con guration similar to nuclei cultured in 3D scaffolds, which had a more folded structure and favored osteogenesis despite weak cytoskeleton assembly. 44 The nuclear deformation affected nuclear lamin A/C multimerization which regulates the mechanical properties and mechano-sensing andtransduction of the nucleus. 21,45 Since nuclear lamins provide anchoring sites for multiple chromatin domains, alteration in their polarization pattern may also affect the chromatin organization and transcription. 41 In response to micropillar topography, cells assembled a weak cytoskeleton, which resulted in the shuttling of HADC3 to the nucleus attributed to the release of HDAC3 from the IκB complex that dissociated from less assembled F-actin laments on micropillars. 46 Additionally, previous studies reported the existence of cytosolic EZH2 that form methyltransferase complex in association with Vav1 and Talin1, which regulate cell adhesion and actin polymerization. 47,48 Therefore, the strong F-actin assembly and FAs formation on at surfaces may facilitate the shuttling of EZH2 to the cytosol. Changes in nuclear localization of these histone-modi cation enzymes signi cantly affected cellular epigenetics with a decrease in activating histone acetylation and an increase in repressive histone methylation in the deformed nucleus. Correspondingly, there was a decreased transcriptional activity of the majority of development-related genes. These results suggest a decreased cell plasticity in the deformed nucleus which may prevent off-targeted differentiation of hMSCs. 49 Further studies are needed to determine if histone modi cations drive these transcriptional changes or if the histone modi cations themselves are a downstream product of transcriptional change.
Chromatin reprogramming in deformed nuclei was directly observed using PWS microscopy and ChromTEM. The results con rmed that compared to a at surface, pillars cause a signi cant decrease in D in the whole nuclei. Our previous study of structural and temporal changes of chromatin architecture using dual-PWS has shown that hMSCs have increased variations in both chromatin packing density and macromolecular motion within the nucleus than osteoblasts derived from them. 50 Based on the previously established CPMC model of transcription, D is directly related to transcriptional responsiveness. Furthermore, using cancer as a testbed, it has been demonstrated that chromatin packing scaling increases the phenotypic plasticity of cancer cells by determining the responsiveness of the cells to external cues. Cancer cells with higher D are more likely to survive cytotoxic stressors because of their ability to both upregulate survival genes within a critical time period (i.e. their transcriptional malleability) and their inherent transcriptional heterogeneity which increases the spread in available transcriptional states that a cell population can sample from. 17 In accordance with CPMC model predictions, RNA-seq analysis combined with results from our microscopy experiments showed a decrease in D in hMSCs nuclei using pillars is accompanied by an increase in the stem cells' ability to respond to osteogenic differentiation induction at an early time point which is in accordance with the CPMC model. The decrease in D of stem cells cultured on micropillar surfaces resulted in decreased expression of downregulated genes which are downregulated upon differentiation induction. Both our model predictions and experimental results demonstrate that differentiation on micropillars is more e cient than differentiation on at surfaces, as micropillars can decrease chromatin packing in hMSCs and consequently increase downregulation of genes associated with the stem cell state. Therefore, contact guidanceinduced chromatin reprogramming is a powerful tool that can be used to ultimately increase the differentiation e ciency of stem cells. Herein, we demonstrate enhanced osteogenic differentiation of hMSCs due to contact guidance-induced chromatin reprogramming on micropillars fabricated on mPOC, a citrate-based biomaterial (CBB). CBBs have been shown to regenerate bone and are compatible with microfabrication techniques, are biodegradable, and elicit non-toxic, antiin ammatory responses. 51,52 CITRELOCK TM , a biodegradable orthopedic xation device fabricated from a CBB was recently approved by the U.S. Food and Drug Admiration (FDA) (K200725.pdf (fda.gov)). Engineering the topography of orthopedic devices to include micropillars may result in enhanced bone apposition, improving device function and patient outcomes. Overall, our ndings highlight how manipulating nuclear morphology using topographically engineered surfaces impacts chromatin reprogramming and gene transcription to control cell functions, which can in the future be used for various applications such as bone regeneration.
Although our results are promising, further studies are required to determine the role of D in facilitating differentiation. Speci cally, it remains to be explored how nuclear reprogramming, in uences the movement of genes between and within domains in the 3D genome, thus altering their expression patterns. 46 Studies involving chromatin conformation capture methods, gene labeling, combined with sequencing can be synergistically used to explore how a change in the spatial location of groups of genes, such as those associated with lineage/development are able to modulate their expression and eventually determine the cell fate or phenotype as a result of nuclear deformation.

Materials And Methods
Synthesis and Characterization of mPOC Pre-polymer. POC pre-polymer was rstly prepared according to the previous report. 53 Brie y, equal molar of citric acid and 1,8-octandiol were melted at 160 °C. Then, the mixture was transferred into a 140 °C oil bath and reacted for 30 min. The mixture was cooled down and dissolved in ethanol and puri ed by precipitation in DI water. The pre-polymer was lyophilized and used for the methacrylation process. 66 g POC pre-polymer was dissolved in 540 ml tetrahydrofuran (THF) and placed in a 60 °C water bath with stirring. Next, 0.036 mol imidazole was added into the system followed by drop-wise adding 0.4 mol glycidyl methacrylate. After reacting for 6 h, the solvent was removed using a rotary evaporator. The remaining product was puri ed by precipitation in DI water and lyophilized for further application. 5 mg mPOC pre-polymer was dissolved in 1 ml deuterated dimethyl sulfoxide (DMSO-d6) and characterized using proton nuclear magnetic resonance ( 1 H-NMR).
Fabrication and Characterization of Micropillar Substrates. The mPOC micropillars were fabricated using a combination of contact printing and UV lithography. Several features including micropillar size, shape, spacing, and height were designed to regulate cell nuclear deformation. Hard micropatterned master molds were rstly fabricated using photolithography. Brie y, a 4 inches Si wafer was pre-treated using a reactive ion etcher (RIE-10NR, SAMCO Inc.) followed by a spinning coating of SU-8 photoresist (SU8 3010) at a 4000 rpm for 30s. The photoresist was then soft baked at 65 °C (3 min) and 95 °C (2 min), respectively. The micro-features were transferred to the photoresist using the MLA150 mask-less aligner. After UV exposure, the photoresist was post-baked at 65 °C (1 min) and 95 °C (2min), and developed in SU-8 developer followed by a rinse with isopropanol and DI water. Hard bake at 150 °C for 10 min was performed to ensure the photoresist strength for repeated use. A PDMS mold replicating the micro-features was made by curing PDMS precursor (120 °C, 30 min) on the master mold. Finally, 50 μl mPOC pre-polymer (70% v/v in ethanol) mixed with photo-initiator (5 mg/ml camphorquinone and ethyl 4dimethylaminobenzoate) was added onto the PDMS (2 x 2 cm 2 ) mold and covered with an oxygen plasma pretreated cover glass. The polymer was cured by exposure to 405 nm laser light with a power of 1W for 2 min. After peeling off the PDMS mold, the mPOC micropillars were fabricated onto the cover glass. The prepared microfeatures were characterized by SEM (FEI Quanta 650 ESEM) and 3D Optical Microscope (Bruker). The mPOC micropatterns were sterilized using ethylene oxide and kept in DPBS before cell culture. Flat mPOC lms were fabricated using the same method as at PDMS molds. To fabricate substrates for ChromEM imaging, micropillars were directly fabricated on a cover glass using SU8-3010 which can reduce background and facilitate sample sectioning.
Cell Nuclear Deformation on Micropillars. Human mesenchymal stem cells (hMSCs, PCS-500-012) were purchased from the American Type Culture Collection (ATCC) and sub-cultured using a growth medium acquired from the same company. hMSCs (P4-P6) were seeded onto the at mPOC substrates and the mPOC micropatterns with various microfeatures. After one-day culture, the cells were xed with 4% paraformaldehyde, and cell nuclei were stained with 4',6-diamidino-2-phenylindole (DAPI) to show nuclear morphology. Nuclear shape index (NSI) was calculated according to the staining images using the following equation: NSI = 4πA/p 2 , in which A represents the area and p represents the perimeter. A total of 273 nuclei on at surfaces and 295 nuclei on micropillar surfaces from 4 biological replicates were imaged and analyzed to calculate the statistics. In order to acquire 3D nuclear morphology, the stained cells were imaged using a confocal microscope (Leica SP8). The acquired images were analyzed using Fiji ImageJ software (https://imagej.net/Fiji) to measure cell nuclear volume, surface area, project area, height, and the ratio of surface area to volume (3D objects counter). A total of 33 nuclei on at surfaces and 34 nuclei on micropillar surfaces from 3 biological replicates were imaged and analyzed to calculate the statistics. Cytoskeleton Inhibition. hMSCs were cultured on micropillars for one day before treatment of pharmacological agents. Speci cally, F-actin laments, microtubules, and intermediate laments were disrupted with 1 μM Latrunculin A, 1μM colchicine; 4mM acrylamide, respectively. 6 h post-treatment, drugs were washed out and the cells were xed and stained with Hoechst to show the nuclear morphology.
Immunostaining, Imaging, and Quanti cation. Cells on at and micropillar surfaces were xed and permeabilized with 0.1% Triton-X100 followed by blocking with 1% BSA solution. Then, the primary antibody (detail in Supplementary Table S6 and S7) were diluted in blocking solution and incubated with cells at 4°C overnight. After washing with PBS buffer, secondary antibodies and Hoechst were diluted 1:1000 in PBS and incubated with cells at room temperature for 1h. The uorescent images were acquired using a cytation 5 imaging reader and a Nikon eclipse TE2000-U inverted microscope. Histology images were analyzed using ImageJ (1.50 i, NIH, Download (nih.gov)) according to a previous report. 54,55 The nuclear YAP positive cells were de ned as those that had a nuclear/cytoplasmic intensity ratio larger than 2. Lamin A/C staining images were acquired using a confocal microscope and the intensity plot was analyzed using ImageJ. A total of 25 and 26 nuclei on at and micropillar surfaces, respectively, from 3 biological replicates were imaged and analyzed to calculate the statistics. Histone acetylation and methylation markers were imaged to qualitatively evaluate the in uence of nuclear deformation on cell epigenetics. HDAC 1, 2 and 3, and EZH2 were stained to study the upstream effects of histone modi cations. A total of 232 and 238 cells on at and micropillar surfaces from 3 biological replicates, respectively, were imaged and analyzed to calculate the statistics. β-catenin and RUNX2 were probed to study the effects of nuclear deformation on canonical Wnt signaling. A total of 187 and 199 cells on at and micropillar surfaces from 3 biological replicates, respectively, were imaged and analyzed to calculate the statistics. To investigate nuclear shuttling of HDAC3, β-catenin and RUNX2, their nuclear and cytoplasmic uorescence intensity was analyzed according to the staining images. Brie y, a rectangular frame was created in cell nucleus. The overlap region was selected as region of interest (ROI). The mean uorescence intensity (M N ) of ROI was measured. Then, the frame was randomly set in the cytoplasm to detect the mean intensity of plasm (M P ). Subsequently, the mean intensity (M B ) of the unoccupied region (no cell region) was measured as the background. Therefore, the nucleus to cytoplasm intensity ratio (R) can be described as: R=(M N -M B )/(M P -M B ).
Western Blot Analysis. hMSCs on different substrates were lysed using radioimmunoprecipitation assay (RIPA) buffer containing 150 mM NaCl, 0.1% Triton X-100, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulphate (SDS), 50 mM Tris-HCl and 1% protease inhibitors. Protein lysates were then centrifuged to remove cell debris and any other undissolved component. The relative quantity of proteins was measured using citation 5 imaging reader. Equal amounts of proteins from at and micropillar samples were run using a NuPGAE 4-12% Bis-Tris Gel (Invitrogen) and transferred to nitrocellulose membranes (Bio-rad). The membranes were then blocked using 5% milk and incubated with primary antibodies (including GAPDH, lamin A/C, total H3, H3Ac, H3K27me3, HDAC1, 2, and 3, and EZH2, Supplementary Table S6 and S7) at 4 °C with gentle shaking overnight. Secondary antibodies were diluted at a ratio of 1:5000 and incubated with the membranes at room temperature for 1h. Protein bands were visualized using an ImageQuant LAS 4010 Gel Imager (Cytiva). The acquired images were analyzed using 'Gel Analyzer' in ImageJ. The intensity of all target protein band was rstly compared with relevant total H3 and then normalized with at surface that was set as 1. Three biological replicates were used to calculate statistics. PWS Microscopy. Bone Marrow-Derived Mesenchymal Stem Cells (hMSCs) were cultured in 35 mm glass-bottom Petri dishes (Cellvis, Mountain View, CA) with a micropatterned or at surface in growth medium or osteogenic differentiation medium at 37°C and 5% CO 2 .
The PWS microscopy images were acquired on a commercial inverted microscope (Leica, Buffalo Grove, IL, DMIRB) with a Hamamatsu Image EM charge-coupled device camera (C9100-13) coupled to a liquid crystal tunable lter (CRi, Woburn, MA) to collect spectrally resolved images between 500 to 700 nm with 1 nm step size.
Further, broadband illumination is provided by an Xcite-120 LED lamp (Excelitas, Waltham, MA). PWS microscopy was used to capture spatial variations of the refractive index distribution or chromatin packing density heterogeneity (∑) within the nucleus. Further, the statistical parameter of chromatin structure, packing scaling (D) was calculated from ∑. 37 At least 10 independent elds of view were utilized for each experiment and four biological replicates were used for the analysis. D value was calculated for 111 hMSCs from the at surface, and 110 hMSCs on pillar surfaces.
ChromTEM Sample Preparation, Image Acquisition, and Analysis. ChromTEM staining targets nuclear DNA speci cally by utilizing the "click-EM" method. 56 Compared to conventional negative staining, which ubiquitously labels nucleic acid by uranyl acetate and lead citrate, ChromTEM provides us an opportunity to investigate chromatin organization from the perspective of DNA packing at high resolution. The image contrast for ChromTEM at bright-eld for a thin resin section follows Beer's Law, which can be converted to DNA concentration with calibration: Here I(x,y) is the intensity of the resultant image, I 0 is the intensity of the incident beam, o is the absorption coe cient of the sample, p(x,y) is the density distribution, and t is the thickness of the section consisting of the sample. We assumed that for a given resolution, the absorption coe cient is constant. Further incident beam and section thickness was controlled to be the same across all images. Therefore, after taking a negative logarithm of the image followed by subtraction of the mean from the image, we directly obtain the chromatin density uctuations from the image intensity.
The chromatin density uctuations can then be used to estimate ACF (autocorrelation function of chromatin density) using the Weiner Kinchen relation as previously described. 57 D can be evaluated using the power-law relationship of ACF approximated by: where r is the spatial separation. This is followed by linear regression analysis to obtain the chromatin packing scaling for a given region within the nucleus. We were able to evaluate local chromatin packing D at different length scales by linear regressions on ACF in log-log scale.
Cells were prepared for TEM imaging using the ChromEM staining protocol 42 . hMSCs seeded on a micropatterned or at surface in growth medium were washed with Hank's balanced salt solution without calcium and magnesium. Next, hMSCs were xed for 5 minutes at room temperature with EM grade 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1M sodium cacodylate buffer (EMS) and the xation was continued for 1 hour on ice after refreshing the samples with fresh xative, and all consecutive steps were carried out in the ice. Cells were washed with 0.1M sodium cacodylate buffer ve times for 2 minutes each, blocked with potassium cyanide (Sigma Aldrich) blocking buffer for 15 minutes, and stained with DRAQ5 ™ (Thermo Fisher) with 0.1% saponin (Sigma Aldrich) for 10 minutes. Then the samples were washed with blocking buffer. Samples were then moved to a pre-cooled custom-made chamber. Photobleaching was done on samples in the chamber bathed in 3-3' diaminobenzidine tetrahydrochloride (DAB) solution (Sigma Aldrich) and using a Nikon inverted microscope (Eclipse Ti-U with the perfect-focus system, Nikon) with Cy5 lter set, along with a 15 W Xenon lamp and the red lter for epi-illumination. Total photobleaching time per sample was about 20 minutes using a 20X objective. Once photobleached, the samples were washed with 0.1 M sodium cacodylate buffer, about ve times for 2 minutes each, and the samples were then stained with reduced osmium (2% osmium tetroxide and 1.5% potassium ferrocyanide, EMS) for 30 minutes. Further, the cells were washed with DI water. This was followed by serial ethanol dehydration at 30%, 50%, 70, 85%, 90%, and 100% (x2), with the nal 100% ethanol added at room temperature, unlike the preceding steps. In ltration and embedding using Durcupan resin (EMS) were performed and the samples were then cured at 60°C for 48 hours. 50 nm thick resin sections were prepared using an ultramicrotome (UC7, Leica). TEM (HT7700, HITACHI) was operated at 80 kV in a bright eld to acquire high contrast images of the samples. For image analysis, the nuclei were segmented using the Image Segmenter toolbox on MATLAB. Further, peripheral chromatin was segmented by creating a mask of 150 nm (median evaluated from a total of 716 data points of peripheral heterochromatin thickness from 11 cells) in thickness from the nuclear periphery. The ACF was calculated from all the cells in both at and pillar groups for the whole nucleus, and the peripheral chromatin to evaluate the D value by log-log tting from r = 80 nm to r = 200 nm and from r = 50 nm to 75 nm, respectively. For statistical analysis, the D value calculated for 20 hMSC on at suface, and 12 hMSCs on micropillar surface. Imaging was perfomed on four independent 50 nm sections from two biogical replicates from each group (hMSCs on at and hMSCs on micropillar surfaces).
Differential Gene Expression and Gene Ontology analysis. RNA extraction was performed on samples from at and micropillar surfaces in both regular and osteogenic differentiation medium with two biological replicates per condition. Sequencing and library preparation was performed by Northwestern University NUSeq Core Facility.
Illumina HiSeq 4000 Sequencer was used to sequence the libraries with the production of single-end, 50-base pair reads. The quality of reads, in fastq format, was evaluated using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc). Adapters were trimmed, and reads of poor quality or aligning to rRNA sequences were ltered using Trim Galore (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). The cleaned reads were aligned to the human genome (hg19) using STAR. 58 Read counts for each gene were calculated using HTSeq-Counts 59 in conjunction with a gene annotation le for hg19 obtained from Ensembl (http://useast.ensembl.org/index.html). A comprehensive QC report was generated using MultiQC. 60 Differential expression was determined using DESeq2. 61 The cutoff for determining signi cantly differentially expressed genes was an FDR-adjusted p-value less than 0.05. The pathway analysis was done using Metascape. 62 Osteogenic Differentiation of hMSCs. hMSCs were seeded onto both at and micropillar substrates. One-day postseeding, an osteogenic induction medium (Lonza) was used to induce the osteogenic differentiation of hMSCs.
After 7 d induction, the cells were washed with PBS buffer followed by xation with 4% paraformaldehyde for 10 min. Immediately, the samples were merged in a solution of 56 mM 2-amino-2-methyl-1,3-propanediol (AMP, pH~9.9) containing 0.1% naphthol AS-MX phosphate and 0.1% fast blue RR salt to stain ALP. The bright led images were then acquired using a Nikon Eclipse TE2000-U inverted microscope. The ALP activity was tested using the ALP assay kit (K422-500, Biovision) followed by the manual. Brie y, cells that have been cultured in induction medium for 7 d were homogenized using ALP assay buffer. Then, the non-uorescent substrate 4-Methylumelliferyl phosphate disodium salt (MUP) was mixed with the homogenized samples to generate a uorescent signal due to its cleavage by ALP. The uorescence intensity was read by a cytation 5 imaging reader (BioTek) at (Ex/Em = 360/440 nm). The enzymatic activity was calculated based on the standard curve and normalized to total DNA content tested by the Quant-iT PicoGreen dsDNA assay (Invitrogen). Four biological replicates were used to calculate statistics. The expression of ALP and IBSP of hMSCs on at and micropillar substrates was evaluated using quantitative real-time reverse transcription-polymerase chain reaction (RT-qPCR) after 7d induction. The total RNA of the cells was extracted using the Aurum Total RNA Mini Kit (bio-rad) according to the protocol. The concentration and purity of the extracted RNA were tested using the citation 5 imaging reader.
The RT-qPCR was carried out using iTaq Universal Sybr Green One-step Kit following the vendor's protocol. The designed primers for ALP are: Forward, 5'-GACCCTTGACCCCCACAAT-3'; Reverse, 5'-GCTCGTACTGCATGTCCCCT-3'. The designed primers for IBSP are: Forward, 5'-TGCCTTGAGCCTGCTTCC-3'; Reverse, 5'-GCAAAATTAAAGCAGTCTTCATTTTG-3'. We employed GAPDH as the house-keeping gene. The GAPDH primers are: Forward, 5'-GTGGACCTGACCTGCCGTCT-3'; Reverse, 5'-GGAGGAGTGGGTGTCGCTGT-3'. The data was analyzed using the 2 -ΔΔCt method. The expression of target genes was rstly normalized to that of GAPDH, and then to the average values on at substrates. Three biological replicates were used to calculate statistics. Calcium deposition were stained with Alizarin Red S on both surfaces after 3 weeks of induction. The positively stained area in the bright led images was analyzed using ImageJ software.
Lineage-speci c Responsiveness Analysis. Raw reads were aligned and mapped to the human hg38 ENSEMBL genome using bowtie2. Transcripts per million (TPM) for each condition were estimated from mapped reads using RSEM. 63 The lineage-speci c responsiveness coe cient, R LS was de ned as the average transcriptional response to an osteogenic differentiation stimulus of cells on the pillar surface compared to the at surface. Genes with similar initial prestimulated expression, based on their quantile of log 2 (E Induced /E control ), are grouped and their change in average expression in response to the stimulus is quanti ed in at and pillar populations for initially under-expressed and overexpressed genes. Further, GO analysis was done on DE genes with p-value<0.05 and |FC|>1.5 in induced cells compared to control cells to evaluate the transcriptional malleability in biological processes that are involved in early differentiation.
Statistical Analysis. Some of the results are shown as mean ± S.D. using a bar plot. The others were shown using the box and whisker plots that represent median values (horizontal bars), 25th to 75th percentiles (box edges), and minimum to maximum values (whiskers), with all points plotted. Statistical analysis was performed using Kyplot software (version 2.0 beta 15). We also reported standard error of the mean (SEM) with mean values obtained from imaging experiments. Statistical signi cance among each group was determined by unpaired t-test (two-sided) and a value of p<0.05 was considered to indicate a statistically signi cant difference. For RNA-Seq analysis, the cutoff for determining signi cantly differentially expressed genes was a p-value less than 0.05. All experiments presented in the manuscript were repeated at least as two independent experiments with replicates.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Declarations
Data availability