Active topological glass

The glass transition in soft matter systems is generally triggered by an increase in packing fraction or a decrease in temperature. It has been conjectured that the internal topology of the constituent particles, such as polymers, can cause glassiness too. However, the conjecture relies on immobilizing a fraction of the particles and is therefore difficult to fulfill experimentally. Here we show that in dense solutions of circular polymers containing (active) segments of increased mobility, the interplay of the activity and the topology of the polymers generates an unprecedented glassy state of matter. The active isotropic driving enhances mutual ring threading to the extent that the rings can relax only in a cooperative way, which dramatically increases relaxation times. Moreover, the observed phenomena feature similarities with the conformation and dynamics of the DNA fibre in living nuclei of higher eukaryotes.

T remendous interest has been devoted to understanding of the glass transition driven by an increase in packing fraction or a decrease in temperature in soft and deformable systems 1,2 . Ultra-soft particles can be realized experimentally using polymers, such as long polymeric stars or rings, which are highly deformable but possess a fixed topology imposed during the synthesis. While it is known that the polymer topology has strong impact on the stress relaxation mechanism 3 , it has been questioned if the topology can independently induce a glass transition 4 .
Such a topological glass has been conjectured for the system of long, unknotted, and nonconcatenated polymer rings in dense equilibrium solutions 4,5 . The rings cannot cross each other, but are known to thread-one ring piercing through an eye of another ring, which temporarily topologically constraints the motion of the two rings. Mutual threadings of many rings can yield a conformation that is relaxed only if a cascade of threadings is sequentially undone, which could give rise to very long relaxation times, e.g., exponential in the ring length 6 . The glassy behavior in an equilibrium melt of rings has been observed in computer experiments, but only under a pinning perturbation, which immobilizes a fraction f of all rings. Then, for sufficiently long rings and high density a glassy behavior can be extrapolated to f ! 0 4,5 . Unfortunately, the conjectured critical length, 90 entanglement lengths, of unpinned rings is currently beyond the reach of experiments and, although the pinning deepens our theoretical understanding of the glass transition [7][8][9] , creating it experimentally to drive the topological glass transition would be challenging.
Whereas many questions remain open in the traditional glass transition of passive Brownian particles, recently a whole new research direction has been opened by considering system composed of active particles that are driven by non-thermal fluctuations 10 . While, intuitively, activity opposes glassiness by enhancing mobility of the particles, some active models can exhibit a more complex behavior as a function of the active control parameters. For example, increasing the persistence time of the active Ornstein-Uhlenbeck particles can either glassify or fluidize the active system, depending on the particular state point, as a result of nontrivial velocity correlations in the system [11][12][13][14] . Indeed, some system properties, such as the time-dependent effective temperature are pertinent to active fluids and render also the corresponding glass transition distinct from the passive one. In particular, the location and the existence of the glass transition of active fluids are dependent on the microscopic details of the activity mechanism. Nevertheless, close to the transition region, universal features of the passive glassy dynamics have been found recently for active spin-glasses 10 and self-propelled particles in the non-equilibrium mode-coupling theory (MCT) 14 . For instance, the scaling of the relaxation time with activity control parameters is governed by the same exponent as in the passive MCT.
The impact of topology on the active glassy states has been studied almost solely in the context of active particles confined to a topologically nontrivial space, in particular a sphere [15][16][17][18][19] . There, a range of dynamic phenomena arises as a consequence of the competition between directed flows, characteristic for active matter in euclidean space, and the 'hedgehog theorem' 20 that asserts the existence of topological defects in a smooth vectorial field on a sphere. In contrast to these studies, where the topology is a global property of space, here we focus on a system where the active particles themselves are not point-like but feature intrinsic circular topology, the embedding space being Euclidean with periodic boundary conditions.
Glassy dynamics also appears in various biological contexts ranging from the bacterial cytoplasm 21 to collective cell migration in tissues 22 . While on the subcellular level, glass-like properties have been attributed to the high crowding, as well as size and interaction heterogeneity of the constituents 21,23 , the confluent tissues modeled using vertex models exhibit a new type of rigidity transition at constant density without 24 and with active motion 25 , attributed to an interplay of cells' shape and persistence of motion. Although there is no topology involved in these models and, therefore, they are inherently different from our present work, the transition occurs due to shape changes at constant density similar to the case studied here.
For dense solutions of ring polymers, we show that making the rings locally more mobile by introducing a moderate segmental activity, the system reaches a glassy state with dramatically slowed-down relaxation. This novel state of matter, the 'active topological glass', is a consequence of the interplay between internal polymer topology, activity, and the crowded polymer environment. In contrast to the conjectured equilibrium topological glass, here no imposed pinning is necessary and only relatively short rings are sufficient to observe the transition. Moreover, contrary to well-studied polymer glasses 26 , where the monomers are arrested due to their nearest neighbors, here the centers of mass of the whole chains are inhibited due to multibody, long-range effects of topological constraints. After detailed account of the physics of the active topological glass, we discuss its relevance for the organization and dynamics of chromosomes in living eukaryotic cells.

Results
Dynamics after the onset of activity. We start with a large, wellequilibrated, concentrated solution of M ¼ 1600 passive, uncrossable, unknotted, and nonconcatenated rings of length N ¼ 400 using the well-established polymer model as in Halverson et al. 27,28 . A consecutive segment of length N h ¼ 50 monomers is made active on each chain by subjecting it to thermal-like fluctuations of temperature three times higher than the rest of the chains (see Methods for model details).
After switching on the activity, the initial equilibrium uniform spatial distribution of the active (hot) and the passive (cold) segments alters (Fig. 1a). They progressively segregate to compensate for the local pressure imbalance and to decrease entropy production [29][30][31] . Simultaneously, we observe a gradual but dramatic conformational change of the rings as revealed by the mean radius of gyration hR g i and other shape parameters (see Supplementary Note 1) that comes to a standstill after several equilibrium diffusion times τ diff ' 4 10 5 τ (see Methods for the definition). In an equilibrium system of linear block copolymers, the colocalization of like-blocks can drive local density inhomogeneities that affect chain conformations. We show in Supplementary Note 3 that analogous effect is not responsible for the conformational changes in the present non-equilibrium system. Foremost, we do this by simulating a system with a low fraction of active chains where the colocalization of hot segments is not present but the conformational changes persist ( Supplementary  Fig. 9). In addition, we show on a simplified effective equilibrium model that in a fully phase separated system the ring conformations do not differ substantially from the homogeneous equilibrium case (Supplementary Fig. 4).
We observe that the hot segment is usually localized at one of the ends of a tree-like, doubly folded conformation (Fig. 1b). This suggests that the conformational change is caused by the differences between the dynamics of the active and passive segments. As the active segment undergoes stronger thermal fluctuations, its diffusivity is enhanced in comparison to the passive one. In fact, it essentially drags the cold tail through the mesh of other chains (Supplementary Movie 1). We illustrate such emergent directionality of the ring's motion by computing the 'cold-hot' vector ch connecting the centers of mass of the cold and hot segments of each ring. After the onset of the activity, the mean magnitude of ch initially grows (Fig. 2a). During this time, the direction of the motion is correlated with the direction of ch (Inset of Fig. 2a). At later times, jchj decreases very slowly, generating a weak anticorrelation with the displacement vector, which is connected to the strengthening of topological constraints as detailed later. As we show in Supplementary Note 2, systems without topological constraints lose their directionality on a microscopic timescale, but the dense polymer environment generates a more intriguing global dynamics.
To describe the dynamics, we track in time the mean squared displacements of the rings centers of mass hg 3 ðt; t 0 Þi (see Eq. (5) in Methods and note that the mean is taken over the rings only because the dynamics is not stationary in general).  The mean radius of gyration hR g i obtained as an average over all rings at a given time t after the onset of activity (Eq. (4) in Methods). Insets: snapshots of two rings in equilibrium (left) and two at late times (right). The hot segment on the active rings is shown in orange (on the equilibrium rings the orange segment is highlighted for comparison only and has the same temperature as the rest of the system).  shows hg 3 ðt; t 0 Þi as a function of time t measured from various times t 0 after the activity onset. hg 3 ðt; 0Þi increasess initially and after 10 6 τ displays a dramatic slow-down. The crossover time is the same as for the structural changes which underlines the fact that the two effects are dependent. More insight is provided by Fig. 2c, where hg 3 ðt; t 0 Þi=t is plotted as a function of time separately for the active and the inactive segment. The initial decrease corresponds to subdiffusive hg 3 ðt; t 0 Þi $ t 0:8 regime consistent with the equilibrium rings dynamics below the diffusion time 28 . The following regime shows again the directional dynamics of the rings and explains its origin. While the passive segment, and hence the whole center of mass, move superdiffusively (exponent 1:13 ± 0:01), the active one temporarily displays standard diffusion (exponent 1). The both stages last for less than a decade in time before both segments cross over to an arrested state with exponent very close to zero. The superdiffusive regime is a consequence of the specific nonequilibrium dynamics through a mesh of topological constraints. As the detailed balance is violated due to the coupling to different thermostats, the pulling of the active segment forwards is stronger than the pulling of the cold tail backwards. The hot segment robustly explores new sites that are spontaneously freed due to density fluctuations and progressively drags behind itself the cold tail (see Supplementary Movie 1). Furthermore, such motion of the active segment away from its cold tail through the environment of neighboring rings introduces new topological constraints that the cold tail must obey. These constraints restrict the transversal motion of the chain. Finally, when the chain is getting more stretched some time after the activity onset, the motion of the hot segment backwards to the cold tail is compromised by the chain flexibility and, therefore, the motion away from the tail prevails. As a result, the cold tail follows the hot head slowly, but ballistically along a trail imposed by the topologically constrained neighboring rings. At these length scales, the trail is characterized by the size R of the static conformation of the tail and it scales with the contour distance s as RðsÞ $ s ν trail where the exponent ν trail ¼ 0:57 ± 0:01 (Fig. 2d). Therefore, the directed dynamics (s $ t) along such contour is superdiffusive with hg 3 ðt; t 0 Þi $ R 2 ðsðtÞÞ $ t 2ν trail , which is in agreement with our observation (Fig. 2c). Moreover, the onset time of the superdiffusion t % 10 5 τ is consistent with the onset time of a configuration that is more open than a random walk, that is ν trail > 0:5 as seen in Fig. 2d, and the dragging mechanism is consistent with the fact that the endpoints of the active segment are closer to each other than in the equilibrium case (Fig. 2e).
Glassy behavior. Subsequently, the system is slowing down, as revealed by hg 3 ðt; t 0 Þi (Fig. 2b). To characterize the slowing down of the relaxation of the rings in more detail, we measured the selfpart of the intermediate scattering function (ISF) F s ðq; t; t 0 Þ (Eq. (6) in Methods). As shown in Fig. 3a, F s ðq Ã ; tÞ depends on t 0 , similar to the aging in classical glasses 32 . After about 2 10 5 τ ' τ diff the system fails to relax and F s ðq Ã ; tÞ plateaus at a nonvanishing value, defined as the non-ergodicity parameter f q (Fig. 3b). A striking characteristic of the ensuing arrested state is that it features a single, β-relaxation process and the subsequent α-relaxation is absent, in contrast to the common, two-step relaxation scenario 33 encountered for polymer glasses 34 or for repulsive colloids 35 . Indeed, we have not been able to observe the α-relaxation, despite the fact that we have extended our simulations to very long times, over 22τ diff . The absence of a twostep process is a feature associated with continuous, type-A glass transitions, as opposed to the discontinuous, type-B transitions [36][37][38] , and it implies the presence of higher-order singularities, the so-called A 3 -and A 4 -critical points, in the framework of MCT [39][40][41][42] . There are strong indications that the system at hand features such higher-order singularities, a point to which we will return in the Discussion section. At later measurement start times, the non-ergodicity parameter is higher, that is, more wave vectors fail to relax, and we cannot even observe the β-relaxation (more details in Supplementary Note 4). In fact, as the system evolves toward a steady state (10 6 τ < t < 2 10 6 τ), progressive strengthening of the topological constraints in the system takes place, which restricts the rings' motion and leads to the rise of the plateau height. This characteristic is similar to the strengthening of glassy behavior for colloidal systems as the packing fraction grows 43 .
After about 2 10 6 τ, hg 3 ðt; t 0 Þi becomes independent of t 0 and only small changes in hg 3 ðtÞi and F s ðq Ã ; tÞ are noticeable. These are due to the local system explorations of the hot segments ( Supplementary Fig. 2), reminiscent of a confined diffusion with occasional constraint release. hg 3 ðtÞi is strongly subdiffusive ($ t 0:36 ) and very slow, typical for polymeric glasses 26 . The corresponding relaxation time, extrapolated as hg 3 ðτ relax Þi ¼ 4R 2 g , is τ relax % 10 12 τ, being more than six orders of magnitude higher than the equilibrium one. Such a strong dependence of the relaxation time on the control parameter is a hallmark of glassy systems. Linear polymers in an equilibrium melt relax slower than rings due to a reptation relaxation mechanism, but for the lengths considered here, their relaxation time is only about twice as large as that of the equilibrium rings 28 . Therefore, the glassy behavior  To further support the evidence of glassy dynamics, we measure the probability distribution of 1d displacements for various lag times Δ in the glassy regime, that is for t > 2 10 6 τ (Fig. 4a). For short Δ ¼ 2500τ, the distribution is close to Gaussian, characterizing the standard diffusion (also in the nonequilibrium two-temperature case 44 ), whereas for longer lag times (Δ > 10 5 τ) it becomes markedly non-Gaussian (see also Supplementary Fig. 5a). Tiny displacements and their non-Gaussian distribution characterize a local constraint (cage) exploration which is another hallmark of glassy systems 45 . Interestingly, for long lag times the tails of the distribution are just simply exponential as in the equilibrium topological glass induced by pinning perturbations 5 . We attribute these 'fat' tails to a constraint release and a short relocation of the hot segments of an individual chain (Supplementary Movie 1). In addition, we measured the ergodicity breaking parameter EB 5 , defined by Eq. (7) in Methods. The EB characterizes how quickly (averaged over rings), the time average of a single ring g 3 converges to the ensemble average hg 3 ðΔÞi. While in equilibrium EB typically decays as t À1 , the constraints with diverging lifetimes make glassy systems nonergodic with EB $ t 0 . We plot EB in the steady state regime (Fig. 4b), where we see a dramatic slowing down with exponent around À0:1 even for short lag times (see also Supplementary Threading analysis. We now show that the mutual ring threadings are responsible for the glassiness. We analyzed the ring threadings using computationally spanned minimal surfaces on the ring contours (Fig. 5a). An intersection of one ring's contour through another ring's minimal surface defines a threading of the second ring. This method was used to clarify the extent and the role of threadings in equilibrium ring melts 46,47 (details in Methods).
We characterize the threading depth in terms of the so-called separation length L sep , defined by Eq. (9) in Methods. It approximates how much of the threading rings material is on one side of the threaded ring. Then, the ratio Q ¼ L sep =ðN À L sep Þ defines the relative portion of the material on one side compared with the other side of the threaded ring's surface. This ratio provides a model-independent view on threadings because its distribution in equilibrium is insensitive to the polymer model above the entanglement length scale N e 47 . We found that more ring pairs are involved in threadings and that they are progressively deeper compared with equilibrium ( Fig. 5c, d,  respectively). Moreover, we found a positive correlation of the location of threadings with the local mechanical stress (Supplementary Fig. 3).
The glassy behavior should be connected to the emergence of a system-spanning cluster of rings that fail to relax due to mutual threadings. We define two rings belonging to the same cluster if at least one of them threads the other one with depth L sep ! L cutoff . Therefore, we discriminate the cluster structure by the depth of threadings. Figure 5b shows the relative size of the biggest cluster as a function of L cutoff for different times after the activity onset. For low L cutoff , the whole system is one cluster as each ring has shallow threadings with many of its neighbors. However, in equilibrium (t 0 ¼ 0) the threadings are rarely deeper than 3N e and therefore for L cutoff > 100 ' 3N e the system can be viewed as a set of many small disconnected clusters. This sharply contrasts with the structure in the glassy regime (t 0 > 10 6 τ), where all rings form a single cluster practically independent of the choice of L cutoff as a significant number of threadings of any depth occurs (Fig. 5b).
While the cluster profile remains stable in the glassy regime, we still observe threading and unthreading events between ring pairs. Numbers of these events, however, balance each other, resulting in a steady state (Fig. 6a). Interestingly, the threading lifetime distribution shows a power-law with a peak at late times, meaning that the majority of threadings are persistent and survive for the total duration of our simulations with a minority having a short lifetime. The bimodal character is likely a consequence of the fat power-law tail as revealed by the shape of the distribution measured at different times after the activity onset ( Fig. 6b and Supplementary Movie 2). In addition, the threading two-point correlation function Φðt; t 0 Þ (defined in Methods) exhibits incomplete relaxation at all times, showing a dynamic threading steady state with persistent threadings (Fig. 6c).
Although in equilibrium the threading depth is correlated with the diffusion slow-down of individual rings 46 , in the case of partly active rings, some specific shallow threadings (for example, see Fig. 7a) can significantly increase the relaxation time. The directionality of the rings tightens these threadings and they can be relaxed only if the active segment backtracks the passive tail. However, the directionality and the presence of other rings oppose the back-tracking. These threadings are likely members of the persistent class of threadings, but we could not determine the latter class yet. We suspect, that such threadings could also be relevant in the dramatic increase of the viscosity in stretched untangled melt of rings 48 (after this paper has been already accepted, the hypothesis has been confirmed in ref. 49 ). To prove at least that ring topology and threadings are essential for the glass transition, we took a configuration of the system in the glassy state and cut the bond connecting two cold monomers in the middle of the cold segment on each ring. We further simulated the system which now consisted of M linear triblocks of length N with two cold segments at the ends and a hot segment in the center. The chain conformations change only moderately (Supplementary Note 5), but it is clear from hg 3 ðt; t 0 Þi=t (Fig. 7b) that the chains start to superdiffuse, releasing the accumulated mechanical stress (Supplementary Movie 3). Later on, the dynamics eventually crosses over to standard diffusion in analogy to orientational relaxation of self-propeled active particles 50 . As the non-topological properties of the system remained unchanged, we conclude that the phase segregation is not a crucial element stabilizing the glass. The glassiness is driven by the enhanced threading due to the ring topology and the violated detailed balance 29 .

Discussion
We now return to the question of the order (continuous or type-A vs. discontinuous or type-B) of the glass transition for the system at hand. The usual control parameters driving vitrification in molecular or colloidal systems are the temperature and the density, and the typical glass-transition scenario there is discontinuous: the intermediate scattering function in the ergodic state develops a plateau, which grows in height and extends longer in time approaching the glass transition as the control parameters are changed 26,33,51 . In the continuous case, the nonergodicity factor grows smoothly from zero to finite values. This second scenario is less common and its realization requires the presence of additional control parameters, such as porosity and randomness [36][37][38] , tunable attraction widths in the   Fig. 6 Threading dynamics. a Threading gain Δn þ th and loss Δn À th (see Methods for all definitions), as well as the net gain Δn þ th À Δn À th as a function of time after the onset of activity. b Threading survival time distribution πðt; t 0 Þ measured from different t 0 . The peak at late times represent the fraction of ring pairs threaded until the end of the analyzed data (3:4 10 6 τ). c Threading correlation Φðt; t 0 Þ as a function of t measured from different t 0 . interactions 52,53 or confining periodic potentials 54 for which the amplitude and the wavelength can be independently varied. In such cases, lines of continuous glass transitions have been found in parameter space, and they are associated with higher-order singularities of the A 3 -or of the A 4 -type. The latter appear as endpoints of type-A transition lines that merge with type-B lines, as endpoints of type-B lines separating two glasses or as endpoints of A 3 -lines in the latter case.
For the system at hand, several control parameters can be tuned: the fraction M a =M of partly active rings; the ratio T h =T c of the temperatures of the hot and cold segments; the fraction N h =N of hot segments on a ring; and the number of monomers N of the rings. The richness of the system makes the possibility of existence of higher-order singularities in principle possible. A detailed investigation in the vast space spanned by these is beyond the scope of this work; we focused mainly on the first two cases above. In Figs. 8a, b, we show the effect of gradually increasing the fraction of partly active chains, which induces a glassy state, as witnessed by the saturation of the mean-square displacement (Fig. 8a) and the growth of a non-ergodic plateau (Fig. 8b) as the ratio M a =M exceeds a number as small as 1=160. There is no evidence of the development of an intermediate plateau in the ergodic state preceding the glass, in full analogy with type-A transitions seen in the aforementioned systems [36][37][38][52][53][54] . The presence of a subdiffusive regime in the mean-square displacements (Fig. 8a) and of a logarithmic crossover intermediate scattering function (Fig. 8b) offer additional corroboration that the transition for this choice of the remaining system parameters is continuous, and thus higher-order singularities are present. We have found similar behavior (not shown) varying the ratio T h =T c . Naturally, this does not rule out that in other parts of the phase diagram the transition is governed by A 2 -singularities and it is thus of B-type. This would give rise to a number of additional scenarios for the behavior of the relaxation functions, including the possibility of multiple relaxations observed in related models 54,55 . The presence of deep and tight threadings in our system bears an intuitive analogy with colloidal attractions 53 , random pinning [36][37][38]51 or polymer-mediated bond formation 55,56 in systems featuring similar glass-transition phenomenologies. A detailed investigation of this issue, however, is a problem for the future.
The present active topological glass is remarkable by its distinct role of activity in comparison to the known active glasses composed of self-propelled particles 12,14 . There, the activity opposes the glassiness as indicated by the increase of the effective temperature defined through a long-time limit of a generalized fluctuation-dissipation relation. At the same time, however, increasing the persistence time can lead to a decrease of the effective temperature and therefore promote the glass formation, and conversely, decrease in persistence time favors fluidization. In contrast, the microscopic model of the active topological glass has zero persistence time, but nevertheless drives the vitrification. This can be the consequence of the polymeric nature of the particles and the topological constraints that together create some persistence as illustrated by the superdiffusive regime. Moreover, the activity clearly causes increase of the number and the severity of topological constraints. This 'topological crowding' then can be viewed as a specific, effective attraction or pinning that further promotes the glass as mentioned above.
As regards the system with only a fraction of active rings present, the threading cluster analysis (more details in Methods and Supplementary Note 6) has revealed that it is always the active rings that are involved in the formation and maintenance of the largest cluster (Fig. 8c-e). The onset time of the glassy regime increases with decreasing M a (Fig. 8a). This delay is related to the slower building up of the largest cluster as fewer active chains participate ( Supplementary Fig. 10). The ring length N governs the number of topological interactions (see Supplementary Note 7 for varying ring length and Supplementary  Fig. 12). Similar to linear polymer melts in equilibrium, due to extended chain configurations, each partly active ring overlaps with a number (extensive in N) of other rings. As a cooperative unthreading of rings is necessary for the relaxation to occur, we expect the relaxation time to grow at least exponentially with N 6 . Such a strong dependence is also known from the melts of polymeric stars, where it is due to the slow process of arm retraction that, however, can take place even without cooperative motion of the other chains 57 . While our model system is interesting from the fundamental physics point of view, it is also inspired from biology and may well bear important ramifications for the organization of the DNA fiber (chromatin) of higher eukaryotic cells. The large-scale static properties of the equilibrium melt of unknotted rings, such as the territorial organization, the scaling of hR g i with N or the so-called contact probability (Supplementary Fig. 1d) are consistent with the population-average conformation of the interphase chromatin 58 . This can be due to the common governing role of the topological constraints in both systems 58 . The rationale behind the ring model is the timescale separationthe chromosomes are linear chains and as such equilibrate and tangle by reptation, however, for the length and density of chromatin, such relaxation would take significantly longer than the cell's life time 59 . Therefore, the constraints arising from the uncrossability of the chains can be modeled as permanent on biological time scales. This is done effectively by the closure of the chain's ends which inhibits the reptation. Furthermore, the chromatin association with the nuclear lamina 60 hinders reptation relaxation and the chromatin can be viewed as loops between lamina contact points. In contrast, the topoisomerase II enzyme can resolve the topological constraints, by cutting the fiber, passing segment through sealing up the cut back 61 , although the extent to which it affects global topology on biological time scales in interphase is unclear. Alternatively, the rings can also represent small scale chromatin loops extruded or maintained by Structural Maintenance of Chromosomes protein complexes 62,63 , and/or Topologically Associating Domains 64 that do not link. Current experimental evidence for the topological state (knottedness) of the chromatin fiber also varies. While conformations inferred from contact probability measurements exhibit knots 65 , the knot analysis 66 of the direct observations of fluorescently labeled chromatin segments finds mostly unknotted segments. Unable to refine the scales of the topological constraints completely at this point, we assumed their existence for the typical observation times and examined the consequences. Therefore, we mention phenomena on chromatin at various scales, for which the interplay of the topology and the activity could be relevant. The segmental activity with thermal-like fluctuation spectra has been measured to be stronger in the normal living cell nuclei 67 as opposed to energy-deprived cells, and has its origins in the energy dependent processes, such as chromatin repair or remodeling, DNA transcription, or loop extrusion. We conjecture that the phenomena observed in our partly active system could also be relevant in biological context, on the basis of the following similarities with our model: genes exhibit size increase upon transcription decoupled from the chromatin decondensation 68 ; a highly transcribed gene shows directed motion 69 ; overall chromatin loci exhibit heterogenous subdiffusive dynamics 67 reminiscent of glassy behavior 23,70 ; the active and inactive chromatin are spatially segregated 71 ; and the chromatin exhibits a doubly folded structure at small scales 72 . Naturally, these effects have also alternative explanations. The phase separation and glassy dynamics could be observed for a copolymer models where different chromatin segments have different interaction potentials based on their epigenetic state 23,70,73,74 or by interaction with binding proteins 74,75 . The double folded structure is likely to be attributed to supercoiling due to the torsional stress induced on the fiber in the process of transcription or loop extrusion 62,76,77 . Nevertheless, we show that the activity in combination with topological constraints at the fiber level can complement the observed phenomena and should be considered in a more complete picture of the chromatin organisation 78,79 . As the models above typically do not consider topological constraints and activity, our findings represent a completely new mechanism for the observed phenomena. Certainly, more work is required to determine the relative contributions of the different mechanisms in the various cases. Potentially, the different nature of the observed glassy states, namely the monomer glass due to caging in copolymer models 23,70 and topological glass here, could be used to discern which one could be present or dominant in the chromatin context. Activity complemented with topological constraints at the microscopic level can lead to rich system dynamics. We have demonstrated this on a model system of dense, unknotted and nonconcatenated ring polymers with active segments. First, the directionality of polymers in a dense environment arises from the topological constraints and the isotropic noises of unequal strength. Similar to ref. 2 , the superdiffusive motion is connected to a major particle deformation at high density. In contrast, however, in our work, the superdiffusivity is triggered by the active noise. Second, a novel state of matter-the activitydriven topological glass-arises from the activity-enhanced ring threading. In contrast to the conjectured equilibrium topological glass 5 , our present model allows the creation of a topological glass for rings of accessible lengths (7N e ) using activity. Moreover, only a low number of partly active rings is necesseary, making the present model suitable for experimental testing with extracted bacterial DNA 80 or synthetic ring polymers 3 , driven by molecular motors fueled by ATP hydrolysis. The effective temperature ratio T h =T c of a factor of three that we used here is within reach, since ATP hydrolysis releases more than 10k B T 81 . However, as our preliminary results suggest, even smaller temperature ratios can be sufficient. Other means of selective heating could be attempted by fluctuating external fields or infrared irradiation selectively coupling to individual groups in the polymer. The latter mechanism can yield a fluid material with reversible vitrification upon light exposure. What is the proper topological order parameter of the glass transition, or what are the fragility properties of the active topological glass, are just a few intriguing questions to be addressed in the future. The present work paves the way for a development and investigation of these novel, molecular, topology-based, responsive materials.

Methods
Model. All particles interact via a purely repulsive Lennard-Jones potential where θðxÞ is the Heaviside step function. The connectivity of polymers is maintained by the finitely extensible nonlinear elastic potential where K ¼ 30ε=σ 2 and r m ¼ 1:5σ. These parameters make the chains essentially uncrossable. The angular potential is with k θ ¼ 1:5ε. The studied systems are monodisperse at fixed volume with the total monomer density ρ ¼ 0:85σ À3 . We used very large systems of M ¼ 1600 chains to avoid unphysical self-threadings of extended rings due to periodic boundary conditions. As shown in Supplementary Fig. 1c, our systems are large enough to assure that the rings are smaller than the simulation box at all times. All the simulations were performed at constant volume with two Langevin thermostats using the large-scale atomic/molecular massively parallel simulator (LAMMPS) engine 82 . The equations of motion were integrated with the time step Δt ¼ 0:005τ, where τ ¼ σðm=εÞ 1=2 .
To model heterogeneous activity of the rings, on each polymer a consecutive segment of length N h ¼ N=8 monomers is considered active by subjecting it to stronger thermal-like fluctuations (isotropic, uncorrelated, white noise) in comparison to the rest (N À N h ) monomers of the ring that remain passive. In particular, the active monomers are connected to a Langevin thermostat of temperature T h ¼ 3 (units of ε and the Boltzmann constant set to unity are being used throughout), while passive ones are coupled to a second Langevin thermostat with temperature T c ¼ 1. The coupling constants of both thermostats are γ ¼ ð2=3Þτ À1 . From our earlier study 83 we know that such values of γ and T h lie in the range of the onset of a non-equilibrium microphase separation in the activepassive mixtures of polymers of 40 monomers, but colloidal systems would not demix 29,31 . Our preliminary studies show that even a weaker temperature contrast is sufficient for the glassy state to occur. We leave a more detailed characterization of the phase diagram for the future study.
Model details. The employed model was frequently used not only for the melt of linear chains but also for the melt of rings 27,28 at T ¼ 1. Therefore, we already know a range of useful properties of the equilibrium system, such as the entanglement length N e ¼ 28 ± 1 and the typical diffusion times of the rings. In equilibrium, the diffusion time τ diff ðNÞ of a ring is defined as the mean time required for its center of mass to diffuse over 2R g , where is the mean radius of gyration. Above, r i denotes the position of the i-th monomer and R is the position of the center of mass of the ring. The angles hÁ Á Á i stand for the average over different chains. As reported in refs. 27,28 , To quantify dynamical evolution of the system, we consider squared displacements of the centers of mass of the rings, g 3 ðt; t 0 Þ, at a given time t provided that the measurement started at t 0 (t 0 ¼ 0 is the activity onset, or, in the case of cut rings, the moment of the cutting): where RðtÞ is the position of the center of mass of a ring at time t with respect to the center of mass of the whole system at that time. Typically, we report the mean squared displacement hg 3 ðt; t 0 Þi averaged over the rings only, i.e., without additional averaging over multiple time origins as the dynamics of the system are not stationary in general. The relaxation dynamics of the system is also characterized by the self-part of the intermediate scattering function defined as where R m is the position of the center of mass of the m-th ring. We evaluated it at q Ã σ ¼ 0:35 corresponding to the maximum of the static structure factor (see Supplementary Fig. 6 for other q-values).
We define ergodicity breaking parameter as EBðtÞ ¼ hg 3 ðt; t 0 ; ΔÞ 2 i À hg 3 ðt; t 0 ; ΔÞi 2 hg 3 ðt; t 0 ; ΔÞi 2 : Here, g 3 ðt; t 0 ; ΔÞ, being mainly a function of the lag time Δ, represents the time average: where R is the position of the ring's center of mass with respect to the global center of mass.
Threading analysis. The minimal surfaces are spanned on the fixed contours of the rings from the molecular dynamics simulations and then minimized using the overdamped surface tension evolution under the constraint of fixed disc topology. We used the Surface Evolver software 84 and followed the protocol in 47 ensuring that for these ring lengths the area of the final surface is close to the minimum. The separation length is defined as where L t i is the (threading) length between the i-th and the (i þ 1)-th penetrations of the surface (see Fig. 5a). L sep approximates how much of the threading rings material is on one side of the threaded ring. The approximation lies in the assumption that two following surface piercings are in the opposite direction with respect to the surface normal vector. In many cases, there are only two surface penetrations between two rings (see Supplementary Fig. 3) and in such case the approximation is exact because of the non-concatenation constraint of the rings conformations.
The threading cluster analysis is performed in the following way. At first, each ring is analyzed for threadings with other rings. Two rings are assigned to the same cluster if at least one of them threads the other one with L sep > L cutoff . Then, the size (in terms of the number of members) of each cluster is determined and the biggest cluster is found. For mixtures of active and passive rings the size of the biggest cluster containing any kind of rings is considered (for example, Fig. 8b) and separately the biggest cluster of those clusters that contain only passive rings is calculated, such as in Fig. 8c.
To characterize the dynamics of threadings we compute the threading gain Δn þ th ðtÞ and loss Δn À th ðtÞ as function of time. To do so, we compare the set TðtÞ of pairs of rings that are threaded at time t with the set Tðt þ ΔtÞ of threaded pairs at time t þ Δt. Then Δn þ tn ðtÞ ¼ jTðt þ ΔtÞ À TðtÞj and Δn À tn ðtÞ ¼ jTðtÞ À Tðt þ ΔtÞj, where jxj is the number of members of set x. Note that Δn ± th ðtÞ is different from the derivative of the number of threaded neighbors, because there exist a state with the derivative zero, but high threading exchange, which is exactly the case in the steady state. For all calculations of the threading dynamics, we used Δt ¼ 2:5 10 4 τ and maximum simulation time 3:4 10 6 τ.
A related quantity characterizing the threading correlation is Φðt; t 0 Þ ¼ 1 n tn ðt 0 Þ P n tn ðt 0 Þ i;j¼1 T ij ðt 0 ÞT ij ðt 0 þ tÞ, where T ij ðtÞ is one, if threading between rings i and j exists at time t and zero otherwise, and n tn ðtÞ is the number of threaded neighbors at time t. The Φðt; t 0 Þ gives the time correlation function for the existence of threadings.
In addition, we compute the threading survival time distribution πðt; t 0 Þ measured from different t 0 . To construct πðt; t 0 Þ, we look at all threaded ring pairs at time t 0 and track how long they remain threaded before they unthread for the first time.