Shaping human telomeres: from shelterin and CST complexes to telomeric chromatin organization

Organization and DNA-binding properties

The majority of shelterin subunits interact minimally with other subunits of the complex, with each subunit having a domain that interacts with adjacent subunit(s) to form a branched structure (FIG. 2a). The telomeric ssDNA-binding domain of shelterin lies mainly in the amino-terminal OB1 and OB2 domains of POT1 (REF.¹⁸) (FIG. 2b), and the addition of TPP1 increases the affinity of POT1 for DNA tenfold^57,58. (An OB domain or OB-fold domain is a small oligonucleotide-binding or oligosaccharide-binding structural motif found in many proteins.) Both POT1 by itself and POT1–TPP1 have a preference for capping the telomeric 3′ end, but they also bind to internal sites of telomeric ssDNA⁵⁹. POT1 interacts with amino acids 266–320 of TPP1 (TPP1_266–320) through two structured domains (across POT1 OB3 and Holliday junction resolvase-like (HJR) domains)^60,61, with no known interactions with other shelterin subunits. The discovery of an HJR domain in POT1 (REFS^60,61), which is structurally adjacent to its telomeric ssDNA-binding domain (FIG. 2c), is intriguing given the possibility that a T-loop can progress to become a telomeric Holliday junction⁶². The TPP1 N-terminal OB domain contains the TPP1 glutamate (E) and leucine (L)-rich patch (TEL patch), which recruits telo merase to telomeres^32–34, whereas its carboxy terminus (TPP1_510–544)^15,46,53 interacts with the TIN2 N terminus^63,64 (TIN2N). TIN2, which is effectively the hub of shelterin assembly, links TPP1 with TRF1 and TRF2 homodimers. The TIN2N domain, which structurally resembles the TRF homology (TRFH) domains of TRF1 and TRF2, interacts with both TPP1 and TRF2_350–366 (REF.⁶³). In addition, another region of TIN2 (TIN2_256–276) interacts with the TRF2 TRFH domain⁶⁵. Interestingly, whereas the TRF1 and TRF2 TRFH domains are structurally similar⁶⁶, they do not share a similar level of interaction with TIN2 (REFS^43,65): the TRFH domain of TRF1 binds TIN2_256–276 20-fold tighter than the TRFH domain of TRF2 (REF.⁶⁵).

Both TRF1 and TRF2 homodimers have a Myb DNA-binding domain at their C terminus for sequence-specific binding of telomeric dsDNA (YTAGGGTTR; Y = C or T, R = A or G)^{48,49,51,67–70}. The TRFH domain is connected to the Myb domain by a long polypeptide chain (~100 amino acids in TRF1 and ~200 amino acids in TRF2), which has no apparent structural features; this flexibility allows each Myb domain of a TRF1 homodimer or of a TRF2 homodimer to bind telomeric DNA with various distances and orientations with regards to the other Myb domain^20,67,71 (FIG. 2c). This flexibility may be crucial for TRF1 and TRF2 binding chromatin comprising nucleosomes with short linker DNAs, as discussed below (see Other human telomere proteins).

TRF1 and TRF2 differ in their N termini. TRF1_1–67 is acidic (Asp–Glu-rich), whereas TRF2_1–45 is basic (Arg-rich). This difference translates to differences in how TRF1 and TRF2 interact with DNA or mediate the formation of special DNA architectures. The TRF1 N-terminal acidic domain has been implicated in the recruitment of the poly(ADP-ribose) polymerase tankyrase 1 (REF.⁷²) and inhibition of telomeric DNA condensation⁷³. The N-terminal basic domain of TRF2, which is better studied, is involved in the protection⁷⁴ and regulation^75,76 of T-loops, interaction with histones⁷⁷ and inhibition of DNA double-strand break repair by non-homologous DNA end-joining in conjunction with RAP1 (REF.⁵⁴). Given its positive charge, one might suggest that the basic domain participates in the interaction of TRF2 with canonical dsDNA such as telomeric DNA. However, the TRF2 N-terminal basic domain does not directly participate in TRF2 recruitment to telomeres but is instead involved in engaging non-canonical DNA structures such as T-loops⁷⁸, Holliday junctions⁷⁹ and G-quadruplexes (G4s)^80–82. The proposed ability of RAP1 to tune the specificity of TRF2 for telomeric dsDNA by neutralizing the basic domain⁸³ suggests that RAP1 has other roles in regulating the affinity of TRF2 to these non-canonical DNA structures.

Shelterin possesses the special ability to engage both telomeric ssDNA and dsDNA. Consequently, mammalian shelterin complexes consisting of POT1 with either TRF1 or TRF2 exhibit high affinity to telomeric DNA comprising both ssDNA and dsDNA regions^53,84,85, and the minimal reconstituted human shelterin complex (shelterin_core, consisting of TRF2–TIN2–TPP1–POT1) binds ssDNA–dsDNA junctions with nanomolar affinity⁵³. Shelterin_core binding to telomeric ssDNA–dsDNA junctions is not easily competed by telomeric ssDNA or dsDNA⁵³. This provides the biochemical basis for how the shelterin complex could specifically target various telomeric DNA structures that present both ssDNA and dsDNA regions and provides insights into how shelterin recruits telomerase to the telomeric tail (for a detailed discussion, see Telomerase inhibition and stimulation).

Architecture of the shelterin complex

Although the atomic structures of several shelterin protein domains have been solved^{18,57,60,61,63,66,68,86,87}, the overall architecture of mammalian shelterin as a complex remains unknown. Furthermore, two fundamental questions remain: which shelterin complexes and subcomplexes are functional in vivo, with what subunit stoichiometry; and with which telomeric DNA structures does each of them interact?

Telomerase inhibition and stimulation

Shelterin has multiple roles in regulating telomerase activity at telomeres. On the one hand, shelterin can mediate T-loop formation^20,22 or end-cap the telomeric 3′ tail^17,47,52,59 to restrict telomerase access. On the other hand, it can recruit telomerase to the telomere^32–34 and drive telomerase processivity^53,57. The recruitment of telomerase to the telomere by shelterin is mediated directly by the TEL patch on the TPP1 OB domain^32–34. The recruitment process occurs in the S phase of the cell cycle, is highly dynamic and involves multiple kinetically distinct stages. Single-molecule live-cell imaging of human telomerase reverse transcriptase (hTERT; the catalytic protein component of telomerase) in cells revealed that telomerase samples telomeres multiple times using a 3D diffusive search mechanism, before stably engaging the telomeric tail through base pairing with the template region of the telomerase RNA (hTR) component^105,106; both the probing interactions and stable binding require interaction with the TEL patch. Similar experiments with hTR corroborated the hTERT observations and also revealed early stages of hTR biogenesis and telomerase assembly¹⁰⁷.

Although the direct role of TPP1 in telomerase recruitment is starting to be understood, its own recruitment to telomeres and assembly with the other components of shelterin are less clear. In vitro studies show mouse shelterin complexes sampling the DNA in a 3D search before stably engaging telomeric sequences⁸⁵. Another study has shown that human TRF1 and TRF2 exhibit slower diffusive behaviour on relatively short regions of telomeric DNA¹⁰⁸. Clearly, the mobility of shelterin on telomeric DNA and on telomeric chromatin, including sliding and hopping, are important questions for the future. For example, given the repetitive sequence of the telomere, how does TPP1 help telomerase find the telomeric tail instead of ‘distracting’ it to internal sites? The DNA sequence might be the same but the structure at the end of the telomere is unique, where both telomeric ssDNA and dsDNA exist (FIG. 1). Hence, the answer might lie in the necessity of an assembled shelterin complex consisting of TRF1 and/or TRF2, and POT1, which would direct the complex to search for the telomeric ssDNA–dsDNA junction based on the virtue of highest binding affinity⁵³ and eventually directing telomerase to the telomeric end.

Interestingly, despite its high affinity to telomeric ssDNA, POT1 localization to telomeres in vivo depends on its association with TPP1 (REFS^45,46,52), indicating that TPP1–POT1 likely exists as a heterodimer during its recruitment to telomeres. Nonetheless, the recruitment of TPP1 to the very end of the telomere may involve multiple stages. For example, as an initial step, TPP1–POT1 could transiently associate with TIN2–TRF1–TRF2 complexes that are bound along telomeric dsDNA regions, sliding or hopping along the telomeres until it finds the telomeric ssDNA–dsDNA junction; it would then form a particularly stable protein–DNA complex, in which TPP1–POT1 is bound to the ssDNA and the TRFs are bound to the dsDNA. Whether TPP1–POT1 located at the ssDNA–dsDNA junction is close enough to the DNA 3′ end to load telomerase remains to be determined.

A particular conundrum is how shelterin switches between its inhibitory and stimulatory roles in regulating telomerase activity at telomeres. Insights might come from the dynamic multiple activities of POT1: it is essential for telomeric ssDNA protection by preventing aberrant accumulation of replication protein A (RPA)^109,110; it participates in cell-cycle coordinated RPA removal through TERRA and heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1)¹¹¹; and it associates with TPP1 to stimulate telomerase processivity⁵⁷. POT1 function depends on its positioning on telomeric ssDNA. For example, POT1 binding to internal telomeric ssDNA allows telomerase to bind to the DNA 3′ end, whereas POT1 binding to the 3′ end of telomeric ssDNA excludes telomerase⁵⁹. The 3′ end binding by POT1–TPP1 gives telomere protection, whereas the internal binding allows telomerase recruitment and stimulation.

However, the role of POT1 extends beyond just positioning, as it is also able to disrupt G4 formation by telomeric ssDNA^112,113, which otherwise can exclude telomerase binding^113,114. The DNA-binding behaviour of POT1 is further changed upon association with TPP1, causing POT1 (together with TPP1) to slide back and forth along the ssDNA to iron out G4 structures¹¹⁵. But is G4 always an inhibitory structure to telomerase? Apparently not. Recent studies have shown that a G4 is formed in nascent telomeric ssDNA synthesized by telomerase and can aid telomerase processivity^116,117. It was further proposed that TPP1–POT1 interaction with this G4 can facilitate the fidelity of telomerase translocation and nascent DNA protection¹¹⁶, which would contribute to TPP1–POT1 stimulation of telomerase processivity¹¹⁸. The intermediate-resolution model of human telomerase¹⁰⁴ already suggests the existence of a cavity to house this G4 (REF.¹¹⁶). More recently, single-molecule fluorescence studies have shown that the telomerase RNA template can directly interact with parallel-stranded G4 structures and facilitate G4 disruption upon telomerase translocation¹¹⁹. A high-resolution structure of telomerase in conjunction with shelterin complexes would provide more insights into these complex processes.

Promiscuity of human CST ssDNA binding

ssDNA-binding proteins such as RPA and its bacterial homologue SSB have multiple ssDNA-binding modes^128,129. Thus, it is perhaps not surprising that CST, which possesses multiple OB domains, exhibits similar ssDNA-binding behaviour^{37,38,130–132}, giving it the ability to engage a plethora of DNA sequences and structures^{35,37,133,134}. A peculiar finding is that CST binds preferentially to ssDNA–dsDNA junctions, which may explain how CST recognizes specialized DNA structures at DNA replication and breakage sites¹³³. CST binds short ssDNA (<30 nucleotides) with specificity for G-rich sequences, such as telomeric sequences^35,134. It can also engage non-telomeric DNA sequences^37,133 within longer ssDNA (>30 nucleotides). One model to explain the promiscuous DNA-binding properties of CST postulates that CST has a core site that binds tightly to G-rich ssDNA and also has one or more secondary sites that can accommodate longer ssDNA with less sequence specificity. Indeed, the recent cryo-EM structure of human CST revealed a single telomeric ssDNA-binding site in OB-F and OB-G of the CTC1 subunit, and the presence of five additional OB domains¹³² (FIG. 3a,b). This model would also explain how CST can engage ssDNA both at telomeres and genome-wide. Other than binding to extended ssDNA structures, CST is also able to directly engage and unfold G4 DNA structures¹³³. This function is crucial for the ability of CST to restore stalled replication forks at telomeres (and genome-wide), presumably by removing the obstacles formed by G4 at the lagging strand¹²⁴.

Although each subunit of CST has OB domains (which individually interact with ssDNA to some extent), its functional ssDNA-binding capacity depends on its heterotrimeric assembly^35,135,136. This could be a functional manifestation of CTC1 topology, whereby STN1–TEN1 stabilizes the CTC1 C-terminus region that contains the core ssDNA-binding site of CST¹³² (FIG. 3b). This is consistent with previous experiments showing TEN1 stabilizing CTC1–STN1 ssDNA binding¹³⁷. It is also possible that CST achieves functional ssDNA binding by leveraging multiple cooperative binding sites across the three subunits, similar to RPA¹³⁸. The current cryo-EM structure of the full CST complex only reveals four nucleotides bound by CTC1, despite the presence of a longer, 18-nucleotide telomeric ssDNA¹³². CST needs more than four nucleotides to stably engage telomeric ssDNA³⁵, and thus it will be important to determine how and where else CST interacts with ssDNA and how ssDNA wraps around CST. This will not only provide insights into how the heterotrimeric assembly enhances CST affinity for ssDNA^37,131 but also how CST can mediate its functions at telomeres and genome-wide at sites of DNA replication and DNA damage repair.

Upon binding to telomeric ssDNA, human CST forms a ring-like decameric DNA–protein supercomplex (FIG. 3b), in which telomeric ssDNA facilitates formation of CST dimers (key building blocks of the decamer)¹³². Whether the known functions of CST are accomplished by this supercomplex or by CST heterotrimers is currently unknown. However, given the intricate assembly pathway of the decamer and its potential to bury and compact a substantial length of ssDNA, we suggest that decamerization may be involved in DNA protection and in pol α-primase recruitment by CST. For example, a decameric form of CST could recruit multiple pol α-primase proteins (FIG. 3c) to a single telomeric ssDNA site, which would initiate several concurrent and coordinated C-strand fill-in processes, leading to a rapid fill-in of telomeric ssDNA. Loading of multiple DNA polymerases has been observed in the T4 replisome, whereby two DNA polymerases play tag to facilitate a smoother turnover for the lagging-strand DNA synthesis process¹³⁹. Interestingly, yeast CST also forms a ring-like protein complex, albeit as a dimer, but it is unclear whether ssDNA binding is needed for yeast CST dimerization¹⁴⁰.

CST relationship to RPA-like complexes

The human CTC1 subunit has seven OB domains and interacts with STN1 and TEN1 subunits in a 1:1:1 stoichiometry in the heterotrimeric complex¹³² (FIG. 3b). The stoichiometry is similar to that of human RPA and Tetrahymena thermophila and budding yeast CST^{103,128,140,141}, but not to Candida glabrata CST¹⁴². The architecture of human CST indicates that the C-terminal half of CTC1 serves as a structural scaffold for STN1 and TEN1, in which STN1 serves as an adapter between TEN1 and CTC1 (REF.¹³²) (FIG. 3b). Two peculiar findings are that the N-terminal and C-terminal domains of STN1 (STN1n and STN1c, respectively) separately interact with CTC1 and that STN1c can switch between two separate docking sites on CTC1, which is presumably decided by CST binding of ssDNA¹³². This is the first structural insight into a complete RPA-like complex, and it will be important to determine whether other STN1 homologues (for example, RPA32 or T. thermophila P45) possess this unique architectural assembly, especially as they have very similar structural topology^131,141,143.

Unlike STN1–TEN1, CTC1 (the largest subunit in the heterotrimeric complex) is more structurally diverse among RPA-like complexes across species^{131,132,140,141,144}, suggesting that CTC1 has evolved to accommodate various distinct functions. Based on telomere functionality, the budding yeast Cdc13 has often been described as a potential structural homologue of mammalian CTC1, despite the lack of sequence similarity¹⁴⁵. The cryo-EM structure of human CTC1 now shows that it resembles the topology of RPA70 more than that of Cdc13 (REFS^132,140,144), suggesting that human CTC1 and RPA share a similar evolutionary origin distinct from that of Cdc13. Human CTC1 also contains a conserved zinc ribbon motif in its OB-G domain (FIG. 3a)¹³², which is similar to that found in RPA and T. thermophila Teb1 (RPA70 homologue). This motif structurally participates in the stimulation of T. thermophila telomerase processivity, although it does not directly bind telomeric ssDNA¹⁴⁶. The function of the zinc ribbon motif in CST, as in the case of RPA¹²⁸, is currently unknown but may involve stimulation of pol α-primase processivity. More studies will be needed to explore the function(s) of the CST zinc ribbon motif, particularly its roles in pol α-primase recruitment and stimulation^39,147.

Terminating telomerase activity by CST

A proposed mechanism of termination of telomerase activity is competition between telomerase–TPP1–POT1 and CST for telomeric ssDNA, with CST effectively displacing telomerase from its substrate³⁵ (FIG. 3d). Supporting this model, in vivo studies revealed that a CST disease mutation elevates telomerase recruitment to telomeres¹³⁶ and that CST recruitment to telomeres coincides with telomerase shut-off at the late S/G2 phase³⁵. As every telomere is likely extended by telomerase only once per cell cycle^105,148, it is premature to say whether CST reduces telomerase activity by preventing re-engagement of telomerase to telomeres or by disrupting an elongating telomerase complex. In addition, it is interesting to note that mouse and human use different shelterin-mediated pathways to recruit CST to telomeres: mouse uses POT1b (REF.¹⁴⁹) whereas human uses TPP1 (REF.¹⁵⁰).

Given that G4 formation at a nascent telomeric tail can aid telomerase processivity¹¹⁶, an interesting question is whether CST unfolding of G4 has any role in the termination of telomerase activity. How CST differs from TPP1–POT1, which can passively disrupt G4 (REF.¹¹⁵), to achieve opposite regulatory effects on telomerase activity is unclear (FIG. 3d). One possibility could be that the actions of TPP1–POT1 and CST are applied at different stages of the telomerase catalytic process. In addition, direct interactions between CST and shelterin³⁵ may also have a role. Further insights might come from understanding how CST and POT1 engage and unfold G4 structures. To reveal such details, it would be important to probe the dynamic interplay between CST, shelterin and telomerase on a single telomeric substrate.

Insights into how human CST interacts with telomerase may come from T. thermophila, where CST is surprisingly an integral part of the telomerase RNP^103,145. Furthermore, T. thermophila P50 (a putative TPP1 homologue¹⁵¹) is also part of the telomerase holoenzyme¹⁰³ and directly interacts with CST (similarly to human TPP1 (REF.³⁵)). Thus, if interactions occurring within a single T. thermophila telomerase holoenzyme are occurring intermolecularly in human telomeres, several insights are revealed. First, analogous CST–TPP1 interactions may contribute to human telomere length regulation^35,150,152. Second, the human CST ssDNA-binding site would face towards the 3′ end of the ssDNA, as proposed in the T. thermophila telomerase holoenzyme model¹⁰³, where it could plausibly bind the product of telomerase extension. In addition, human and T. thermophila CST both have a flexible STN1c domain, which, based on the proposed human CST decamer assembly pathway¹³², could have a regulatory role in CST interactions with telomeric ssDNA and telomerase. Structures of mammalian telomerase with CST and shelterin bound to a single piece of telomeric ssDNA would test these speculations.

CST coordination of C-strand fill-in

It is often proposed that CST participates in the recruitment of pol α-primase to telomeres, presumably through CST interactions with shelterin or telomeric ssDNA^{35,120,137,152,153}, although this has been challenged¹²¹. Even though co-immunoprecipitation studies have pointed to STN1, shelterin proteins and telomerase as potential candidates to recruit pol α-primase to telomeres¹⁵⁴, it remains unclear whether these are direct protein–protein interactions or mediated by telomeric DNA associations. A clearer picture exists for pol α-primase recruitment to DNA replication origins, where the replication initiation factors MCM10 and AND-1 mediate its recruitment^155,156. Interestingly, recent studies have implicated CST in the assembly of AND-1 and pol α-primase at replication origins, by forming direct interactions with the MCM complex, AND-1 and pol α¹²². In addition, CST also participates in recruiting pol α-primase to DNA double-strand break sites by interacting with the shieldin complex^126,127. All of the above findings support the hypothesis that CST has a direct role in recruiting pol α-primase to telomeres.

Pol α-primase serves two primary functions: the primase catalytic subunit first synthesizes an RNA primer, which is then used by the polymerase catalytic subunit for DNA polymerization. Protein recruitment aside, CST is also involved in stimulating the pol α-primase switch from RNA to DNA synthesis^39,157,158. This represents only part of the C-strand fill-in process, the entire mechanism of which remains unclear. Some key questions about C-strand fill-in include how many copies of CST and pol α-primase are present at a single telomeric site, how CST helps primase to initiate RNA synthesis and then switch to DNA synthesis, and whether a G4 formed along the internal telomeric ssDNA will be an obstacle for DNA polymerization by pol α-primase. Answers to these questions would come from structural and biochemical studies of larger assemblies involving CST, pol α-primase, telomerase and shelterin.

Some insights are obtained when CTC1 residues that are involved in the pol α-primase interaction¹³⁵ are mapped onto its structure, revealing that CST associates with pol α-primase across at least two separate sites¹³². Interestingly, eukaryotic pol α-primase has a flexible bilobal architecture¹⁵⁹, which leads us to speculate that CST could serve as a scaffold to support pol α-primase’s architecture and, possibly, also provide a mechanism to regulate the transfer of RNA primers by modulating the distance between the catalytic and primase subunits of pol α-primase^39,158,159.