neRNAP subunit-encoding open reading frames were identified using existing data base annotations and tblastn searches of the Nanoarchaeum equitans genome sequence (SeqID: AE017199.1; see Supplementary Table S1 for more details).

Markov Chain Monte Carlo MCMC simulations were carried out as described previously (Sullivan and Weinzierl, 2020). Briefly, the simulations employed the PROFASI forcefield in the PHAISTOS package (Boomsma et al., 2013). Due to the origin of the proteins from hyperthermophilic organisms, the simulation temperature was set to 355 K (81.85°C). The resulting trajectory data (based on 50,000 calculated states per simulation) was analyzed for secondary structure elements using cpptraj (Roe and Cheatham, 2013) and processed/visualized with custom scripts on Python Jupyter notebooks.

The protein-coding portions of RNAP subunits A′, A″, B′, B″, D, H, L, N and p were PCR amplified from purified N. equitans genomic DNA (a gift from Prof. M. Thomm, University of Regensburg) as full-length, non-tagged sequences and cloned as NdeI-BamHI (neA″, neB′, neB″, neF, neK), or NdeI-EcoRI (neA′, neD, neE, neH, neL, neN, neP) fragments into the bacterial expression vector pET21a. Recombinant proteins were expressed with IPTG-induction in E. coli BL21-DE3 Rosetta 2 (Merck) under standard conditions (Werner and Weinzierl, 2002). Subunits neA′, neA″, neB’ and neB″ were purified as insoluble inclusion bodies. Briefly, bacterial cells expressing these recombinant subunits were resuspended in T/G₀ (25 mM Tris-base, 200 mM glycine, 10 mM magnesium acetate, 100 μM zinc acetate, 14 mM ß-mercaptoethanol and 10% glycerol at pH7.5) and sonicated. The inclusion bodies were washed extensively with 1 x deoxycholate buffer (1 mg/ml deoxycholate, 15 mM ß-mercaptoethanol) and water/15 mM ß-mercaptoethanol before solubilizing them in T/G₀ in the presence of saturating urea or 6 M guanidine-hydrochloride. Subunits neD, neL, neH, neN and neP were expressed similarly as soluble recombinant proteins. Bacterial cells expressing these recombinant subunits were resuspended in P300 Buffer (300 mM potassium acetate, 20 mM Tris-acetate pH 7.9, 10 mM magnesium acetate, 100 μM zinc acetate, 14 mM ß-mercaptoethanol and 10% glycerol) and sonicated. The supernatant containing the solubilized proteins were heat-inactivated of at 70°C for 10 min to precipitate the bacterial proteins present in the extract (the hyperthermophilic ne subunits remain completely soluble during this treatment).

The urea-solubilized inclusion bodies, or the soluble subunits, were passed over ∼5 ml SP- or Q-Sepharose (Fast flow, Amersham) in chromatography columns. Proteins were eluted in a salt gradient from T/G0 to T/G1000 using a DuoFlow BioLogic FPLC system (BioRad). The purified subunits were assembled by mixing them in the presence of 8 M urea in a dialysis cell (Slide-A-Lyzer 3500MCOW frames [Pierce], or 96-well microdialyser (SpectraPor) on a Theonix robotic platform (Aviso) for high-throughput assembly (Nottebaum et al., 2008; Weinzierl, 2013)), followed by lowering the urea concentration by gradual dilution in the dialysis buffer (Werner and Weinzierl, 2002; Naji et al., 2007; Nottebaum et al., 2008; Weinzierl, 2013). Equimolar amounts of the large subunits were mixed with small subunits, which were in at least four-fold excess to the large subunits, under denaturing conditions. The highest yield of enzymatically active neRNAP (due to increased folding efficiency) was obtained in the presence of 500 mM salt (either sodium chloride, potassium- or sodium acetate) in the refolding buffer (Supplementary Figure S1). The assembly of large complexes was monitored by analytical size exclusion chromatography on Superose-6 and Superose-12 columns (Amersham) at a flow-rate of 0.5 ml/min. When required, soluble protein complexes were concentrated further using centrifugal YM-50 Centricon (Millipore) units according to the manufacturers instructions.

Refolded RNAPs were assayed for transcriptional activity by measuring incorporation of a-³²P-UTP into RNA. Refolded RNAP was added to 1 x transcription buffer (1 x TB) containing 500 μM ATP, CTP, GTP, 1 μM UTP, 27 nM a-³²P-UTP (6000 Ci/mmol, Amersham), 1.5 μg nuclease-activated calf thymus DNA (Fluka), 120 mM potassium acetate, 10 mM magnesium acetate, 10 mM Tris-acetate pH 7.5 and 10 mM DTT, which was incubated at 37–65°C for 45 min. The final reaction volume was 50 μL. The reactions were stopped by addition of 15% (w/v) trichloroacetic acid followed by 30 min incubation on ice. The precipitate was collected on 96-well GF/F glass fiber filter plates (Whatman), washed twice with excess 10% TCA, once with 95% ethanol, and quantitated in a scintillation counter in presence of scintillant fluid (Opti-fluor, Packard Bioscience). These steps were fully automated on a Theonyx liquid handling platform (Aviso) (Nottebaum et al., 2008; Weinzierl, 2013). Independent repetitions (“biological replicates”) of the same transcription reaction are reproducible within a 12% error margin. Transcripts originating solely from abortive initiation are not precipitated using this method. Therefore, only transcripts from elongation-competent RNAPs (longer than ∼20 nucleotides) give rise to a signal in this assay.

Crystallization buffer sets ICL-1, -3, -4, and -5 (Hampton Research, Aliso Viejo, United States; Supplementary Figures S2A–F) were used as 10 x concentrates for high-throughput transcription assays based on nicked DNA templates as described previously (Werner and Weinzierl, 2002; Nottebaum et al., 2008; Tan et al., 2008; Boomsma et al., 2013). Briefly, the assay measures the incorporation of a-³²P-UTP into acid-insoluble RNA by liquid scintillation counting in a robotically implemented workflow. Similar to the strategy used when employing such buffer sets for crystallization screens, the initial screen was only carried out with one assay per buffer set. Buffer sets that gave high levels of activity were subsequently tested in triplicate to confirm the result.

The structural consequences of several of the substitutions were determined by comparing secondary structure propensities of sequences from N. equitans to equivalent domains from M. jannaschii (mj). Both species are hyperthermophiles thus containing similar sequence-encoded features that stabilize their protein structures at elevated temperatures. Markov Chain Monte Carlo (MCMC) simulations is the method of choice for a systematic and comprehensive exploration of conformational space (Boomsma et al., 2013) (Figure 3). As expected, the presence of a proline in the neBH causes a substantial disruption of a region that displays high a-helical propensity in the mjBH (Tan et al., 2008). On a structural level, the presence of proline in neA′ in position 810 (neA′ P⁸¹⁰) is predicted to cause a substantial destabilization of the a-helical conformation of the Bridge Helix in a slightly more N-terminal location (mostly affecting neA′ R⁸⁰⁸; Figure 3B). Simulating a ne Bridge Helix with a “corrected” in silico point mutation (neA′ P⁸¹⁰-A) restores the predicted conformational population to one that is very close to the mj Bridge Helix (Figure 3C). This proves that the unusual conformational properties are predominantly due to neA′ P⁸¹⁰ position, rather than any of the other differences in the primary amino acid sequence. The region in the ne Bridge Helix most distorted corresponds to the orthologous region in M. jannaschii (mjA′-R⁸²⁰) which is a structure with one of the highest a-helical propensities of the entire domain (Figure 3A). High-throughput mutagenesis studies of mjA′-R⁸²⁰ in mjRNAP have shown it to be highly sensitive to point mutations, with only phenylalanine and tryptophane substitutions not resulting in substantial loss of catalytic activity (Tan et al., 2008; Weinzierl, 2013).

Similar comparisons of the ne and mj Trigger Loop conformations yield a less clear-cut result (Figure 3D,E), although the unusual position of a proline in the ne motif near the edges of the domain (neA”-P⁴⁷) again is likely to contribute a destabilizing influence (Figure 3E).

For neFL3, the structural consequences of replacing highly conserved residues that are virtually invariant in other archaeal and eukaryotic polymerases in a non-conservative manner (for example, in FL3: C-D⁴⁴⁵, G-D⁴⁵², V/I-R⁴⁵⁸, and N-A⁴⁶⁰; Figure 1C) suggest that this would cause distinct changes in the functional contributions of these residues to catalysis. Especially one of these substitutions neB” G⁴⁵²-D is predicted to reduce the flexibility of the central region of neFL3 considerably in comparison to the orthologous sequence of mjFL3 (Figures 3F,G). Similarly, Metal B contains two highly conserved acidic residues that coordinate of binding of the Mg²⁺ ion brought along by the incoming NTP, but in N. equitans one of them is converted to glutamine (neA’ Q217) and thus is predicted bind the metal less strongly (Figure 1D).

Overall, based on previous insights from a range of structural and functional studies from archaeal and eukaryotic RNAPs representative of the majority of such organisms, a picture of a structurally diverged catalytic site in nanoarchaeal RNAPs emerges that suggests that the catalytic site may be more flexible in some areas (prolines in the neBridge Helix and neTrigger Loop domains, stiffer in the diverged neFL-3 domain and potentially compromised electrostatically by a diminished neMetal-B motif).

The conformational distortions caused by potentially disruptive radical substitution suggest that the Nanoarchaeum RNAP may display only very low - or even no - catalytic activity. On the other hand, the presence of all known RNAP subunits in an otherwise minimal genome implies selective pressure responsible for maintaining an active transcriptional machinery. Technical problems with obtaining N. equitans in quantities sufficient for biochemical analysis preclude a direct purification of native enzymes from cells. We therefore decided to investigate this question by adopting the in vitro assembly approach that has been applied successfully for the assembly of RNAPs from other hyperthermophilic archaea (Werner and Weinzierl, 2002; Naji et al., 2007). The in vitro assembly of neRNAP followed essentially the same procedure that we employed successfully in the past for mjRNAP (Werner and Weinzierl, 2002). Each of the subunits essential for catalytic activity was expressed as a recombinant protein in E. coli, followed by chromatographic purification and in vitro assembly by controlled dialysis from denaturing conditions (Figure 4A). Under these conditions, a portion of the neRNAP subunits assembled into a complex that - comparable to mjRNAP (Figure 4B) - eluted as a distinct peak of activity during size exclusion chromatography (Figure 4C). As expected from its hyperthermophilic origin, the temperature optimum for catalytic activity was around 76°C (Supplementary Figures S3). Initial transcription experiments with neRNAP suggested that the standard buffer conditions (120 mM potassium acetate, 10 mM magnesium acetate, 20 mM Tris-acetate, pH 8.6) were probably suboptimal because we observed a ∼ 7-fold lower specific activity for neRNAP as compared to mjRNAP when assembled in parallel. We therefore attempted to optimize the assay conditions over a wider range of pH values, salt concentrations and in the presence of various additives. The concept of “sparse-matrix” sampling is well established in the macromolecular crystallization community where the method is used to identify the optimal (yet initially unknown) conditions to obtain macromolecular crystals for structural studies (Jancarik and Kim, 1991). Such approaches have also been employed usefully to identify optimal renaturation conditions (Hofmann et al., 1995), or for stabilizing macromolecular complexes (Chari et al., 2015). Here, we employed such a strategy to identify the best assay conditions for neRNAP that included a wide range of different concentrations of various cations and anions, buffers at different pHs, and the presence of a variety of detergents and stabilizing reagents. A series of buffer sets (ICL-1, ICL-3, ICL-4, and ICL-5; Hampton Research), comprised of 386 different cocktails (see Supplementary Figures S2A–F for composition), were used as 10 x stock solutions after supplementing them with Mg²⁺ and Zn²⁺ in automated high-throughput transcription assays. Most of the mixtures include an inorganic or organic salt, a buffering compound (with pH ranges from 4.5 to 9.5) and a “precipitant”, such as polyethylene glycol. In our assays, the precipitant may display stabilizing effects on protein structure - especially quaternary structures - under hyperthermophilic assay conditions.

A summary of the results (see Supplementary Figure S4 for the complete data set) shows that neRNAP had a clear preference for a group of three buffers (ICL-3 #A1, A2, A3) that contained 20 mM sodium fluoride, potassium fluoride and ammonium fluoride, respectively (Figure 5A). This apparent preference for fluoride is unique to neRNAP because mjRNAP only performed moderately (30–50% in comparison to standard conditions) in these buffers (Figure 5A). To test this potential requirement for fluoride further, neRNAP activity was assayed in transcription buffers containing varying amounts of fluoride salts. Optimal neRNAP stimulation was achieved with 200–300 mM potassium fluoride or ammonium fluoride (Figure 5B and Supplementary Figure S5). The stimulation of neRNAP activity by fluoride ions raised the question of whether other halogen ions (chloride, bromide, or iodide) would have a similar effect on neRNAP. This, however, was not the case, suggesting that the stimulating effect on the catalytic activity of neRNAP is indeed highly specific for fluoride.