OPENPichia: licence-free Komagataella phaffii chassis strains and toolkit for protein expression – Nature Microbiology

Strains and media

The wild-type K. phaffii strains NRRL YB-4290, NRRL Y-7556 and NRRL Y-11430 were obtained from the Agricultural Research Service, CBS 2612 was obtained from the Westerdijk Institute (Netherlands) and NCYC 2543 was obtained from the National Collection of Yeast Cultures. All mentioned strains were cultured and maintained on YPD or YPD agar.

All entry and expression vectors were propagated and are available in the E. coli DH5α strain. MC1061 and MC1061λ strains were also successfully used and generally showed higher transformation efficiency as well as easier green–white or red–white screening than was the case for DH5α. All E. coli strains were cultured and maintained on Luria–Bertani (LB) agar.

The following antibiotics were used at a concentration of 50 µg ml−1 for the selection in E. coli: Zeocin, nourseothricin, hygromycin, kanamycin, chloramphenicol and carbenicillin. The following antibiotics were used at a concentration of 100 µg ml−1 for the selection in K. phaffii: Zeocin, nourseothricin, hygromycin, geneticin and blasticidin.

Several media were used: LB (1% tryptone, 0.5% yeast extract and 0.5% NaCl), yeast extract peptone dextrose (YPD; 1% yeast extract, 2% peptone and 2% d-glucose), yeast extract peptone glycerol (YPG; 1% yeast extract, 2% peptone and 1% glycerol), BMY (1% yeast extract, 2% peptone, 1.34% yeast nitrogen base without amino acids and 100 mM potassium phosphate buffer pH 6), buffered minimal glycerol yeast extract medium (BMGY; BMY with 1% glycerol), BMDY (BMY with 2% d-glucose), buffered methanol-complex medium (BMMY; BMY with 1% methanol) and limiting glucose (1% yeast extract, 2% peptone, 100 mM phosphate buffer pH 6, 50 g l−1 Enpresso EnPump substrate and 5 ml l−1 Enpresso EnPump enzyme solution). For plates, 1.5% agar was added to the LB media and 2% to the YPD media; when Zeocin selection was used, the media were set to pH 7.5.

All oligonucleotides and synthetic DNA fragments were ordered from Integrated DNA Technologies. All synthetic DNA fragments (gBlocks and Genes) were designed and adapted for synthesis using the Codon Optimization Tool and gBlocks Gene Fragments Entry Tool available at the website of Integrated DNA Technologies Europe.

Illumina sequencing

The strains were cultured overnight in YPD medium and the genomic DNA (gDNA) was extracted using an Epicentre MasterPure Yeast DNA Purification Kit. Sample preparation (DNA fragmentation, adaptor ligation, size selection and amplification) and next-generation sequencing (5 × 106 150-bp paired-end reads) was done by Eurofins using Illumina technology. The raw sequence reads were uploaded to the NCBI database under the accession number PRJNA909165. The reads were checked for quality using fastqc38, from which the %GC and number of reads were obtained. From the number of reads, the average overall coverage was calculated using the formula (frac{mathrm{{reads}}times {mathrm{read}},{mathrm{length}}left({mathrm{bp}}right)}{{mathrm{length}}; {mathrm{of}}; {mathrm{genomic}}; {mathrm{DNA}}+{mathrm{mitochondrial}}; {mathrm{DNA}}left({mathrm{bp}}right)}).

Next-generation sequencing analysis

The reads were trimmed using Trimmomatic39 to remove adaptors, leading and trailing low-quality bases (cut off quality of three), low-quality reads (four-base sliding window quality of <15) and reads below 100 bp. Next, the reads were aligned to a reference and the mutations were identified using Breseq40 in consensus mode. The genome sequence published by Sturmberger et al.12 was used as a reference. The reference sequences for killer-like plasmids and the mitochondrial DNA were obtained from Sturmberger et al.12 and Brady et al.16, respectively. The reported coverage depth was calculated using the Breseq algorithm. This is done by fitting a negative binomial distribution to the read-coverage depth observed at unique reference positions. The mean of this binomial fit is used as the coverage depth. The copy number of killer-like plasmids was estimated by comparing their coverage depth with the average of the four chromosomes. The coverage depth for each molecule was calculated as the mean of a binomial fit for the coverage depth for each reference position.

Phylogenetic tree

To generate a phylogenetic tree, the sequencing data from this study were combined with the previously published raw reads14 and also aligned as described above. From the predicted mutations of both datasets, a whole genome alignment was constructed, from which a phylogenetic tree was calculated using the Mega X41 software package. A maximum likelihood algorithm was used with a Hasegawa–Kishino–Yano substitution matrix.

Creation of the NCYC 2543 Δhis4 strain

The NCYC 2543 Δhis4 strain was generated using the split-marker method that was described previously by Heiss and colleagues42. The homology arms of the HIS4 gene were selected from Näätsaari et al.43 and the reference genome of the CBS 7435 strain. First, a construct containing the two homology arms with a floxed nourseothricin acetyltransferase marker was created. Two overlapping fragments, which overlap for a length of 594 bp, containing one of the homologies and a part of the antibiotic marker were then generated by PCR using Taq polymerase (Promega). These fragments were purified through phenol–chloroform precipitation. Briefly, following the addition of an equal volume of phenol:chloroform:isoamyl alcohol (25:24:1), the solution was mixed, centrifuged (5 min at 12,000g) and the liquid phase was isolated by decanting. A one-tenth volume of 3 M sodium acetate pH 5.5 and two volumes of 100% ethanol was added the sample, which was then mixed and centrifuged (15 min at 12,000g). Finally, the pellet containing the amplified DNA was washed with 70% ethanol, air-dried and resuspended in water.

Both purified fragments were transformed into NCYC 2543 competent cells by electroporation, and the transformants were streaked to single clone onto YPD plates containing nourseothricin and cultured at room temperature for 2 days. The resulting clones were replica plated onto CSM-his plates for growth screening and cultured for 2 days at room temperature. Strict non-growers were checked by colony PCR for replacement of the HIS4 gene with the antibiotic marker cassette.

The nourseothricin acetyltransferase marker was finally removed by transient expression of a Cre-recombinase. This gene was cloned into a plasmid with an autonomously replicating sequence44 and a Zeocin-resistance cassette, which was then transformed into the Δhis4 strain. The transformants were incubated overnight on a YPD plate containing Zeocin and the resulting colonies were transferred to YPD plates without antibiotics. The removal of the antibiotic cassettes of the plasmid and HIS4 knockout was verified with replica plating on YPD containing the respective antibiotics and double-checked via colony PCR.

Creation of the NCYC 2543 hoc1
tr strains

The NCYC 2543 hoc1tr strains were generated using the split-marker method described in the previous section. The left homology arm of the HOC1 gene was chosen such that it contained about 1 kb upstream of the premature stop codon. K. phaffii gDNA was used as the PCR template. The right homology arm was chosen so that it contained about 1 kb downstream of the premature stop codon. The left and right homology arms were respectively fused by PCR to the first and last two-thirds of the floxed nourseothricin acetyltransferase marker. The PCR fragments were gel purified and the DNA was recovered using a Wizard SV Gel and PCR Clean-Up System (Promega) according to the manufacturer’s instructions. Both purified fragments were transformed into NCYC 2543 competent cells by electroporation, and the transformants were streaked to single clone onto YPD plates containing nourseothricin and cultured at room temperature for 2 days. The resulting clones were screened through colony PCR using a forward primer that annealed upstream of the left homology arm and a reverse primer that annealed to the nourseothricin selection marker. The nourseothricin acetyltransferase marker was removed by transient expression of a Cre-recombinase as described in the previous section. The engineered HOC1 locus was confirmed for both strategies by colony PCR and Sanger sequencing. The sequences for the PCR primers and split-marker cassettes are in Supplementary Tables 6 and 7.

Growth analysis

The different K. phaffii strains were cultured on YPD agar for 2 days, inoculated in triplicate into a 5 ml preculture in test tubes containing BMDY and cultured overnight at 28 °C with shaking at 225 rpm The optical density at 600 nm (OD600) of each culture was measured and 250 ml BMDY was inoculated at a starting OD600 of 0.05. Samples of 1 ml were immediately isolated from each culture to measure and check the starting OD600. Next, the culture was cultivated in shake flasks at 28 °C with shaking at 225 rpm; samples of 1 ml were isolated every 2 h for 22 h and again after 26 and 29 h. All samples were diluted accordingly and measured within an OD600 range of 0.05–1.00.

Recombinant protein expression

The expression vectors were made using a MoClo toolkit, based on Golden Gate cloning as described in this paper (Supplementary Data 2). Briefly, the protein-coding sequences were ordered synthetically with Part 3b-type BsaI overhangs (NEB, R3733) and cloned into the entry vector with BsmBI (NEB, R0739). Next, expression vectors were made by assembly of the Level 0 parts.

The cloning procedure was as follows: 1 µl T4 DNA ligase (400 U; NEB, M0202), 2 µl T4 DNA ligase buffer (NEB, M0202) and 1 µl restriction enzyme (20 U) were added to 20 fmol backbone (pPTK081 for entry vectors or any P8 backbone for destination vectors). An excess of insert (>1,000 fmol PCR amplicon or synthetic gene, or 10 pmol annealed oligonucleotides) was added for a BsmBI assembly, whereas equimolar amounts (20 fmol) of each entry vector were added for a BsaI assembly. BsmBI assembly mixtures were incubated according to the following protocol: >25 cycles of 42 °C for 2 min (digest) and 16 °C for 5 min (ligation), followed by 60 °C for 10 min (final digest) and 80 °C for 10 min (heat inactivation step). BsaI assembly mixtures were incubated similarly, except that the digestion steps were performed at 37 °C.

K. phaffii electrocompetent cells were generated using the previously described lithium acetate method45. Briefly, precultures were inoculated in 5 ml YPD and cultured overnight in an incubator at 28 °C with rotation at 250 rpm. The precultures were diluted and cultured to an OD600 of approximately 1.5. Cells were harvested by centrifugation (1,519g for 5 min at 4 °C) from 50 ml of the culture, resuspended in 200 ml of a lithium acetate and dithiothreitol solution (100 mM lithium acetate, 10 mM dithiothreitol, 0.6 M sorbitol and 10 mM Tris–HCl pH 7.5) and incubated at 28 °C for 30 min with rotation at 100 rpm. The cells were then collected by centrifugation (1,519g for 5 min at 4 °C), washed twice with 1 M ice-cold sorbitol and finally resuspended in 1.875 ml of 1 M ice-cold sorbitol. DNA (0.5–1 µg) was added to aliquots of 80 µl and electroshocked (1.5 kV, 200 Ω and 25 µF). A 1 ml volume of 1 M sorbitol was immediately added to the samples and the suspension was incubated at 28 °C for 2–5 h. Next, the cells were plated on YPD agar containing the appropriate antibiotic and colonies were isolated after 2 days of incubation at 30 °C.

To enable the comparison of expression levels, only colonies with single-copy integration of the construct were selected. The copy number was determined by quantitative PCR on a LightCycler 480 system (Roche) using primers that bind PAOX1 and PGAP. The genes OCH1 and ALG9 were used as references. NCYC 2543 gDNA was included as a single-copy positive control. A single-copy plasmid integration will yield one additional copy and more than two copies would be the result of multiple plasmid integrations. Amplification efficiencies were determined using serial dilutions of gDNA samples. Reactions were set up in 10 μl with final concentrations of 300 nM forward primer, 300 nM reverse primer, 1×SensiFast SYBR no-ROX mastermix (Bioline), 10 ng gDNA and the following cycling conditions: 3 min at 95 °C, followed by 45 cycles of 95 °C for 3 s, 60 °C for 30 s at a ramp rate 2.5 °C s−1 and 72 °C for 1 s, and ending with 0.11 °C s−1 from 65 °C to 95 °C for melting curve determination (5 acquisitions s−1). Copy numbers were calculated using the ΔΔCt method46.

The different strains expressing the recombinant proteins were cultured on YPD agar plates for 2 days, inoculated in triplicate into a 5 ml preculture of BMDY and cultured at 28 °C overnight with shaking at 225 rpm. Next, the cultures for PAOX1-driven expression were inoculated in 2 ml BMDY, cultured for 24 h in a microtiter plate, transferred to 2 ml BMMY and incubated for 48 h in a microtiter plate. After 24 h in BMMY, an extra 1% methanol was added. The cultures for PGAP-driven expression were instead inoculated in 2 ml limiting glucose medium and incubated for 48 h in a microtiter plate. The OD600 was measured for all cultures and the supernatant was collected by centrifugation (2,500g for 5 min). The samples were incubated with EndoH (produced in-house) to remove N-glycans and analysed by SDS–PAGE.

ELISA-based quantification of GBP

Each well of a Nunc MaxiSorp 96-well plate was coated with 75 ng anti-penta-His (Qiagen, 34660) in PBS solution and incubated overnight at 4 °C. The wells were washed three times with 200 µl wash buffer (PBS + 0.05% Tween-20) and any residual liquid was removed. The samples were blocked with 100 µl Reagent Diluent (1% Probumin (Millipore, 82-045-1) in PBS pH 7.2) for 2 h. This was followed by three washes with 200 µl wash buffer and the removal of any residual liquid. Dilutions of the yeast supernatant were prepared in 96-deep-well plates, and 100 µl of a 100,000-fold dilution was applied to each well, followed by incubation for 1 h with gentle shaking in a table-top plate shaker. The wells were washed three times with 200 µl wash buffer and the residual liquid was removed. The samples were provided with 100 µl of 250 ng ml−1 MonoRab rabbit anti-camelid VHH coupled to horseradish peroxidase in Reagent Diluent and incubated for 1 h with gentle shaking in a table-top plate shaker. Each well was washed three times with 200 µl wash buffer and the residual liquid was removed. 3,3′,5,5′-Tetramethylbenzidine substrate was prepared according to the manufacturer’s instructions (BD OptEIA) and 100 µl was applied to each well, followed by a 10 min incubation. Finally, 50 µl stop solution (2 N H2SO4) was added to each well and the plate was read at 450 nm using a plate reader. The absorbance units were background corrected. All strains were compared in a Kruskal–Wallis omnibus test (two-sided), followed by a pairwise (two-sided) comparison corrected with Dunn’s multiple comparison procedure.

Flow cytometry to compare surface display of human lysozyme

Electroporation of a surface display plasmid to multiple K. phaffii strains (NRRL Y-11430, NCYC 2543 and OPENPichia) was performed using the lithium acetate method described in the ‘Recombinant protein expression’ section. We chose the previously reported47 pPSD-FLAG-hLYZ-V5-Sag1 plasmid as a test case. It expresses the wild-type human lysozyme protein flanked by an N-terminal FLAG tag and a C-terminal V5 tag, and is fused at the C-terminal end to a Sag1 anchor under the control of the AOX1 promoter. The copy number of the surface display construct in the resulting strains was determined as described earlier. Clones that were determined to have one integrated copy of the surface display construct were inoculated in BMGY supplemented with 50 µg ml−1 Zeocin in technical triplicates and cultured for 24 h at 28 °C with shaking at 200 rpm. The cultures were then transferred to BMMY supplemented with 50 µg ml−1 Zeocin, set to 10 OD600 units ml−1 and further cultured at 28 °C for 24 h with shaking at 200 rpm. After 12 h, the cultures were spiked with an additional 1% methanol. After induction, the cells were harvested by centrifugation at 1,500g for 5 min and washed three times with ice-cold washing buffer (PBS containing 1 mM EDTA pH 7.2 and one cOmplete Inhibitor EDTA-free tablet (Roche) per 50 ml buffer). The cells were kept on ice during the entire staining procedure. Unstained controls, single-stain controls and an empty vector control were included.

The cells (at an OD600 of two) were stained with mouse monoclonal anti-V5 (1/500; AbD Serotec, MCA2892) and rabbit polyclonal anti-FLAG (1/200; Sigma-Aldrich, F7425) in ice-cold staining buffer (wash buffer containing 0.5 mg ml−1 BSA) for 1 h at 4 °C. They were then washed three times with ice-cold staining buffer and stained with goat anti-mouse AF568 (1/250; Thermo Fischer Scientific, A-11031), goat anti-rabbit AF488 (1/500; Thermo Fischer Scientific, A11008) and Live/Death stain eFluor506 (1/1,000; Thermo Fischer Scientific) for 1 h at 4 °C. This was followed by three washes with ice-cold staining buffer before analysis on a BD FACSMelody instrument. The data were analysed using the FlowJo software. The gating strategy is shown in Extended Data Fig. 6.

Comparison of NCYC 2543 and NCYC 2543 hoc1
tr in a fed-batch process

Fermentations were conducted using a SciVario Twin 3 l fermenter (Eppendorf) containing 800 ml basal salts medium as described in the Pichia Fermentation Process Guidelines (Invitrogen Corporation, 2002). Yeast extract (Neogen, NCM0218A) was further added at a concentration of 10 g l−1 to supplement the batch medium.

To prepare the inoculum seed culture, a 1 l baffled flask containing 100 ml of BMGY, 1% yeast extract, 2% peptone, 1.34% yeast nitrogen base, 1% glycerol, 100 mM potassium phosphate pH 6.0), supplemented with 4 × 10–5% biotin, was inoculated with the expression clone of interest at an initial OD600 of 0.1. The culture in the flask was incubated at 28 °C with agitation at 200 rpm for 20–24 h until the OD600 reached the range of 20–30.

The batch phase of the fermenter was initiated by inoculating batch medium with inoculum seed at an initial OD of one. The cultivation temperature was maintained at 25 °C with an airflow rate of 1 vvm. The pH was automatically controlled at 6.0 by the addition of 25% wt/wt ammonium hydroxide as required. The dissolved oxygen levels were maintained at 30% saturation through control of agitation (600–1,200 rpm) and the addition of pure oxygen. Foam formation was prevented by the addition of an antifoam solution (Struktol, J673A).

Once the initial glycerol (40 g l−1) was fully consumed, marked by a rapid increase in the percentage of dissolved oxygen, the fed-batch phase commenced with the introduction of a 50% glucose solution (wt/wt) supplemented with 12 ml l−1 PTM1 solution. The feed rate was adjusted to 20 ml h−1 l−1 batch volume and linearly increased to 40 ml h−1 l−1 batch volume over a duration of 48 h to introduce 1 l of feed solution. All process parameters were maintained at the levels established during the batch phase throughout the entire fermentation process.

RT–qPCR analysis of HOC1 mRNA

The four strains were inoculated in BMGY medium, in triplicate, from an overnight preculture and cultured for 20 h at 28 °C and 200 rpm. The cells (10 OD600 units) were pelleted and washed with RNase-free water. Total RNA was prepared using a RiboPure-Yeast Kit (Invitrogen, AM1926), followed by a DNase treatment using a TURBO DNA-free Kit (Invitrogen, AM1907) according to the manufacturer’s instructions. Complementary DNA was then prepared using an iScript cDNA Synthesis Kit (BioRad, 1708891). The RT–qPCR reaction was performed for technical triplicates of each biological replicate using the following conditions: activation for 5 min at 95 °C, followed by 40 cycles of 10 s at 95 °C, 15 s at 55 °C and 20 s at 72 °C, and a final elongation step for 40 s at 72 °C. The transcript level variance of eight reference genes for normalization (UCB6, TDH3, QCR9, ALG9, PGK1, TAF10, ACT1 and TPI1) was analysed using the geNorm algorithm, as implemented in the qbase+ software48, to identify the genes whose transcript levels were least affected under the experimental conditions used. Based on these data (not shown), the HOC1 transcript levels were normalized using the geometric mean of the genes QCR9 and ALG9. The levels of HOC1 transcript were determined using two primer pairs. Determination of amplification efficiencies and conversion of raw Cq values to calibrated normalized relative quantity was performed using the qbase+ software. Statistical analysis of the calibrated normalized relative quantities was done using the GraphPad Prism 9 software package. All primers used are listed in Supplementary Table 6.

Transformation efficiency testing

Competent cells were prepared using the lithium acetate method described in the ‘Recombinant protein expression’ section. Each strain was transformed with 200 ng linearized plasmid and several dilutions of the transformation mix were plated on either non-selective YPD agar or YPD agar containing 100 µg ml−1 Zeocin. For each transformation, colonies were counted from the plates where clear individual colonies could be observed after incubation at 30 °C for 2 days. Both the selective and non-selective plates were counted to correct for a potential difference in the number of competent cells per transformation.

A linear model (estimated using ordinary least squares) was fitted in the statistical software R49. The log-transformed normalized transformation efficiency (natural logarithm of the number of transformants per million clones) was used as the outcome variable, and the strain and promotor type, including an interaction effect were used as the predictor variables. The model explains a statistically significant and substantial proportion of variance (coefficient of multiple correlation (R2) = 0.94, F(7,38) = 81.33, P < 0.001 and adjusted R2 = 0.93). Model-predicted group means with 95% confidence intervals were obtained using the ggeffects package with heteroscedasticity-consistent variance estimators from the sandwich package (vcovHC, type HC0)50,51.

Capillary gel electrophoresis-laser induced fluorescence detection-based glycan analysis of cell-wall mannoproteins

Strains were inoculated in YPD or YPG medium, from their respective precultures, at an OD600 of 0.05 and cultured overnight at 28 °C and 200 rpm. The next day, 500 OD600 units per strain were pelleted (10 min at 1,500g) and the mannoproteins were isolated as follows. The pellets were washed three times with Milli-Q water, after which 20 mM citrate buffer pH 6.6 was added at 1 ml per 150 µg of wet cell weight. The resuspended cells were autoclaved for 1.5 h at 120 °C in cryovials and then centrifuged for 10 min at 16,000g. Three volumes of ice-cold methanol were added to the supernatant fractions and the vials were incubated for 15 min at 20 °C. The mannoproteins were spun down at 16,000g for 10 min and the pellets were left to dry until transparent. The pellets were resuspended in 50 µl RCM buffer (8 M urea, 360 mM Tris–HCl pH 8.6 and 3.2 mM EDTA) and stored at 4 °C until further analysis.

N-linked oligosaccharides were prepared from the purified mannoproteins following blotting to polyvinylidene fluoride membrane in the wells of 96-well plate membrane plates and analysed by capillary electrophoresis with laser-induced fluorescence detection using an ABI 3130 capillary DNA sequencer as described previously22.

Alcian blue assay

The assay was performed as described previously23, with the following adaptations. Briefly, Alcian blue was prepared in 0.02 N HCl at a concentration of 63 µg ml−1 and the solution was centrifuged to remove insoluble precipitates. An overnight culture of each strain was cultured in YPD medium at 28 °C and 200 rpm. The next day, the cells were pelleted and the supernatant was removed. The cells were washed with 0.02 N HCl and the pellet was resuspended in 0.02 N HCl to 10 OD600 units ml−1. The cells (100 µl; 1 OD600) were transferred to a 96-well V-bottomed plate, to which 100 µl of the Alcian blue solution was added. Following incubation at room temperature for 15 min, the plate was centrifuged at 3,220g for 15 min, after which the pellets were visually checked.

Congo red and Calcofluor white test

The test was performed as described elsewhere52, with slight adaptations. Briefly, the strains were cultured overnight in BMGY. The next day, dilutions were made to obtain between 1 × 105 and 10 cells in 5 µl BMGY. Drops of 5 µl were spotted on the different plates, which were incubated at 30 °C for 3 days. Congo red (Sigma, C6767) and Calcofluor white (Fluka, 18909) were present at final concentrations of 75 µg ml−1 and 10 µg ml−1, respectively.

Electron microscopy

Transmission electron microscopy

The strains were cultured overnight in BMGY at 28 °C and 200 rpm. High-pressure freezing, as described previously53, was carried out in a high-pressure freezer (Leica EM ICE). The cells were pelleted and frozen as a paste in 150 µm copper carriers. High-pressure freezing was followed by quick freeze substitution as described previously54. Briefly, the carriers were placed on top of the frozen FS solution inside a cryovial containing 1% double-distilled water, 1% OsO4 and 0.5% glutaraldehyde in dried acetone. After reaching 4 °C for 30 min, the samples were infiltrated stepwise over 3 days at 0–4 °C in Spurr’s resin and embedded in capsules. The polymerization was performed at 70 °C for 16 h. Ultrathin sections of a gold interference colour were cut using an ultramicrotome (Leica EM UC6), followed by post staining, in a Leica EM AC20 system, with uranyl acetate at 20 °C for 40 min and lead at 20 °C for 10 min.

The sections were collected on formvar-coated copper slot grids. The grids were viewed using a JEM-1400Plus transmission electron microscope (JEOL) operating at 60 kV.

Scanning electron microscopy

The strains were cultured overnight in BMGY at 28 °C and 200 rpm. The cells were fixed overnight in 1.5% paraformaldehyde and 3% glutaraldehyde in 0.05 M sodium cacodylate buffer pH 7.4. The fixed cells were centrifuged for 2 min at 1,000g between each of the following steps. First, the cells were washed three times with 0.1 M sodium cacodylate buffer pH 7.4 and then incubated for 30 min in 0.1 M sodium cacodylate pH 7.4 containing 2% OsO4. The osmicated samples were washed three times with Milli-Q water before a stepwise ethanol dehydration (50%, 70%, 90% and 2 × 100%). This was followed by two incubations in hexamethyldisilazane solution (Sigma-Aldrich), as a final dehydration step, after which the samples were spotted on silicon grids (Ted Pella) and air-dried overnight at room temperature. Finally, the samples were coated with 5 nm platinum in a Q150T ES sputter coater (Quorum Technologies) and placed in a Gemini 2 Cross beam 540 microscope (Zeiss) for scanning electron microscopy imaging at 1.50 kV using an SE2 detector.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.