PRDM9 forms a trimer by interactions within the zinc finger array

(A) We designed an additional tandem DNA fragment (tandem-Hlx1-BamHI) with two binding sites (purple) separated by a restriction enzyme site (green). This new fragment was derived from the original tandem-Hlx1, in which the binding sites were re-arranged within the unspecific sequences (grey). (B) EMSA-binding reactions were performed using 1 μl of the PRDM9 construct Halo-ZnF^Cst 1–11 and 3 nM of the two different tandem fragments (tandem-Hlx1 or tandem-Hlx1-BamHI) both with a size of 232 bp (lane 1–2, tandem-Hlx1; lane 3–4, tandem-Hlx1-BamHI). The tandem-Hlx1-BamHI fragment was digested with BamHI enzyme before (red, lane 5–6) or after (blue, lane 7) adding PRDM9 to the reaction. The BamHI digest was performed in 1× TKZN buffer (10 mM Tris, 50 mM KCl, 50 µM ZnCl₂, and 0.05% NP-40, pH 7.5) using 0.5 U/μl BamHI-HF enzyme (NEB) incubated at 37°C for 15 min. All binding reactions were carried out in 1× TKZN buffer and 50 ng/μl polydIdC as unspecific competitor. After an incubation of 60 min at RT, the samples were electrophoresed for 60 min at 100 V in 0.5× TBE. The rest of the protocol was performed as described in Supplementary_Methods section of the Supplementary Information. For both tandem fragments, we observed a lower- and supershift band (lane 2 and 4, respectively). Note that for tandem-Hlx1-BamHI, we see an additional band, which could be due to a faster migration of a more compact DNA secondary structure. The restriction enzyme digest of the tandem-Hlx1-BamHI resulted in two separate fragments (75 and 157 bp in size) with only one Hlx1-binding site (lane 5). The incubation of digested DNA with PRDM9 resulted in two shifts, representing the binding of PRDM9 to the shorter- (75 bp) and longer DNA (157 bp) (lane 6). When digesting the DNA fragment after complex formation (lane 7), the banding pattern changed from the pattern of lane 4 to a banding pattern identical to lane 6, strongly suggesting an independent binding of two PRDM9 protein complexes on the tandem DNA fragment that can get separated with the digest.

(A) The PRDM9 constructs YFP-ZnF^Cst, ZnF^Cst, YFP-ZnF^Cst 1–11, ZnF^Cst 1–11, and YFP-ZnF^Cst 2–11 were bound to 10 DNA molecules (extended assay I) increasing in length from 75 bp–1,460 bp but harbouring one (red) or two (purple) successive 34-bp Hlx1 target sites (single- or tandem-Hlx1, respectively) as it is shown in Fig 2B. The PRDM9–DNA complexes were separated along a 5% native polyacrylamide gel for 120 min at 100 V resulting in lower- and supershift bands (red and purple rectangles, respectively). Each binding reaction included a lower and upper reference band (220 bp and 4,368 bp, respectively; blue arrows) to normalize for migration distance. For each experiment, only one type of PRDM9 construct was used as stated above each EMSA membrane. (B) Because of the low MW of the shortest constructs ZnF^Cst 2–8 and ZnF^Cst 2–6, the assay was limited to 5 instead of 10 different DNA fragments (small assay I), including a lower- and upper reference band of 75 and 2,585 bp, respectively. “Plus” and “minus” indicate the presence or absence of protein for the respective DNA molecule used. The bound and unbound DNA fragments were separated for 60 min at 100 V along the 5% native polyacrylamide gel. For each experiment, only one type of PRDM9 construct was used as stated above each EMSA membrane. (C) Extended assay I was repeated without using a protein sample. Single- and tandem-Hlx1 only fragments were detected showing some DNA impurities, which were considered when analysing the signals of complex formation. Note that all conditions used for the EMSA experiments can be found in Table S10. The raw data and estimated correlations are presented in Tables S2 and S4.

To test for data validity of four replicated multimer assay II experiments, the expected migration distance for a PRDM9 dimer (#2, green), trimer (#3, red), and tetramer (#4, blue) along the native gel was calculated and plotted against the observed data (Table S3, #PRDM9) in a linear correlation using the OriginPro software. The perfect linear fit (black) only correlates with the expectation of a trimeric complex formation.

Box plot of the tested PRDM9 constructs representing the distribution of measured PRDM9 units within a multimer complex of assay I. Different PRDM9 constructs are colour-coded: green, ZnF domain of PRDM9^Cst; blue, tandem ZnF array of PRDM9^Cst without ZnF0; and red, truncated ZnF array of PRDM9^Cst. The raw data and results can be found in Tables S2 and S4.

FCS autocorrelation curves of Halo-eYFP-ZnF^Cst 1–11 before (black line) and after (red line) the addition of 3 M urea and of eYFP before (dark grey) and after (dark brown) the addition of 3 M urea. The number of fluorescent molecules 〈N〉 within a focal volume was estimated as the amplitude value (equivalent to 1/G(τ) at τ = 0) from these FCS correlation curves. The ratio (N_denatured/N_native) between the number of PRDM9 oligomers per focal volume and the number of PRDM9 subunits after oligomer disruption was used to calculate the number of PRDM9 monomers per oligomer. The complete FCS data are presented in Table S5.

(A) eYFP-labelled and unlabelled ZnF^Cst 2–11 (62 and 37 kD, respectively) constructs were co-expressed using competent Escherichia coli cells and visualized on a Western blot with an anti-His antibody. Black arrows indicate the co-expressed proteins in the WC lysate. (B) In an EMSA experiment, crude lysates of the co-expressed (1 μl of WC; WC fraction, including cell debris of bacterial expression), as well as, single-expressed (3 μl of supernatant [SN] each; SN of bacterial expression) constructs were bound to 10 nM of single-Hlx1 75-bp DNA in a reaction containing 1× TKZN buffer, which was supplemented by 50 ng/μl unspecific competitor polydIdC. Binding of the co-expressed PRDM9 constructs (lane 2) resulted in two shifted bands with the same migration distance as observed for each independent PRDM9 construct (lane 3, ZnF^Cst 2–11; lane 4, eYFP-ZnF^Cst 2–11).

(A) To analyse the PRDM9–DNA complex with mass spectrometry, 7 μl of Halo-ZnF^Cst 1–11 was incubated with a 2 µM 75-bp DNA of the Hlx1 hotspot for 60 min at room temperature, separated on a native 5% polyacrylamide gel, and visualized by Coomassie blue staining (arrow). As a control, the protein without DNA was loaded on the gel not giving a distinct band under native conditions. “Plus” and “minus” indicate the presence or absence of PRDM9 and DNA within the binding reaction. (B) Shown is the mass spectrum of the PRDM9–DNA complex (Halo-ZnF^Cst 1–11 + single-Hlx1 75 bp), which was cut out and isolated of the native gel shown in (A), treated with trypsin digestion, DTT reduction and carbamethylation using iodacetamide, and finally analysed using a MALDI-TOF Axima Performance instrument (Shimadzu). The spectrum shows the PMF of the PRDM9–DNA complex with all prominent peaks matching the peptides of our protein. (C) To confirm the PMF data, four colour-coded MS/MS spectra of prominent m/z values were obtained: 1,338.61 (blue), 1,767.84 (purple), 1,810.76 (green), and 1,908.01 (orange). Note that the raw data and m/z values can be found in Tables S6 and S7.

To further confirm that the PRDM9 trimer binds only one target DNA molecule, we tested four different combinations of short- and long DNA sequences. Specifically, we combined two of five possible DNA fragments, all carrying one Hlx1-binding target flanked by unspecific DNA of different lengths with a size of 75 bp (15 nM), 189 bp (15 nM), 273 bp (7 nM), 543 bp (1.5 nM), and 856 bp (1.5 nM). Note that for this experiment, we used a different construct (2.5 μl of Halo-ZnF^Cst 1–11) than in Fig 4. Both PRDM9^Cst constructs have the same number of ZnF. Coloured rectangles highlight the different combinations: green, 75 bp + 189 bp; pink, 75 bp + 273 bp; yellow, 189 bp + 856 bp; and blue, 273 bp + 543 bp. “Plus” and “minus” above the EMSA image indicate the different components in the reaction (plus = present or minus = absent). For clarity, we highlighted unbound and bound DNA fragments in red and blue, respectively. The banding pattern of all four DNA combinations results in only two shifts (one for each fragment), supporting that PRDM9 binds only one DNA fragment at a time, as proposed in model 1 of Fig 4.

In this EMSA titration experiment, the DNA is titrated over the protein by using two different types of DNA, hot long (75 bp) and cold short DNA (39 bp). The “hot” DNA is biotinylated and is detected in EMSAs, whereas the cold DNA is not biotinylated and cannot be visualized on the EMSA membrane. Binding reactions were performed by using a constant amount of hot long DNA (15 nM single-Hlx1 75 bp; lane 1 = hot DNA only) and PRDM9 protein (250 nM of MBP-eYFP-ZnF^Cst WC, lane 2 = hot DNA + PRDM9) and increasing amounts of cold short DNA (single-Hlx1 39 bp; lane 3, 15 nM; lane 4, 75 nM; lane 5, 150 nM; lane 6, 300 nM; lane 7, 600 nM; lane 8, 900 nM; lane 9, 1,200 nM; and lane 10, 1,500 nM), supplemented by 1× binding buffer (10 mM Tris, 50 mM KCl, and 1 mM DTT, pH 7.5) and 50 µM ZnCl₂, 0.05% NP-40, and 50 ng/μl nonspecific competitor polydIdC. The reactions were incubated at RT for 60 min and the protocol was continued as described by Striedner et al (2017). Note that PRDM9 binds to Hlx1 DNA with 75 bp and 39 bp with the same affinity, as was shown by Striedner et al (2017). The black arrow indicates unbound “hot” DNA and the orange arrow indicates the shifted band of the PRDM9+“hot” DNA complex. The intensity of the shifted band decreases with an increase in “cold” DNA, as it competes with the “hot” DNA for the binding of PRDM9. Although the total amount of DNA increases compared with the protein concentration, the migration of the PRDM9–DNA complex does not change, suggesting that only one DNA molecule is bound by the protein trimer even at higher DNA concentrations. Complexes with more than one DNA molecule would render different shifts (e.g., 2× long, 2× short, and 1× long + 1× short). Note that similar data have been published by our group earlier; however, for a different purpose (Striedner et al, 2017; Tiemann-Boege et al, 2017).