Melting temperature analysis

The melting temperatures (T_ms) of duplexes containing single SAMRS:standard and SAMRS:SAMRS pairs were obtained from the literature [35] (Supplementary Fig. S1). As a full set of thermodynamic parameters would require melting studies of oligonucleotides that are too short to serve efficiently as primers, we expanded on this literature using a heuristic approach.

Experimentally, the T_ms of standard oligonucleotide duplexes and the corresponding SAMRS:standard duplexes were measured in PCR buffer of JumpStart Taq DNA polymerase (1 µM of each oligonucleotide, 10 mM Tris-HCl, 50 mM KCl, pH 8.3 at 25°C, 1.5 mM, or 5.0 mM of MgCl₂). The oligonucleotides had the following sequences (upper case letters G, A, C, T, and N indicate standard nucleobases; lower case bold letters g, a, c, t, and n indicate SAMRS components; the standard complementary sequences are not shown):

Set 1 sequence: 5′-GAG CTG AGG TCA GTG T n n n n C-3′

Set 2 sequence: 5′-GAG CTG AGG TCA GTG N n a t n N-3′

Set 3 sequence: 5′-GCT CGA ATT GCA CCC T n n n n C-3′

The melting curves were visualized using fluorescent dye (0.5× EvaGreen) in a Roche LightCycler^® 480 with the following temperature profile: (i) denature and anneal duplexes: 95°C for 3 min, cool to 40°C with melting-curve setting (10 acq/°C; ∼4–5°C/min), heat again to 50°C and hold for 10 min; (ii) slowly denature duplexes from 50°C to 90°C with melting-curve setting (100 acq/°C; ∼1°C/min). Each set of duplexes was measured 3 times. Standard:standard and SAMRS:standard duplexes were run in parallel on the same 96-well plate. T_ms were obtained from the slow denaturing ramps (ii) using the automatic calculation method of the Roche LightCycler (MeltFactor set at 1.2, QuantFactor set at 20). ΔT_m values were calculated in Microsoft Excel for each ramp individually (Supplementary Tables S1-1, S1-2, and S1-3).

We then analyzed the sequences to find heuristically the best adjustments for each nearest neighbor pair that would most closely match the T_m deltas obtained experimentally from 84 sequences (Supplementary Tables S1-1 and S1-2). Code broke each sequence into nearest-neighbor doublets (e.g. TcataT is broken into the doublet: [Tc] [ca] [at] [ta], and [aT]). An initial T_m “effect estimate” was iteratively applied to all 48 possible doublets. The average difference between the input T_m deltas and those calculated from an initial set of estimates was recorded. The program then adjusted the T_m of randomly chosen doublets in 0.1 increments (adding 0.1 from the first and subtracting 0.1 from the other). If the adjustment improved the correspondence of the effect estimate and the data, the adjustment was retained. If it did not, the opposite adjustment was attempted (subtracting 0.1 from the first and adding 0.1 from the other). Adjustments that gave improvements were retained; the others were discarded. This process was continued until no improvements were found after 50 iterations through all doublets. Then, the process was repeated until no better estimates could be found in 20 000 attempts.