Adapter flanked cDNA molecules are attached to streptavidin beads via biotin modification of the Sol-S adapter. Yellow triangles indicate the GsuI recognition motif engineered into the Sol-L adapter, while the connected black arrow represents the distal cut site. Adapter barcodes and corresponding reads are classified as AA (green), CC (red), or TT (blue). Reads from the initial library and all three long march steps are aligned to form an 84 bp contig.

(A) Barcodes for each round of the long march. The first two bases, masked during genomic alignment, were analyzed for all reads aligning to the P. falciparum genome. Barcodes are classified as AA (green), CC (red), TT (blue) and NN (gray), where NN represents any barcode other than AA, CC, or TT. For each round of marching, the dominant barcode was that of the adapter added during that round. (B) Histogram of offset, overlapping alignments between 400,000 reads from the round 3 sub-library and 400,000 reads from the initial library. Reads were aligned to the P. falciparum genome and the difference between the starting positions of their 5′ termini was measured in cases where a round 3 read mapped distal to an initial library read. The resulting three peaks represent reads successfully marched once, twice, or three times. The gray line demonstrates that similar analysis of two pools of 400,000 reads from the initial library show no offset peaks. (C) Example of contig joining by adjacent marched reads from the same amplicon. A segment of P. falciparum chromosome 14 from 2,450,540 to 2,450,690 (representing a portion of the “hypothetical protein” gene PF14_0572) demonstrates the long march's utility in increasing contig size. Reads from all four libraries mapping to the area are shown. The four bottom reads derive from the libraries marched zero, one, two, and three times, respectively. While the gray reads cover much of the region shown, the adjacent marched steps from the last gray amplicon, shown in black, are required to cover the entire area and stitch together neighboring contigs.

Identical numbers of genome-aligned reads were randomly sampled from the round 3 sub-libraries and the initial libraries to simulate varying degrees of sequencing depth. The number of genomic base pairs covered by at least one read (y axis) was computed and plotted against the number of randomly selected input reads (x axis) for A) Plasmodium falciparum and B) hepatitis B virus (HBV) samples. Because of the small dataset sizes for HBV, each dataset of a given size was randomly filled and analyzed 1000 times; graphed coverage is an average for those datasets.

(A) Effect of overlap length on the probability of erroneous assembly of non-overlapping reads. For datasets with the indicated numbers of unique sequences, the probability was calculated of each sequence being erroneously joined to another in the dataset (left) or of at least one read in the dataset being erroneously joined to another (right). (B) Effect of initial pool complexity on the length of contigs. For each indicated number of amplicons in the initial pool, a simulation was performed assuming 1 million reads, and contigs were built by joining adjacent reads (see Methods). Each distribution of contig lengths, expressed in number of unique sequences assembled into the contig, was derived from a single simulation. (C) Effect of initial pool complexity on dataset redundancy. Simulations were performed as in (B) for each of the indicated amplicon pool complexities, and the fraction of unique sequences that were observed more than once is indicated. (D) Effect of allowed mismatches on the probability of erroneous assembly of non-overlapping reads. Probabilities were calculated assuming datasets of 1 million unique sequences. Allowed mismatches were single-nucleotide substitutions in the context of an ungapped alignment. (E) Effect of cleavage/ligation efficiency on the distribution of reads across the four steps of a three-round march. “Step 0” refers to unreacted molecules after three rounds of marching, while “Step 1”, “Step 2”, and “Step 3” refer to molecules that have been cleaved/ligated in one, two, or all three of three march rounds, respectively. (F) Effect of initial pool complexity on the length of contigs given a non-uniform distribution of reads across four steps. Contig lengths were determined through simulation as in (B), but using the probability of obtaining a read from each step as determined in panel (E) assuming a cleavage/ligation efficiency of 0.5. (G) Expected correspondence between round-associated barcode tags and the step no. of tagged reads. For instance, round no. = step no. = 1 if a molecule was cleaved/ligated in the first round and only the first round and was therefore tagged with the first round barcode and was advanced by one step along the amplicon template.