High-performance xenograft read-sorting for DNA methylation analysis

In the world of cancer research, scientists often need to study tumors in a living environment that mimics the human body. To do this, they take human cancer cells and transplant them into mice, creating what are known as patient-derived xenografts. These tiny living laboratories allow researchers to watch how a tumor grows, how it reacts to drugs, and how it might become resistant to treatment. However, a significant hurdle arises when scientists try to analyze the genetic material from these mixed samples. The final product is a chaotic soup of DNA strands, some belonging to the human tumor and others belonging to the mouse host. To understand the tumor's behavior, researchers must first separate the human strands from the mouse ones with extreme precision. This task becomes even more difficult when studying DNA methylation, a chemical tag that acts like a switch, turning genes on or off. To read these tags, scientists use a chemical process that changes the DNA sequence itself, scrambling the letters in a way that confuses the standard computer programs used to sort genetic data.

A team of researchers has now developed a smarter way to untangle this genetic mess, specifically for these chemically altered DNA samples. They enhanced an existing software tool called Xengsort, transforming it into a highly efficient sorter that can distinguish between human and mouse DNA even after the chemical changes have occurred. The team tested their improved tool against several other established methods using both computer-generated mixtures and real-world data from breast cancer studies. They found that while all the tools were generally accurate at identifying the source of the DNA, the new version of Xengsort was dramatically faster and required far less computer memory. In one large-scale test involving real patient samples, the new tool finished the sorting job in just over five hours, whereas a competing method took nearly thirty hours to complete the same task.

The core of the problem lies in how DNA methylation is read. To see the chemical switches, scientists treat the DNA with a substance that converts one specific letter, cytosine, into another, thymine, if that letter is not chemically tagged. This process is essential for understanding how tumors adapt, but it fundamentally alters the DNA sequence. Imagine trying to sort a pile of mixed-up books where someone has changed the spelling of every word in half of the books; a standard librarian would struggle to tell which book came from which author. Similarly, standard computer programs that match DNA sequences to a reference genome often fail because the chemical conversion makes the human DNA look different from the human reference, and the mouse DNA look different from the mouse reference. This confusion can lead to human tumor strands being misidentified as mouse strands, or vice versa, which would ruin the scientific analysis.

To solve this, the researchers rebuilt the internal dictionary of the Xengsort tool. Instead of just looking for exact matches of DNA letters, the new tool anticipates the chemical changes. It creates a dual-layered index that accounts for both the original DNA sequence and the sequence after the chemical conversion. When a piece of DNA enters the system, the tool checks it against both possibilities simultaneously. It breaks the DNA strand into small chunks and asks a simple question: does this chunk look like it belongs to a human or a mouse, considering the chemical changes? By trusting the most unique chunks of DNA and ignoring the ones that look too similar in both species, the tool can confidently sort the strands into human, mouse, or ambiguous categories. This approach allows the software to skip the slow, heavy process of aligning every single strand to a full genome, which is what other methods require.

The researchers put their tool to the test in two ways. First, they created millions of simulated mixtures of human and mouse DNA strands with known origins, allowing them to measure exactly how often the tool made mistakes. They tested these mixtures at different depths and ratios, simulating everything from a sample with very little human DNA to one that was almost entirely human. The results showed that the tool was incredibly accurate, losing very few human strands and rarely mistaking mouse strands for human ones. It performed just as well as the other top tools in terms of accuracy, but it did so in a fraction of the time. While other methods took hours to process the data and required massive amounts of computer memory, Xengsort completed the same work in under an hour on a standard computer setup.

Next, the team applied the tool to real data from breast cancer patients whose tumors had been grown in mice. These samples came from different types of studies, including some where the DNA was chemically converted and others where it was processed using enzymes. In these real-world scenarios, the tool again proved its worth. It successfully sorted the DNA from complex samples, including those with very low amounts of human tumor cells mixed with a lot of mouse tissue. The researchers noted that the tool was particularly robust when dealing with samples that had been treated with drugs to alter the DNA methylation patterns. In a direct comparison of speed and resource usage, the new tool was significantly more efficient. It processed a large dataset in 5.5 hours using 43 gigabytes of memory, while a competing method took 29.1 hours and required 49 gigabytes of memory. Another method, which relied on a combined reference genome, was fast but demanded a massive 73 gigabytes of memory, which can be a limiting factor for many laboratories.

The study also highlighted that the choice of sorting method matters less for accuracy than it does for efficiency. All the tested tools, including the new Xengsort, achieved high accuracy in separating the DNA strands. However, the time and computing power required to get those results varied wildly. For researchers working with large datasets or those who need to process samples quickly to guide treatment decisions, the speed and low memory footprint of Xengsort offer a clear advantage. The researchers concluded that their workflow is now ready for widespread use, suggesting that scientists should first sort their DNA strands by species using this tool before moving on to the detailed analysis of the tumor's methylation patterns. By making the sorting step faster and less resource-intensive, the tool removes a major bottleneck, allowing scientists to focus more time on understanding the cancer itself rather than waiting for the computer to finish its work.