Small cell lung cancer (SCLC) remains a challenging cancer type to treat effectively. While patients initially respond to chemotherapy and immunotherapies, the disease almost always returns, often becoming resistant to treatment. Unlike other cancers, SCLC lacks clear subtyping approach based on specific genetic mutations that might indicate certain treatment options. However, recent data suggests that analyzing the activity of certain proteins called transcription factors (TF) may help distinguish SCLC subtypes. Unfortunately, methods to study these TF profiles are limited because SCLC tumors are typically diagnosed from very small samples of tumor tissue that are difficult to analyze in detail. So, in an effort to design more widely available approaches to study SCLC, the MacPherson and Ha labs collaborated to characterize cell free DNA (cfDNA) in their recent publication in Science Advances.
cfDNA is fragmented DNA released from dying cells, and can be found in circulation of everyone, but often increases during periods of disease such as infections or cancer, when cell death is more frequent. SCLC generates particularly high levels of cfDNA, likely owing to the highly metastatic nature of this cancer type. Dr. MacPherson explained that cell-free, circulating tumor DNA (ctDNA) is now commonly used to phenotype various cancer types, but mostly for gene mutation profiling. Researchers in the MacPherson and Ha labs were interested in using this to study TF profiling in SCLC. Normal DNA is wrapped around nucleosomes, which protects cfDNA in plasma from being degraded. When genes are more active, key regions are nucleosome-free – which allows easier access of TFs to the DNA. Dr. MacPherson further explained that, meanwhile, “for a gene that's highly expressed, transcriptional machinery need to bind to TF binding sites and the transcriptional start site of the gene which tend to be nucleosome free, resulting in degradation.” Meaning, that to recognize these highly active genes, they can essentially look for ‘missing’ regions of sequences in the cfDNA. To take advantage of this, the team developed a novel cfDNA nucleosome profiling by sequencing regions surrounding thousands of TF-binding sites (TFBS) and transcription start sites (TSS).
The team focused on four key TF important in SCLC: ASCL1 NEUROD1, POU2F3 and REST. “We decided to build into our targeted assay, pulling down transcriptional start sites of almost all genes in the genome, as well as transcription factor binding sites for four key transcription factors in small cell lung cancer,” explained Dr. MacPherson. They started by utilizing patient-derived xenografts (PDX) models that they grow in mice, where they could assess the tumor cfDNA by computationally separating it from the mouse DNA. Additionally, they ran their targeted sequencing panel on a large cohort of SCLC and NSCLC plasma samples. To generate an all-in-one assay to phenotype SCLC for key mutations and TF activation, they also included exons for cancer mutated genes to see if mutation profiles could be characterized using their targeted panel. When they analyzed PDX and plasma samples where either immunohistochemistry (IHC) or RNA sequencing data was also available, they generated TF subtype groupings based on their corresponding gene expression levels.
Using their targeted sequencing panel, they saw that for samples known to be positive for a particular TF, composite sequencing coverage at the corresponding binding sites were decreased, resulting in a differently shaped curve. For NEUROD1, POU2F3 and REST, the magnitude of change correlated with their gene expression levels, indicating they could even distinguish high and low-expressors based on this method. Additionally, by using unsupervised clustering, the team was able to separate samples by their TF activity, highlighting the strength of this platform when compared to more traditional approaches. Next, the team wanted to assess whether their panel could also capture tumor cell gene expression using TSSs and found that changes in coverage indicated results that were consistent with conventional approaches. Together, these data assured that this method could reliably distinguish patients based on TF activity and gene expression. The research team then clustered the nucleosome profiles of the samples at individual TFBS and TSSs and found a smaller subset of sites which they termed as highly informative, referring to the ability to use this narrowed list to infer TF activity and gene expression. In many of their samples, from this unbiased approach they found that the highly informative TSSs included many which were associated with the key TFs they were interested in. Dr. MacPherson remarked on this result, where “taking unbiased approaches to identify informative regions,” allowed them to “see major players in small cell lung cancer and targets of our master TFs of interest, which was gratifying to see.” Dr. Ha added that “It was remarkable to us that only a smaller set of informative genomic regions was needed for our computational models. This has implications for more cost-effective and easier translation into the clinic to benefit patients.”
The team also developed a computational model to predict whether a sample was more similar to NSCLC or SCLC based on cfDNA profiles. This computational method estimates ‘histology composition’ from the cfDNA nucleosome profiles, which would indicate whether a sample was NSCLC-like (score of 0) or SCLC-like (score of 1). They trained their model on the PDX and some nonmalignant patient samples and found that this could strongly distinguish the two cancer types from the plasma samples. They even observed an intermediate score (0.74) in one PDX derive from a patient whose NSCLC had transformed to SCLC, indicating that capability to detect this type of event in patients, which is a common clinical complication, often driving resistance to treatment in NSCLC. Lastly, the team wanted to see whether their method could distinguish TF-subtypes, without the need for histology or RNAsequencing. They designed a prediction model which would estimate an activity level (0 to 1) for each individual TF, based on both TFBS and TSSs which are associated with the corresponding TF. Again, they trained their model on the PDX and nonmalignant samples, and then retrained on plasma samples where they had matching known tissue TF activity levels and found that their model did quite well in distinguishing appropriate subtypes for ASCL1, NEUROD1 and REST. They then applied their model to their larger set of NSCLC and SCLC plasma samples and found that it predicted TF-subtypes which would be both consistent with SCLC compared to NSCLC TF activity, as well as an expected predominance of ASCL1 and NEUROD1 subtypes in SCLC. They also observed that their model could pick up samples which appeared to be positive for more than one TF, which was reassuring as this is a common clinical observation. Together, their method seemed to be remarkably good at performing the analyses for which the lack of tissue samples would have previously limited.
Overall, the data shown here highlights a promising new approach to studying SCLC using cfDNA, which could overcome the limitations of tissue-based analyses. Dr. Ha explained that “for tumors that may not always have clear actionable genomic alterations, this assay expands the boundaries for using ctDNA.” Both Drs. MacPherson and Ha were enthusiastic that while the research is at an early stage, with improvements, this method could be relatively easily implemented in the clinic, allowing for noninvasive diagnoses, monitoring and potentially the recognition of subtype-based therapeutics. “This work has important implications for non-invasive routes to detect clinically relevant subtypes have different transcriptional programs,” Dr. Ha explained. Dr. MacPherson added that “I see a key role for this type of test right now as a tool for broad phenotyping of responses to either investigational agents or even to standard of care to link either a gene mutation or a transcriptional subtype to strong responses or resistance, which would help in the design of the next version of a clinical trial.” The research team was also keen to express their gratitude for the collaborative nature of the Hutch which facilitated the teamwork between his and Dr. Ha’s labs. This work was jointly led by Joseph Hiatt and Anna-Lisa Doebley in the MacPherson and Ha labs. “It's very easy in this environment at the Hutch to have really rich, deep collaborations, and I think this is an excellent example of that,” Dr. MacPherson reflected.
This spotlight work was funded by the National Institutes of Health, the Kuni Foundation Discovery Grants for Cancer Research, the Conquer Cancer Foundation and the CRUK.
Fred Hutch/University of Washington/Seattle Children’s Cancer Center Consortium members Drs. Rafael Santana-Davila, Keith Eaton, McGarry Houghton, Gavin Ha and David MacPherson contributed to this work.
Hiatt JB, Doebley AL, Arnold HU, Adil M, Sandborg H, Persse TW, Ko M, Wu F, Quintanal Villalonga A, Santana-Davila R, Eaton K, Dive C, Rudin CM, Thomas A, Houghton AM, Ha G, MacPherson D. Molecular phenotyping of small cell lung cancer using targeted cfDNA profiling of transcriptional regulatory regions. Sci Adv. 2024 Apr 12;10(15):eadk2082. doi: 10.1126/sciadv.adk2082. Epub 2024 Apr 10. PMID: 38598634; PMCID: PMC11006233.