Cross-validated test; DELFI score distribution; Annual transitions.

Whole genome cfDNA analyses. Clinical and demographic characteristics of individuals analyzed.

Performance of blood-based lung cancer screening test. A, Sensitivity and specificity of the test in the clinical validation set (N = 382) overall and by clinical subgroup. Point estimates are reported with 95% Wilson confidence intervals. Overall sensitivity and specificity denoted by solid vertical lines. B, Sensitivity of the test in the lung cancer cases in the clinical validation set (N = 248) evaluated across cancer histology, and T, N, and M categories. Point estimates are reported with 95% Wilson confidence intervals. Overall sensitivity of 84% denoted by the solid horizontal line. C, Left, sensitivity of the test in the lung cancer cases in the clinical validation set (N = 246) by cancer group stage. Middle, bar plot showing the stage distribution of lung cancer as observed in populations undergoing lung cancer screening with LDCT (based on NLST study) that are used to weigh observed stage-specific sensitivities. Right, lung cancer screening relevant stage-weighted sensitivity in clinical validation set. D, Comparison of the NNS with LDCT conditioned on test positive or negative result when applied in the lung cancer screening eligible population. Test performance showed consistency across clinical subgroups and expected increased performance with increasing burden of disease (tumor (T), node (N), metastasis (M) and group staging). After weighting, the stage distribution to reflect a screening population, test performance remained high and demonstrated the ability to reliably identify those individuals more likely to have lung cancer detected on LDCT.

Population health benefits of a blood-based test in lung cancer screening. A, Care pathway reflecting the recommended standard of care for lung cancer screening with LDCT that is received by 6%–10% of eligible individuals annually, as well as potential pathway employing initial blood-based test and follow-on events. B, The predicted number of cancers detected by screening scenario: LDCT alone (“base case”); LDCT + low test uptake; LDCT + high test uptake. C, Predicted cancers diagnosed at stage I versus Stage IV by screening scenario: LDCT alone (“base case”); LDCT + low test uptake; LDCT + high test uptake. D, Predicted decrease in lung cancer deaths represented by screening scenario: LDCT alone (“base case”); LDCT + low test uptake; LDCT + high test uptake. E, Simulated comparison of the predicted number needed to scan with LDCT to detect one lung cancer: LDCT alone (“base case”); LDCT + low test uptake; LDCT + high test uptake. Population-level modeling demonstrates significant health benefits when a blood-based test is available as an alternative for lung cancer screening.

High-dimensional fragmentation features reflect lung cancer biology and are incorporated in the machine learning classifier. A, Heatmap representation of the deviation of cfDNA fragmentation features across the genome for the classifier training set with lung cancer or noncancer individuals compared with the mean of classifier training noncancer individuals. Each row represents a sample, whereas columns show individual genomic features. The cross-validated DELFI score and clinical characteristics are indicated to the left of the fragmentation deviation heatmap. B, Left, TCGA-derived observations of chromosomal arm gains (red) and losses (blue) in lung adenocarcinoma (LUAD; n = 518) and squamous cell cancer tissues (LUSC; n = 501). Right, the observed chromosome arm gains (red) and losses (blue) in the classifier training individuals separated by histology. C, A heatmap representation of the principal component eigenvectors of the fragmentation profile features. Regression coefficients from the final classifier indicating how the principal components of the fragmentation profiles and z-scores of the chromosomal arms were combined are provided in the top and right margins of the heatmap, respectively. Positive values for the coefficients are represented in red, whereas negative values are represented in blue. Agreement across copy number chromosomal gains and losses in TCGA lung cancers, observed z-scores in the cfDNA of patients with lung cancer, and chromosome arm model coefficients reflect biologic consistency between chromosomal changes in lung cancer, cfDNA fragmentation profiles, and classifier features.

Participant demographics and clinical characteristics.

Genome-wide fragmentation profiles are altered in patients with cancer and reflect underlying chromatin structure. A, The fragmentation profile (ratio of short to long cfDNA fragments in 5 Mb bins) across the genome was evaluated in the classifier training plasma samples of lung cancer (n = 181) and noncancer individuals (n = 395). The noncancer individuals had similar fragmentation profiles, whereas patients with lung cancer exhibited significant variation. B, Comparison of cfDNA profiles with Hi-C A/B chromatin compartment reference data from lung cancer tissue or peripheral blood cells. Track 1 shows A/B compartments extracted from LUSC cancer tissue (48). Track 2 shows a median lung cancer component extracted from the LUSC plasma samples of 7 patients with lung cancer from the classifier training set with high tumor fraction by ichorCNA (49). The 7 LUSC cases with high ichorCNA have values of 0.051, 0.439, 0.230, 0.259 0.439, 0.167, and 0.057. Track 3 shows the median profile for 10 noncancer plasma samples from the training set. Track 4 shows A/B compartments for lymphoblast cells (48). These four tracks show chromosome 22 as an example, with darker shading indicating informative regions of the genome where the two reference tracks differ in domain (open/closed). C, 100-kb regions were selected using the reference LUSC and lymphoblast A/B tracks as having the same chromatin state or opposite chromatin state. Within these regions, the deviation of the fragmentation value from a noncancer cfDNA reference (n = 10) was plotted per region per individual (noncancer n = 20, LUSC n = 7). Values around 0 have little variation from the noncancer reference. Negative values indicate a region has a more open chromatin state than the reference and positive values indicate a region has a more closed chromatin state than the reference. These data suggest that although cfDNA profiles of healthy individuals reflect the chromatin structure of blood cells, those of patients with lung cancer represent a mixture of cfDNA patterns of chromatin compartments from lung cancer as well as blood cells.

Lung cancer screening via annual low-dose computed tomography has poor adoption. We conducted a prospective case-control study among 958 individuals eligible for lung cancer screening to develop a blood-based lung cancer detection test that when positive is followed by a low-dose computed tomography. Changes in genome-wide cell-free DNA fragmentation profiles (fragmentomes) in peripheral blood reflected genomic and chromatin characteristics of lung cancer. We applied machine learning to fragmentome features to identify individuals who were more or less likely to have lung cancer. We trained the classifier using 576 cases and controls from study samples and validated it in a held-out group of 382 cases and controls. The validation demonstrated high sensitivity for lung cancer and consistency across demographic groups and comorbid conditions. Applying test performance to the screening eligible population in a 5-year model with modest utilization assumptions suggested the potential to prevent thousands of lung cancer deaths.Significance: Lung cancer screening has poor adoption. Our study describes the development and validation of a novel blood-based lung cancer screening test utilizing a highly affordable, low-coverage genome-wide sequencing platform to analyze cell-free DNA fragmentation patterns. The test could improve lung cancer screening rates leading to substantial public health benefits.See related commentary by Haber and Skates, p. 2025

Background: Less than 10% of eligible persons undergo annual lung cancer screening by low-dose computed tomography (LDCT). Greater uptake of LDCT is hampered in part by its cost, inaccessibility, and balance of benefit to risk. A blood-based, low-cost, widely available initial blood test could boost screening participation and improve the net benefit of screening, if it were sensitive for cancer detection and affordable. The DELFI (DNA evaluation of fragments for early interception) technology uses low-coverage, whole-genome sequencing and machine learning to identify patterns of circulating cell-free DNA (cfDNA) fragmentation indicative of cancer. We report initial results of the cfDNA analysis from DELFI-L101 (NCT04825834), a prospective, observational, national case-control study to train and test DELFI classifiers for lung cancer detection. Methods: Eligible participants were adults ≥50 years old with current or previous smoking histories of ≥20 pack-years and recent or planned thoracic CT imaging. At enrollment, medical history was recorded and blood samples were collected for DELFI analysis. A classifier for lung cancer detection was developed using repeated 10-fold cross-validation. A split study approach for the purposes of independent validation of the classifier is forthcoming. Results: The study cohort included 242 patients with lung cancer and 652 individuals without cancer. Study participants largely represented those of a lung cancer screening population, with 45% stage I/IA. Most participants were ≥65 years old with roughly equal proportions of men and women. There was broad representation across lung cancer risk factors among both cases and controls. The cross-validated area under the receiver operator characteristic curve (AUC) was 0.81 for lung cancer detection. AUCs for adenocarcinoma and squamous cell carcinoma were not significantly different, but the AUC for small cell lung cancer was significantly higher than that for adenocarcinoma (p<.001) and squamous cell carcinoma (p=.02). Clinically meaningful sensitivity to detect all stages of disease was achieved. Conclusions: A classifier developed using samples collected prospectively distinguished between lung cancer cases and controls with robust cross-validated performance across all stages and lung cancer subtypes. A cfDNA DELFI fragmentome test could represent an affordable, high-performing blood test that may improve lung cancer screening. Citation Format: Peter J. Mazzone, Kwok-Kin Wong, Jun-Chieh J. Tsay, Harvey I. Pass, Anil Vachani, Allison Ryan, Jacob Carey, Debbie Jakubowski, Tony Wu, Yuhua Zong, Carter Portwood, Keith Lumbard, Joseph Catallini, Nicholas C. Dracopoli, Tara Maddala, Peter B. Bach, Robert B. Scharpf, Victor E. Velculescu. Prospective evaluation of cell-free DNA fragmentomes for lung cancer detection. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5766.

e20521 Background: Annual lung cancer screening can save lives, but fewer than 10% of eligible persons participate each year. More widespread screening is hindered by cost, inaccessibility, and uncertainty over individual-level benefit vs risk. Screening rates could be raised by a simple, inexpensive initial blood test, if it were sensitive for cancer detection. The DELFI (DNA evaluation of fragments for early interception) technology uses low-coverage, whole-genome sequencing and machine learning to identify patterns of cell-free DNA (cfDNA) fragmentation associated with cancer. Here we report preliminary cfDNA analysis results from DELFI-L101 (NCT04825834), a prospective, observational, multistate case-control study to train and test DELFI classifiers for lung cancer detection. Methods: Enrollees were ≥50 years old with current or previous smoking histories of ≥20 pack-years and recent or planned chest CT imaging. Medical history was recorded at enrollment, and blood samples were collected for DELFI analysis. Repeated 10-fold cross-validation was used to develop a classifier for lung cancer detection. A split study approach is planned for independent validation of the classifier. Results: At this time, 242 individuals with lung cancer and 652 without cancer have enrolled. Most participants were ≥65 years old, and the proportions of men and women were similar. Lung cancer risk factors were present among both cases and controls. Like the lung cancer screening population, approximately half of lung cancer cases were stage I. Median DELFI scores were higher among individuals with lung cancer than no cancer, overall and across groups stratified by age or body mass index (BMI). Cross-validated area under the receiver operator characteristic curve was 0.81 for lung cancer detection. Clinically meaningful sensitivity to detect lung cancer was attained across all disease stages, with sensitivity increasing stepwise with stage. Conclusions: We developed a classifier based on cfDNA fragmentome patterns analyzed using DELFI that could differentiate between lung cancer cases and controls with cross-validated performance across age groups, BMI categories, and cancer stages. A blood-based DELFI fragmentome test could serve as a low-cost, high-performance blood test with potential to improve lung cancer screening efficiency. Clinical trial information: NCT04825834 .

We thank Drs. Grskovic, Hiller, and Woodward for their Letter to the Editor. We respectfully disagree with several of their statements and wish to point out a number of factual errors in their claims about both this publication1 and our clinical validation.2 Grskovic et al suggest that “analytical performance characteristics of laboratory assays do not necessarily translate to clinical performance.” We disagree that this statement is applicable to our assay and note that in any assay, where the affected/unaffected analyte distributions overlap, as is the case with our assay, greater precision necessarily leads to better test performance. They then argue that the clinical impact of their analytical performance would be minimal when using a 1% donor-derived cell free DNA (dd-cfDNA) cutoff, but contradict this argument by noting that “practitioners rely on quantitative assessment of dd-cfDNA level, [and] serial change in dd-cfDNA.” Below, we clarify the points made by Grskovic et al. First, our calculation of the coefficient of variation (CV) for within-run variability used appropriate method-specific cfDNA inputs, with differences reflecting our higher yielding cfDNA isolation methodology. While 83% of the clinical samples referred to in the Grskovic et al study contained at least 8ng cfDNA, the input used in their analytical validation,3 95% of our clinical samples contain at least 30ng cfDNA, the input amount used in this study. Thus, our reported within-run CV of 1.85% was calculated using more conservative bounds than the 9.2% CV reported by Grskovic et al.3 Second, the across-run CV of 4.29%1 for actual patient samples reported herein represents samples under 2% dd-cfDNA and thus is comparable to the CV of 7.7% reported by Grskovic et al.3 Further, the across-run CV of 1.99% at 45ng input cfDNA for contrived samples reported herein1 is comparable to the CV of 4.5% at 60ng input cfDNA reported by Grskovic et al.3 Grskovic et al then try to support their contention that better analytical performance may not translate to better clinical performance; however, their claims around our clinical validation study2 either fall short of this end or are incorrect. The published area under the curve (AUC) for the assay we used was measured as 0.87 in a study with > 200 biopsy-matched samples,2 as compared to the smaller sample size and AUC of 0.74 for the assay reported by Bloom et al,4 clearly demonstrating better discriminatory power for our assay in a clinical setting. The study by Bloom et al4 indeed demonstrated a negative predictive value (NPV) of 84% when distinguishing rejection from nonrejection in a high-risk cohort with a 28% prevalence of rejection; we note that this compares to an NPV of 94% for our assay, when projected to the same cohort. Grskovic et al also suggest that the rejection samples in our clinical validation study2 were strictly from for-cause biopsies, while many of our nonrejection samples were from patients undergoing surveillance biopsies, and that this may serve to artificially inflate our AUC. Their assertion is incorrect: 34% (13/38) of our rejection samples were from patients undergoing surveillance biopsies,2 and the proportion of rejection cases in both the for-cause (24%; 25/103) and surveillance biopsy cohorts (11%; 13/114) match expected prevalences. Furthermore, performance numbers were calculated separately for the for-cause and surveillance protocol biopsy cohorts,2 and they are similar, indicating that there was no cohort-based AUC inflation. A similar validation of their assay in a surveillance setting is not available. Additionally, their claim that we deviated from Banff 2017 criteria is mistaken. If we exclude the cases in our clinical validation that Grskovic et al find questionable—those with both T-cell-mediated rejection and borderline antibody-mediated rejection—our sensitivity for T-cell-mediated rejection cases remains 100% (8/8)2 as compared to 27% (3/11) in Bloom et al.4 As described above, the claims made in the Letter by Grskovic et al are either incorrect or misleading; the published performance numbers indicate superior performance of this assay.1,2 In summary, we look forward to more studies demonstrating the performance and value of dd-cfDNA. ACKNOWLEDGMENT Paul R. Billings, MD, PhD, Solomon Moshkevich, MBA, Jonathan Sternberg, PhD, Adam Prewett, MBA, Rosalyn Ram, PhD, Ryan Swenerton, PhD, and Maxim Brevnov, PhD, participated in helpful discussions.

Determine if differences in fetal fraction (FF) are observed in donor oocyte pregnancies compared to the general population. Retrospective analysis Noninvasive prenatal testing (NIPT) samples from singleton pregnancies were analyzed at a single reference lab. NIPT was performed using a SNP-based method with FF measured as previously described.1 FF from 1611 donor oocytes was analyzed and compared to a large set of reference cases matched for maternal weight (MW) and gestational age (GA). A z-score was calculated for each donor oocyte compared to its reference data. If no impact to FF from the use of donor or IVF, the average z-score is expected to be zero. Statistical analysis was performed using a z-test to establish if this was the case. For donor cases the average z-score was -0.4. A z-test determined this deviation from normal to be significant (p < 0.00001), showing that donor cases have lower FF than their corresponding reference data. The average MW was 154.3 lbs. (range 79.2-370.4 lbs.), average GA was 12.9 weeks (range 9-33 weeks) and average FF was 8.4%. The adoption of NIPT over other screening and diagnostic methods continues to grow, especially among women using donor oocyte/IVF. This population's preference for NIPT may stem from increased anxiety, higher false positive rates with traditional serum screening and avoidance of diagnostic procedures carrying miscarriage risk. Therefore, understanding the differences in FF in this population is critical.2-5

Standard noninvasive methods for detecting renal allograft rejection and injury have poor sensitivity and specificity. Plasma donor-derived cell-free DNA (dd-cfDNA) has been reported to accurately detect allograft rejection and injury in transplant recipients and shown to discriminate rejection from stable organ function in kidney transplant recipients. This study used a novel single nucleotide polymorphism (SNP)-based massively multiplexed PCR (mmPCR) methodology to measure dd-cfDNA in various types of renal transplant recipients for the detection of allograft rejection/injury without prior knowledge of donor genotypes. A total of 300 plasma samples (217 biopsy-matched: 38 with active rejection (AR), 72 borderline rejection (BL), 82 with stable allografts (STA), and 25 with other injury (OI)) were collected from 193 unique renal transplant patients; dd- cfDNA was processed by mmPCR targeting 13,392 SNPs. Median dd-cfDNA was significantly higher in samples with biopsy-proven AR (2.3%) versus BL (0.6%), OI (0.7%), and STA (0.4%) (p < 0.0001 all comparisons). The SNP-based dd-cfDNA assay discriminated active from non-rejection status with an area under the curve (AUC) of 0.87, 88.7% sensitivity (95% CI, 77.7–99.8%) and 72.6% specificity (95% CI, 65.4–79.8%) at a prespecified cutoff (>1% dd-cfDNA). Of 13 patients with AR findings at a routine protocol biopsy six-months post transplantation, 12 (92%) were detected positive by dd-cfDNA. This SNP-based dd-cfDNA assay detected allograft rejection with superior performance compared with the current standard of care. These data support the feasibility of using this assay to detect disease prior to renal failure and optimize patient management in the case of allograft injury.

Background. Early detection of rejection in kidney transplant recipients holds the promise to improve clinical outcomes. Development and implementation of more accurate, noninvasive methods to detect allograft rejection remain an ongoing challenge. The limitations of existing allograft surveillance methods present an opportunity for donor-derived cell-free DNA (dd-cfDNA), which can accurately and rapidly differentiate patients with allograft rejection from patients with stable organ function. Methods. This study evaluated the analytical performance of a massively multiplexed polymerase chain reaction assay that targets 13 962 single-nucleotide polymorphisms, characterized and validated using 66 unique samples with 1064 replicates, including cell line-derived reference samples, plasma-derived mixtures, and transplant patient samples. The dd-cfDNA fraction was quantified in both related and unrelated donor-recipient pairs. Results. The dd-cfDNA assay showed a limit of blank of 0.11%, a limit of detection and limit of quantitation of 0.15% for unrelated donors, and limit of blank of 0.23%, a limit of detection and limit of quantitation of 0.29% for related donors. All other metrics (linearity, accuracy, and precision) were observed to be equivalent between unrelated and related donors. The measurement precision of coefficient of variation was 1.8% (repeatability, 0.6% dd-cfDNA) and was <5% for all the different reproducibility measures. Conclusions. This study validates the performance of a single-nucleotide polymorphism-based massively multiplexed polymerase chain reaction assay to detect the dd-cfDNA fraction with improved precision over currently available tests, regardless of donor-recipient relationships.

We analyzed maternal plasma cell-free DNA samples from twin pregnancies in a prospective blinded study to validate a single-nucleotide polymorphism (SNP)-based non-invasive prenatal test (NIPT) for zygosity, fetal sex, and aneuploidy. Zygosity was evaluated by looking for either one or two fetal genome complements, fetal sex was evaluated by evaluating Y-chromosome loci, and aneuploidy was assessed through SNP ratios. Zygosity was correctly predicted in 100% of cases (93/93; 95% confidence interval (CI) 96.1%–100%). Individual fetal sex for both twins was also called with 100% accuracy (102/102; 95% weighted CI 95.2%–100%). All cases with copy number truth were also correctly identified. The dizygotic aneuploidy sensitivity was 100% (10/10; 95% CI 69.2%–100%), and overall specificity was 100% (96/96; 95% weighted CI, 94.8%–100%). The mean fetal fraction (FF) of monozygotic twins (n = 43) was 13.0% (standard deviation (SD), 4.5%); for dizygotic twins (n = 79), the mean lower FF was 6.5% (SD, 3.1%) and the mean higher FF was 8.1% (SD, 3.5%). We conclude SNP-based NIPT for zygosity is of value when chorionicity is uncertain or anomalies are identified. Zygosity, fetal sex, and aneuploidy are complementary evaluations that can be carried out on the same specimen as early as 9 weeks’ gestation.