Using Disease-Agnostic Genomic Liquid Biopsy in Complex Diagnoses: Real-Life Example of cfChIP-Seq in a Fever of Unknown Origin

Introduction

The advent of noninvasive liquid biopsy methodologies aims to revolutionize the diagnostic process. Most of the progress in liquid biopsies is on assays for concrete diagnostic questions, such as genetic aberrations (eg, noninvasive prenatal testing), disease-specific tests (eg, detection of colorectal cancer), and patient-tailored tests (eg, cancer recurrence, rejection of a transplanted organ). Such assays fall neatly into the role of a laboratory test, with a well-defined range of outcomes. Consequently, they are evaluated and regulated using clear clinical metrics and considerations.¹

The clinic can also benefit from a different type of liquid biopsy that is disease-agnostic and provides rich information that can be interrogated to evaluate the presence of various conditions and pathologies. Unlike multicancer early detection tests, these do more than test for a list of diseases. They report on underlying molecular entities, such as RNA,² nonhuman DNA,³ fragmentomics,⁴ and epigenetics.^5,6 By their nature, these assays reflect the patient's complex molecular state. Such methods can take a role akin to that of traditional biopsies in diagnosis, where a pathology report describes a range of findings that may either pinpoint a specific pathology or, when less definitive, narrow the range of possible diagnostic hypotheses.

Below, we illustrate this concept through a case in which cell-free ChIP-seq,⁶ which provides a genome-wide assessment of promoter activity in the cells of origin, was applied to a patient with fever of unknown origin (FUO). The test provided a rapid and definitive diagnosis of diffuse large B-cell lymphoma (DLBCL), and if applied earlier in the diagnostic sequence (Fig 1), it might have altered the clinical outcome.

FIG 1.

Timeline of patient hospitalization: The patient's hospitalization course, from admission (day 0) to demise (day 27), is outlined. Key diagnostic procedures and clinical manifestations are shown. The working diagnosis evolved as new information became available. CT, computed tomography; DLBCL, diffuse large B-cell lymphoma; FDG, fluorodeoxyglucose; MRI, magnetic resonance imaging; PET, positron emission tomography.

Methods

Immunoprecipitation, Next Generation Sequencing Library Preparation, and Sequencing

Sample collection and handling, immunoprecipitation, library preparation, and sequencing were performed by Senseera Ltd as previously reported,⁶ with certain modifications that increase capture and signal-to-background ratio. Briefly, ChIP antibodies were covalently immobilized to paramagnetic beads and incubated with plasma. Barcoded sequencing adaptors were ligated to chromatin fragments and DNA was isolated and next-generation sequenced.

Computational Analysis

Reads were aligned to the human genome (hg19) using bowtie2 with “no-mixed” and “no-discordant” flags. We discarded fragment reads with low alignment scores (-q 2) and duplicate fragments.

Preprocessing of sequencing data was performed as previously described.⁶ Briefly, the human genome was segmented into windows representing TSS, flanking to TSS, and background (the rest of the windows). The fragments covering each of these regions were quantified and used for further analysis. Nonspecific fragments were estimated per sample and extracted, resulting in the specific signal in every window. Counts were normalized and scaled to 1 million reads in the healthy reference, accounting for sequencing depth differences. Analysis of significantly increased gene promoters was performed as previously described⁶ by comparing the signal level with the distribution of gene signals in a healthy cohort. Statistical correction for multiple testing was performed using Benjamini-Hochberg false discovery rate.⁷ Detailed information regarding these steps is provided in the Data Supplement.⁶ Enrichment analysis was performed using the Enrichr website.⁸

B-cell H3K4me3 ChIP-seq samples were downloaded from the Roadmap Epigenomics project¹¹ (samples E031 and E032⁹). DLBCL H3K4me3 ChIP-seq samples were downloaded through the IHEC data portal (samples: EGAX00001326557, EGAX00001326444, EGAX00001326560, EGAX00001364321, EGAX00001326551, EGAX00001326554¹⁰). To identify genes high in B-cell and DLBCL samples (Fig 2E), we selected genes with a group-specific mean expression of >10 and a corresponding healthy plasma reference mean of <3.

FIG 2.

cfChIP-seq analysis. cfChIP-seq results are consistent with DLBCL diagnosis. (A) Scatter plot comparing the gene-by-gene profile of the FUO patient (y-axis) with the average healthy profile (x-axis). Genes whose signal was significantly above healthy were defined as “over-represented” (636 marked genes). The inset on the bottom-right corner compares a random healthy donor with a healthy profile. (B-D) Bar charts showing the results of an enrichment analysis of the “over-represented” gene group versus (B) ARCHS4 tissues, (C) GO biological process, and (D) ARCHS4 cell line annotation databases. The top seven hits are shown in each panel. (E) Overlap between genes high in the patient with FUO and genes high in B-cell ChIP-seq and DLBCL ChIP-seq. B-cell ChIP-seq data were taken from the ROADMAP Epigenomics repository,¹¹ and DLBCL ChIP-Seq data were taken from the Blueprint repository.¹² (F) Genome browser showing the ChIP-seq signal in a few representative genes from the groups shown in (E). DLBCL, diffuse large B-cell lymphoma; FUO, fever of unknown origin.

Ethical Statement

Samples were collected from the patient after obtaining informed consent to participate in this study (Hadassah Medical Center, institutional review board (IRB) #0198-14-HMO to B.G.). Consent for publication was obtained from the patient.

Results

Case Report

A 50-year-old woman presented to the emergency department with a 2-week history of fever and a 4-day history of perineal paresthesia. She also reported associated symptoms like sore throat, muscle pain, and a single episode of diarrhea. Notably, she denied any urinary or bowel dysfunction, shortness of breath, chest pain, abdominal pain, nausea, vomiting, rashes, night sweats, or weight loss.

FUO presents a diagnostic challenge, and despite costly and time-consuming comprehensive investigations, the cause of a patient's fever often remains elusive. While infectious and inflammatory etiologies are the most frequently identified culprits, accounting for 36% and 19% of cases, respectively,¹³ malignancy can also masquerade as FUO. Even after thorough examination and workup, the underlying etiology remains unknown in up to 51% of cases.¹⁴

The patient's medical history was significant for breast carcinoma (invasive duct carcinoma, stage I, triple-positive, pT1bN0Mx) diagnosed 3.5 years prior. She underwent successful treatment with lumpectomy, chemotherapy (including Taxol and Herceptin), radiotherapy, and hormonal therapy (anastrozole). At the time of admission, she was considered to be in complete remission.

In patients with a history of malignancy, FUO might suggest disease recurrence. In the setting of recurrent breast cancer, unexplained fever can raise suspicion for liver metastasis.¹⁵ However, there is growing concern regarding the development of secondary malignant neoplasms, including hematologic malignancies, years after the initial diagnosis of a primary breast tumor. While prior studies have documented an elevated risk of secondary myeloid neoplasms, particularly AML and myelodysplastic syndromes, after breast cancer treatment, data remain limited regarding the risk of lymphoid malignancies in this patient population.^16,17

The initial workup included a thorough clinical history and physical examination, laboratory tests, and imaging studies. The physical examination was unremarkable. Laboratory tests showed mild hyponatremia, slightly elevated alkaline phosphatase, and elevated C-reactive protein, suggesting inflammation or infection. The chest X-ray was unremarkable. An extensive infectious workup, including viral and bacterial panels, was negative. Computed tomography (CT) scans of the head and spine and a magnetic resonance imaging (MRI) of the pelvis did not reveal any space-occupying lesions or a clear source of infection.

Given the persistent fever and neurologic complaints, a lumbar puncture was performed and revealed no signs of meningitis or other pathology. A rheumatologic workup was also negative. A spinal MRI revealed a fluid channel in the distal part of the conus medullaris; however, no corresponding etiology for the patient's clinical presentation was identified.

The patient's condition continued to deteriorate. A full-body CT scan revealed new findings of hepatosplenomegaly but no definitive signs of infection, abscess, or solid tumor. Repeat blood tests showed elevated lactate dehydrogenase, prompting a blood smear followed by a bone marrow biopsy. The biopsy showed no signs of hematologic disorders or malignancies.

Given the persistent perineal paresthesia, a multidisciplinary evaluation by gynecology and orthopedics was undertaken. No localized pathology was identified. During this time, the patient developed urinary retention, necessitating bladder catheterization. Subsequently, on day 11, the patient developed fecal incontinence, requiring urgent spinal decompression surgery. A biopsy from the S1-S2 vertebrae showed normal bone tissue, bone marrow, and connective tissue.

Fluorodeoxyglucose positron emission tomography (PET)-CT imaging revealed widespread increased metabolic activity in various locations, including the breasts, pancreas, peritoneum, adrenal gland, orbit, ovaries, spine, and bone marrow. This finding strongly suggested a rapidly progressing malignancy.

A biopsy was crucial for a definitive diagnosis. However, obtaining tissue samples proved to be challenging because of the location of the suspicious lesions (eg, adrenal gland and orbit). Figure 1 shows the time course of the clinical deterioration and the diagnostic evaluation.

cfChIP-Seq as a Diagnostic Tool

We have developed a liquid biopsy method, cell-free ChIP-seq, which profiles the histone modification patterns of cell-free DNA (cfDNA).⁶ cfChIP-seq exploits the fact that the major form of cell-free DNA in plasma is nucleosomes, complexes of DNA with histone proteins.¹⁸ Within a living cell, nucleosomes are post-translationally modified by the transcription apparatus, reflecting transcriptional activity. On cell death, nucleosomal DNA is protected from cleavage, resulting in mostly mono- and dinucleosomal particles reaching the bloodstream. By isolating circulating nucleosomes originating from active promoters and sequencing the associated DNA fragments, we can profile promoter gene activity within cells of origin, providing a window into cellular processes reflected in the cell-free DNA pool.

Since the publication of the method, we and others established that cfChIP-seq captures cell type– and cell state–specific promoters in relevant diseases—myofibroblast promoters in cardiac patients,^6,19 hepatocyte promoters in a range of liver diseases,^6,20 and cancer-specific promoters in the relevant cancer types, including colorectal cancer, lung cancers, liver cancers, and multiple myeloma.^6,21-25 Moreover, tumor-specific attributes are reflected in the cfChIP-seq profiles, such as high ERBB2 in patients with HER2 amplification.^6,26,27 We demonstrate a correspondence between cfChIP-seq measurement from plasma to gene expression in the tumor using matched blood samples and tissue biopsies.²¹

Application of cfChIP-Seq in This Case

With ongoing diagnostic uncertainty, nonconventional diagnostic assays were considered 3 weeks after admission. After obtaining informed consent (under existing IRB approval), blood was drawn for cfChIP-seq assay and analysis.

Analysis of promoter activity (H3K4me3) in the patient's sample identified 636 gene promoters with significantly higher signal when compared with a cohort of approximately 1,200 healthy donors (fold change >4, q value <10⁻³ after false discovery rate multiple hypothesis correction⁷). These are highlighted in Figure 2A and listed in the Data Supplement (Table S1).

By quantifying parameters directly linked to gene expression, we can compare our findings with existing knowledge regarding each gene's involvement in health and disease. We performed an enrichment analysis of the over-represented gene group versus the ARCHS4 (tissues and cell lines) and GO (biological process) annotation databases using the Enrichr interface⁸ (Figs 2B and 2C). The analysis revealed a significant enrichment of CD19⁺ B cell–related genes (ARCHS4 tissues) and specifically to genes involved in B-cell proliferation (GO Biological Process). These findings strongly suggest that the patient has an increased turnover of B cells or B cell–related cells. Furthermore, enrichment analysis against the ARCHS4 cell line database is strongly associated with various B-cell lymphoma lines, with the most robust enrichment for the PFEIFFER cell line derived from a patient with DLBCL (Fig 2D).

Given these enrichments, we contrasted the patient's profile against ChIP-seq profiles of B cells and DLBCL cells from the literature^11,12 (Figs 2E and 2F). This comparison shows that the patient has high signal in DLBCL-specific (non–B cells) genes, such as RGS13 and HTR3A (111 genes of 327 DLBCL-specific genes, P < 10⁻⁶²).

To conclude, cfChIP-seq analysis strongly indicated rapid, aberrant B-cell proliferation and death, consistent with a diagnosis of DLBCL.

After blood was obtained for cfChIP-seq analysis, conventional diagnostic procedures continued. A small pelvic mass was identified by imaging, allowing for an ultrasound-guided biopsy. While the pelvic mass pathology was pending, the cfChIP-seq results became available and were consistent with DLBCL as described above. Biopsy results obtained the following day corroborated the diagnosis (Fig 3). Tissue biopsy was notable for aggregates and sheets of medium-sized and large atypical lymphoid cells. The typical cells were highlighted by immunohistochemistry stains for leukocyte common antigen and CD20 and did not stain for CD3, pancytokeratin, and GATA3. The MIB1 (ki67) proliferation index was 40%. These pathology findings were sufficient for the diagnosis of DLBCL. A week later, additional stains were examined and were positive for bcl-6 and MUM-1. The final diagnosis was DLBCL, nongerminal center (activated B-cell) type.

FIG 3.

Lymph node pathology results. Pathology stains of the patient's pelvic lymph node biopsy are shown. (A) H&E, tumor cells have pleomorphic large nucleoli with apoptotic bodies. (B) CD20 IHC stain highlights extensive mass involvement by lymphoblasts. (C) MIB-1 IHC stain reflects the high proliferative activity of tumor cells. (D) LCA IHC stain reflects strong CD45 expression. All images were taken at ×20 magnification. H&E, hematoxylin and eosin; IHC, immunohistochemistry; LCA, leukocyte common antigen; MIB, MIB-1, anti-KI-67, monoclonal antibody.

Unfortunately, the patient's condition continued to deteriorate rapidly before treatment could be initiated. She developed complications, including lactic acidosis, elevated liver enzymes, and prolonged bleeding time. Chemotherapy with cyclophosphamide was initiated, but the patient's condition worsened, and she succumbed the following day.

Discussion

This case highlights the potential clinical utility of unbiased cfDNA analysis for diagnosing diseases of unknown origin when initial symptoms are nonspecific. The extensive, conventional workup, including blood tests, imaging modalities, consultations with specialists, and even initial negative biopsies from several sites, demonstrates the limitations of these procedures in pinpointing the culprit in a time-appropriate manner. The cfChIP-seq assay and a reasonably straightforward subsequent analysis pinpointed the disease within days.

Several important aspects must be considered here. First, as a noninvasive test, this assay could have been administered during early diagnosis without harm to the patient as opposed to the many invasive procedures she underwent that were not diagnostic. Experience with hundreds of patients with cancer at different stages^6,28 strongly suggests that DLBCL signs would have been prominent also at such an earlier time point, mainly because widespread disease was prominent on the PET-CT scan that was performed on day 10.

Second, unlike localized tissue biopsies, liquid biopsies are systemic tests. In situations involving unknown disease sites, such methods can provide unique information.^29,30 This aspect also highlights one of the limitations of such methods; they depend on the amount of cell-free DNA shed by the disease. If the pathology does not involve the shedding of cell-free DNA into the circulation, these methods cannot detect it. However, most tumors shed some amount of cell-free DNA,^28,31 and the detection ability can be improved by increasing the sensitivity of the methods through better molecular procedures, computational analysis, and the amount of plasma assayed.

Third, this study was conducted in a research setting, resulting in a 4-day turnaround time. Clearly, as these methods mature into a diagnostic product, this time lag can be dramatically reduced. In our hands, a 24-hour turnaround is feasible, depending on sequencing methods and automation in liquid handling and computational pipelines. Clearly, as with any new technology, there is a significant step involving dissemination, deployment, and logistics before it can be used routinely in clinical settings.

More crucially, the test is disease-agnostic and does not require a diagnostic hypothesis in advance. This flexibility builds on the clear connection between the measured entity and underlying cellular events (eg, gene expression in the cells of origin).^6,21 This connection allowed us to compare the cfChIP-seq profiles against the prior literature and to find that they strongly indicate DLBCL. This example and previous work on other pathologies show that the interpretation of cfChIP-seq can leverage extensive gene expression literature and publicly available databases. Significantly, cfChIP-seq was not specifically developed to detect DLBCL, nor was it trained on lymphoma samples. While there are targeted liquid biopsy assays for detecting DLBCL,³² these methods require active, deliberate investigation of DLBCL.

More broadly, this case showcases the difference between a targeted diagnostic assay (whether for a specific cancer or a list of cancer types) and an agnostic assay that provides a genome-wide molecular profile. Importantly, the same molecular assay can serve both classical diagnostic purposes and more open-ended analysis. The rich data from an agnostic assay (eg, cfChIP-seq) can address multiple targeted hypotheses through different analyses, each focusing on a specific diagnostic question (eg, detection of a particular cancer type). This combination of assay and analysis can be evaluated and validated within existing regulatory frameworks.

A major concern is in the interpretation of such genome-wide profiles. In this case, our interpretation drew on a mechanistic understanding of what the assay measures and how it relates to other molecular aspects (eg, expression). This understanding enables access to a vast and expanding body of knowledge on gene expression across multiple cell types, diseases, and physiologic conditions.^33,34 Today, such an interpretation requires human expertise. However, much of it can be streamlined and automated to support human analysis. Again, the analogy is to pathology, where knowledge and context are crucial for reaching definite conclusions. With the advent of artificial intelligence (AI) models for molecular data,³⁵ most, if not all, of the analyses described above could be performed automatically. The combination of AI with continuously growing online repositories will make such analysis more accessible across different clinical settings.

In conclusion, in this case, cfChIP-Seq testing provided a crucial piece of the diagnostic puzzle, facilitating the diagnosis. Had this test been clinically available and implemented early in the diagnostic protocol, a definitive treatment could have been implemented earlier, possibly altering the ultimate clinical course. In this case, raising the plausibility of a DLBCL diagnosis early would have weakened the strong hypotheses of recurrence of earlier breast cancer and would have led clinicians to widen the search for a relevant biopsy. As we enter a new era of precision medicine, the routine integration of such liquid biopsies into clinical practice is inevitable; while their standardized adoption will require further calibration through extensive clinical experience, revolutionary cases such as this serve as the necessary catalysts to accelerate this diagnostic transition.

AUTHOR CONTRIBUTIONS

Conception and design: Orly Agmon Gutin, Jenia Gutin, Nir Friedman

Financial support: Nir Friedman

Administrative support: Benjamin Glaser

Provision of study materials or patients: Orly Agmon Gutin

Collection and assembly of data: Orly Agmon Gutin, Jenia Gutin, Dania Jaber, Gabriella Snoyman, Bara Hmeidat, Benjamin Glaser

Data analysis and interpretation: Orly Agmon Gutin, Jenia Gutin, Ronen Sadeh, Nir Friedman

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).