Low-coverage whole-genome sequencing of extracellular vesicle-associated DNA in patients with metastatic cancer

Low-coverage whole-genome sequencing (LC-WGS) can provide insight into oncogenic molecular changes. Serum extracellular vesicles (EV) represent a novel liquid biopsy source of tumoral DNA. This study compared copy number alteration (CNA) profiles generated from LC-WGS of formalin-fixed paraffin-embedded (FFPE) tumoral DNA and EV-DNA obtained from cancer patients. Patients with squamous cell carcinoma of the base of tongue (n = 3) and cutaneous squamous cell carcinoma (n = 2) were included. LC-WGS (0.5-1X coverage) was performed on FFPE-DNA and serum EV-DNA. Similarity between CNA profiles was analysed using QDNAseq. FFPE samples had a mean CNA of 31 (range 17–50) over 1.9 × 109 (range 1.0–2.6 × 109) bp in length, and EV samples had a mean CNA value of 17 (range 7–19) over 7.6 × 108 (range 2.9–15 × 108) bp in length. A mean of 8 (range 0–21) CNA over 5.9 × 108 (range 1.6–14 × 108) bp in length was found to overlap between EV and FFPE-derived samples per patient. Although the mean correlation efficient between samples was r = 0.34 (range − .08 to 0.99), this was not statistically significant (p > 0.05). Regions of highest deletion and duplication in FFPE samples were not well reflected in the EV-DNA. Selected CNA regions in EV-associated DNA were reflective of the primary tumor, however appreciation of global CNA and areas of most significant change was lost. The utility of LC-WGS of EV-derived DNA is likely limited to molecular alterations of known interest.

www.nature.com/scientificreports/ Copy number alteration (CNA) defined as a variable copy number of a segment of deoxyribonucleotides > 1 kb in size is a commonly observed major genetic alteration identified in most cancers [13][14][15][16] , that has been postulated to be both a predictive and prognostic biomarker 14,15 . As cancer genomes are known to evolve over time CNA can also be observed to fluctuate, and CNA can also arise due to treatment response or resistance 14 . Therefore, DNA extracted from circulating EV from cancer patients' biofluids can facilitate a convenient and non-invasive continual assessment of CNA changes in a tumor during treatment, and can provide insight into the mechanisms behind disease progression. There are a number of approaches currently available to assess CNA in formalinfixed paraffin-embedded (FFPE) tumoral tissues including microarray-based comparative genomic hybridization (array-CGH), single nucleotide polymorphism (SNP) arrays, molecular inversion probe (MIP) assays, and low coverage whole genome sequencing (LC-WGS) 16,17 . Traditionally CNA is assessed using high yields of DNA obtained from traditional tumoral biopsies, whilst EV-associated DNA is isolated in significantly smaller quantities. Kader et al. described a LC-WGS method which successfully characterized CNA using ultra-low input of tumoral FFPE-derived DNA 18 and similarly, CNA has been demonstrated to be successfully profiled in smaller quantities of circulating-tumor DNA (ct-DNA) using LC-WGS 19 . Thus, LC-WGS could be used to assess EV-associated DNA.
This study aimed to assess whether LC-WGS of EV-associated DNA is reflective of tumoral DNA based on their respective CNA profiles, to define the potential clinical utility of this liquid biopsy approach. We compared CNA profiles generated from LC-WGS of DNA extracted from formalin-fixed paraffin-embedded (FFPE) samples and matched EV isolated from the serum of cancer patients.

Results
Demographic data. Demographic data of patients enrolled in the study are summarized in Table 1 (n = 5).
Three patients had metastatic base of tongue squamous cell carcinoma (BOT), with two patients having Human Papilloma Virus (HPV)-related disease. Two patients had metastatic cutaneous squamous cell carcinoma (cSCC). Patients had a median age of 64 (range 59-77) years, and were predominantly male. Median follow up was 32 (range [22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41] months and at data cut-off, two patients had died of disease and three patients were alive and responding to treatment. Four patients had their EV samples collected for analysis within one month of their FFPE-tumor samples being obtained, and one patient (Study ID: C008) had their EV samples collected 12 months after their FFPE-tumor sample. All BOT patients have received radiation, chemotherapy and immunotherapy as treatment for their disease. All cSCC patients have received surgery and immunotherapy as treatment for their disease.

Validation of isolated EV & EV-DNA characteristics.
Validation of EV isolation from differential ultracentrifugation (UC) was carried out using Transmission Electron Microscopy (TEM), Nanoparticle Tracking Analysis (NTA), and Western blotting with collective results presented in Fig. 1. Isolated EV analyzed in triplicate (Fig. 1A) by NTA show a mean particle size of 215 nm and mean concentration of 2.03 × 10 10 particles/mL. Figure 1B shows a ds-DNA antibody labelled TEM image of a representative isolated EV, which demonstrates a round particle with a diameter of approximately 200 nm. Ds-DNA antibody positive particles are evident within the outline of the vesicular membrane. Figure 1C demonstrates the Western blot results using common exosomal markers (positive for CD9 and negative marker Calnexin) in melanoma cell lysate (SKMel28) as positive control (lane 10) and isolated EV (lane 7).
A minimum of 10 ng of extracted EV-associated DNA was used for each LC-WGS run. One representative patient (#B-010) sample is shown in Fig. 2, which was analyzed using High Sensitivity DNA bioanalyzer (Agilent) to assess sample preparation, quality and fragment size of the EV-associated DNA extracted. The percentage of reads mapped from all patients' EV-associated DNA ranged between 55.6 and 97.1% (mean = 71.6). Median fragment length (bp) ranged between 169 and 178 bp (mean = 176 bp). LC-WGS was performed aiming for 0.5-1 × coverage, and was considered adequate to generate CNA profiles.   Fig. 3 summarises the global CNA count (total number of deletions and duplications) for EV and FFPE samples of each patient, as well as areas with significant duplications and deletions. The majority of CNA in FFPE samples were deletions rather than duplications, and this was reflected in EV samples. However, the number of CNA identified in EV-DNA were considerably less than found in FFPE. Summarised in Table 2 are the regions of deletion/duplication in both number of CNA, and number of basepairs, the areas of overlapped CNA region, Pearsons correlation value (r) calculated of CNA in the overlapped regions, associated p values using paired sample t-test, and regions with significant (defined as high frequency) deletion/duplication between FFPE and EV samples of all patients included in the study. As noted in Fig. 3, EV samples reflected some specific CNA in FFPE but in general were less in number and some CNA were unique to EV samples suggesting the limitation of tumor sampling from both FFPE and EV. Despite the overlap, none of the correlation values were statistically significant with p values all greater than 0.05.

Comparison of CNA profiles of FFPE and EV samples.
The concordance of CNA profiles between matching FFPE tumors and EV samples are shown in Fig. 4. Overall, although common CNA and regions were identified between sample types, there were many areas under-represented by the EV-DNA based analyses. Again, some CNA were unique to each sample type demonstrating sampling limitations. In the HPV-driven metastatic BOT group (n = 2), patient #B-005 FFPE tumoral DNA demonstrated significant regions of deletion detected in Chr 3p and 8p, and duplication in Chr 3q, 5p and 8q, 20p and 20q. These observations are consistent with findings from The Cancer Genome Atlas (TCGA) head and neck cancer cohort and a large HPV-positive and HPV-negative oral squamous cell carcinomas data set published by Gillison et al. 21,22 . This patient's profile also contained recurrent focal amplification for 3q26/28, a  www.nature.com/scientificreports/ region involving squamous lineage transcription factors TP63 and SOX2 and the oncogene PIK3CA, which was also a group of mutations found in the TCGA dataset 21 . In contrast, patient #B-010 had a significant deletion observed in Chr 1q while regions with duplication were observed in Chr 8q, 11q and 20q. These findings are not observed in prior published datasets 21,22 . In the non HPV-driven metastatic base of tongue patient (#B-014), the only finding that is consistent with prior published dataset is in the duplication observed in Chr 3q. In the metastatic cutaneous squamous cell cancer samples, consistent with published literature were regions of duplication found in Chr 7p, 8q, and 20q seen in both FFPE and EV samples 23 .

Discussion
LC-WGS is a rapid and practical approach to characterize large-scale chromosomal alterations across the genome, which can be used to monitor tumoral molecular evolution 14 . Non-invasive liquid biopsy approaches such as cf-DNA, ct-DNA or EV-associated DNA have emerged as viable alternatives to using traditional tumoral tissue which require an invasive biopsy. Studies applying LC-WGS using cf-DNA and ct-DNA analysis have demonstrated the ability to detect tumour-associated CNA changes in cancer patients with various cancer types including prostate cancer, colorectal cancer, lung cancer, urothelial cancer and neuroblastoma [24][25][26][27][28][29][30][31] . Changes in CNA profiles of cf/ct-DNA have also been to be associated clinical disease progress such as cancer proliferation, and treatment response 32,33 . However, LC-WGS is restricted by the use of samples in which the tumor fraction is relatively high (> 10%), thus the search for alternative sources of circulating DNA is critical 33 . A recent study found that EV-DNA contained a higher mitochondrial DNA copy number than that found in cf-DNA of patients with hepatocellular carcinoma using whole genome and capture-based sequencing, suggesting EV-DNA may be an advantageous source of alternative liquid biopsy 34 . Our study is one of the few published studies that has investigated the utility of LC-WGS as a method of CNA profiling using low-input EV-associated DNA from cancer patients' plasma with comparison to matched FFPE samples. EV have been reported to contain tumoral DNA elements such as highly expressed oncogenic gene mutations in several cancer models including melanoma, lung, pancreatic and colorectal cancer [9][10][11]35,36 . However, few studies have directly compared whole genome sequencing concordance between EV-associated DNA and the primary tumor. One study identified tumoral-related copy number variations (CNV) in 3 metastatic pancreatic patients' EV-DNA 37 . Another study which has compared CNA profiles analysis using LC-WGS at 0.6X depth between EV-DNA and FFPE tumoral DNA was published by Lee 38 . This demonstrated that in nine urothelial bladder carcinoma patients, CNA analyses in EV-DNA and cf-DNA reported concordance with tumoral DNA (r = 0.412 and 0.481 respectively, p values not provided) 38 . Further, the authors were able to confirm that areas of CNA corresponded to regions with known mutations of interest including loci involving CDKN2A and RB1 38 .
Our study found that while overlapping regions of CNA in the EV samples and FFPE samples are observed, the correlation coefficients were highly variable (range: − 0.49 to 0.99) with no correlation being found to be statistically significant (p values > 0.05 from paired sample t-test). Regions in FFPE samples with significant duplication or deletion were not reflective of the regions found in EV samples, suggesting that EV-derived DNA probably represents a portion of tumoral DNA and non-tumoral DNA. Changes unique to EV-DNA could arise from non-tumoral DNA, mitochondrial DNA, or due to FFPE and EV sampling limitations. Overall there were significantly more regions of CNA detected in FFPE-DNA than in EV-DNA, likely due to its reported fragmented nature, which renders it difficult to detect larger CNA regions. These observations have identified some limitations of using CNA profiles generated from EV-associated DNA in the clinical setting.
It is difficult to make comparisons between studies, given the different cancer types, different methodologies used for EV isolation, and that global numbers of CNA and length were not reported in the discussed studies above. Our study also demonstrated that EV-DNA could identify CNA regions containing known cancerrelevant genes concordant with FFPE. However, we also showed that the global architecture of the molecular changes observed in FFPE and the regions of greatest change were lost with the limited view provided by the liquid biopsy approach. Specifically, we demonstrated that FFPE had a larger number of CNA spanning over a number of base pairs compared to EV-DNA and highest regions of loss or gains were observed in different regions between samples. EV-DNA can still potentially be used to identify specific mutations of known interest which was not pursued in depth within our study given the low coverage of the whole genome sequence. We recently reported that EV-DNA was able to detect circulating tumoral HPV-DNA in HPV-driven oropharyngeal carcinomas, however the detection sensitivity was significantly lower than cf-DNA analyses (p = <0.001) 39 . One study using LC-WGS CNA profiles of cf-DNA reported good concordance with FFPE in 22 metastatic prostate cancer patients (r = 0.87, p < 0.001) 40 . It is however recognised that cf-DNA contains different DNA elements to EV-DNA and this may explain the difference reflected in the results 41 . Our findings calls for further research in comparing the clinical utility of EV-DNA to cf-DNA which is an already established, well validated source of liquid biopsy. One direction to improve similar studies in the future could be to use immunocapturing of specific EV subpopulations that are postulated to be involved in carcinogesis, thus increasing sensitivity of the EV-DNA analysis for the detection of tumour-derived DNA 42 . However, the ability to isolate circulating DNA that is tumour-specific remains a challenge unless a tumour is defined by the presence of known oncogene-addiction which is not the case for most malignancies.
There are limitations to this study. The study lacked the ability to standardize of amount of input EV-associated DNA used in LC-WGS due to limited yield which may have contributed to the variations in different CNA profiles. Due to the paucity of similar studies published in literature, an optimal amount of EV-DNA input has not yet been determined. We did not specifically select a specific EV subpopulation, which may alter EV-DNA yield and quality, with a recent study showing that large EVs (those > 1000 nm) contain more tumour-derived DNA than small EVs (< 200 nm) 43 . We also note that at our sequencing depth 0.5-1X and an average of 50 million reads (see "Methods and materials" section below), we may not have achieved the most optimal sequencing sensitivity of smaller tumour specific fragments. There are also limitations to using QDNASeq to perform CNA profiling, firstly the bioinformatic package cannot completely correct for technical bias in read coverage, and secondly, the copy number calling can still be impacted by DNA quality 44 . Additionally, our cohort of patients was small in number, included different tumor types and varied treatments. It has been demonstrated that different types of tumors with varying levels of genomic instability may result in different amounts of EV as well as diverse molecular packaging, and thus result in different types of DNA isolated 43  www.nature.com/scientificreports/ impacted by the varied sample collection times. Finally, concordance of mutations would be more comprehensively assessed using orthogonal methods with deeper sequencing resolution or targeted sequencing which were not included in our study.

Conclusion
Based on our results, we suggest that although EV-associated DNA CNA profiles were reflective of FFPE, the overlapping regions were limited in number, highly variable between patients, and did not globally reflect regions with significant alterations found in FFPE. Whilst EV-associated DNA may be useful as a liquid biopsy source to identify tumoral DNA, overall, they do not globally reflect primary tumor DNA CNA profiles. In a clinical setting, EV-associated DNA alone may not be suitable as a source to be used to monitor global tumoral DNA changes. DNA extraction. DNA extraction was carried out as per method previously published by our group 39 . DNA was extracted using the QIAamp DNA Microkit (Qiagen, Australia) according to the manufacturers' instructions. DNA concentration was quantified using the Qubit high-sensitivity dsDNA kit (Thermo Fisher Scientific, Australia). DNA quality of one patient (study ID #B-010) was also characterized using High Sensitivity DNA bioanalyzers (Agilent Technologies, Australia).

Methods and materials
Low-pass whole genome sequencing. Low-pass whole genome sequencing was performed at the Peter MacCallum Cancer Centre using a previously published method 18 . DNA was eluted in 50 μL of low TE buffer/ nuclease-free water and quantified with QuBit high sensitivity (Thermo Fisher Scientific, Australia), and subsequently normalized to a volume of 50 μL in low TE buffer before sonication. DNA was sheared with sonication (Covaris S2 system) for 1 × 40 s, with the following parameters: duty cycles of 10, an intensity of five, and 200 cycles/burst.
Library preparation was performed with the NEBNext Ultra II DNA Library Preparation Kit (New England Biolabs, USA) as per manufacturer's instructions, with minor modifications. To each 50 μL sample of DNA 3 μL of NEBNext Ultra II End Prep Enzyme Mix and 7 μL of NEBNext Ultra II End Prep Reaction Buffer was added, followed by a short thermal cycling run to carry out end repair, 5′ phosphorylation and 3′ dA-tailing (deoxyadenosine monophosphate-tailing). The thermocycling conditions used were; 30 min at 20 °C for one cycle (end repair stage), and 30 min at 65 °C for one cycle (adenylation or a-tailing stage), with a final hold of 4 °C. 1.5 μM of NEBNext Adaptors were then ligated to DNA with 60 μL of NEBNext Ultra II Ligation Master Mix and 1 μL of NEBNext Ligation Enhancer (1 μL) in a 30minute incubation at 20 °C. Adaptor loops were severed by treatment with USER Enzyme which removed the uracil binding the two halves of the loop together. For this step, DNA was incubated with 3 μL of enzyme for 15 min at 37 °C.
Clean-up of the adaptor-ligated DNA was performed using Agencourt AMPure XP Reagent (Beckman Coulter, Australia) as per section 3.8.1 of the manufacturer's instructions ('Library preparation'). The purified adaptorligated DNA was then enriched with 11 cycles of PCR. The cycling conditions were; 30 s at 98 °C for one cycle (initial DNA denaturation), 10 s at 88 °C (DNA denaturation) and 75 s at 65 °C (primer annealing and DNA extension) for 11 cycles, and five minutes at 65 °C for one cycle (final extension phase) prior to a final hold at www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.