Magnesium chloride hexahydrate (MgCl₂⋅6H₂O), sodium chloride (NaCl), acetone, and phosphate buffer saline (PBS) were purchased from Sinopharm Chemical Reagent Co., Ltd. (China). Exonuclease III was provided by Thermo Fisher Scientific Co., Ltd. (United States). Tris borate EDTA (TBE) buffer (5×), tris EDTA (TE) buffer (1×), sodium dodecyl sulfate (SDS, 10%, w/v) buffer, loading dye buffer solutions (6×), agarose, and saline-sodium citrate (SSC, 20×) buffer solutions were obtained from Solarbio Science & Technology Co., Ltd. (China). GelRed Neuclic Acid Gel Stain was purchased from Biotium, Inc. (United States). DNA ladder was provided by Sigma-Aldrich Co., Ltd. (China). QIAamp DNA Blood Mini Kit was purchased from Qiagen Co., Ltd. (Germany). All chemicals in our experiments were of analytical grade and used without further purification. Aqueous solutions were prepared using deionized water (≥18 MΩ, Milli-Q, Millipore). The SERS substrates were provided by Nanova Biomaterials Inc. (United States). High-performance liquid chromatography (HPLC)-purified oligonucleotides were provided by Sangon Biotechnology Co., Ltd. (China).

The Raman microscope equipped with a He–Ne laser (633 nm, 20 mW) and a 100× objective (NA: 0.9) was used to detect the Au NP@Si substrates (HR800, Horiba Jobin Yvon, France). SERS data were collected under a 100× visible objective with a 633 nm laser radiation, with an acquisition time of 10 s and an accumulation count of one time. The scan of the wave number typically ranged from 1,000 to 1,800 cm^–1. The Raman spectral data were analyzed by the LabSpec5.6 Software. For each sample, 50 random spots on the substrate were tested for SERS signal. All SERS spectra have been elaborated by removing the baseline, with an example shown in Supplementary Figure 1. The correlation of the major Raman bands to the chemical bonds of Cy5 is summarized in Table 1. A DNA thermostat hybridization oven (HL-2000 HybriLLinker, UVP, United States) was used for the incubation experiments involving DNA. The substrate surface was characterized by contact angle measurement instrument (DataPhysics OCA). The gel electrophoresis of various sample mixtures was performed on a horizontal electrophoresis tank instrument (VE-186), followed by image acquisition with a Gel Imaging System (Tanon-2500).

The hybridization and the Exo III-assisted cleavage of different DNA samples were evaluated by agarose gel electrophoresis. Agarose (2%, w/w) was prepared in the buffer of 1× TBE and mixed with 3 μL of GelRed stain. Specifically, 10 μL of the DNA sample was mixed with 2 μL of 6× loading dye buffer solution. Different DNA samples were tested by gel electrophoresis under a voltage of 110 V for 25 min. The DNA sample mixtures included P, T, the hybridization of P and T (P + T), the mixture of the P, T, and Exo III (P + T + Exo III), and the residual DNA (R).

The SERS signals of ctDNA testing experiments were harvested from the Au NP@Si substrate. In the first step, the Au NP@Si substrate was modified with the capture DNA (C) sequence. The substrate was precleaned with acetone, rinsed with DI water for three times, and dried in an oven at 200°C for 10 min. The pretreated substrate was incubated overnight with 10 μL of C (10 μM) and 400 μL of the PBS solution at 37°C for bioconjugation through Au–S bonds. The sodium chloride (1 M) solution was used in the aging step for immobilization of C (15 μL, three times with an interval of 20 min). The substrate was incubated overnight in the DNA thermostat hybridization oven at 37°C, followed by sequential washing with PBS and DI water three times. The reaction of cycled enzymatic DNA cleavage/amplification was carried out by mixing 10 μL of P (10 μM), 10 μL of ctDNA in the testing concentrations, 3 μL of Exo III (60U), and 17 μL of the enzyme-reaction buffer at 37°C for 2 h. Afterward, the mixture was incubated at 75°C for 10 min in order to inactivate the Exo III enzyme. The product mixture was obtained after the reaction was diluted with 360 μL of PBS buffer and then incubated with the as-prepared substrate at 37°C for 2 h. The substrate was washed with 2× SSC solution containing 0.1% (w/v) SDS for 5 min, washed with DI water several times, and dried before the Raman signal measurement.

The feasibility of our assays was verified by a series of control groups in the tests, including (A) C; (B) C + P; (C) C + P + M; (D) C + M + Exo III; (E) C + P + W + Exo III; and (F) C + P + M + Exo III. All control groups were performed in the identical conditions.

A series of ctDNA with gradient concentrations were prepared to determine the LOD of our assay. About 10 μL of P (10 μM) and 10 μL of ctDNA with different concentrations were mixed for incubation at 37°C for 2 h. The final concentrations of mutant ctDNA were in the range of 10–100 nM.

The blood samples from healthy human donators were collected and treated according to the standard protocol. All human care and experimental procedures were conducted in compliance with relevant laws and the guidelines approved by the institutional committees for Human/Animal Experiments of the School of Basic Medical Sciences of Soochow University. An informed consent was acquired from all human subjects. The serum was obtained from the whole blood and mixed with H3.3-mutated ctDNA of various concentrations. Then, H3.3-mutated ctDNA was extracted from the serum according to the extraction procedure of QIAmp DNA Blood Mini Kit. Briefly, 100 μL of serum mixed with H3.3-mutated ctDNA was taken into a 1.5 mL centrifuge tube, then 10 μL of protease K and 100 μL of AL buffer were added, and later incubated at 56°C for 10 min. After adding 50 μL of ethanol, the solution was centrifuged for 1 min at 8,000 rpm. About 500 μL of AW1 and 500 μL of AW2 were added to the separation tube and centrifuged for 1 min at 8,000 rpm. After 100 μL of AE eluent was added, the solution was balanced for 1 min at room temperature and centrifuged for 1 min at 14,000 rpm to collect the eluent.

Statistical analyses were assessed with one-way ANOVA by using Prism 5.0 software (GraphPad Prism, United States). All intensities at the specified SERS peaks are presented as mean ± SD.

The assay is featured with integration of the cycled enzymatic DNA cleavage/amplification and SERS for the sensitive detection of ctDNA (Figure 1A). An oligonucleotide probe is designed to be folded in a stem–loop hairpin structure, tagged with cyanine dye Cy5 at the 5’ end. The hairpin structure is stable, minimizing the undesired side hybridization with W sequence. In contrast, the stem–loop structure of the oligo probe can undergo changes to form a new double helix by hybridizing its 3’ end with the target sequence of the mutant ctDNA. Consequently, the protruding 3’ end in the new double helix can specifically be recognized by Exo III enzyme for the cleavage into nucleotides in a stepwise manner. Importantly, after completion of the cleavage process, the target sequence of ctDNA can be released into the solution and recycled for the next round of enzymatic DNA cleavage of the oligo probe. In this way, the residual DNA sequence generated by the digestion of the oligo probe can be accumulated to a great number by this cycled reaction. Hybridization of the amplified R and C preimmobilized on Au NPs@Si will bring the Cy5 tag close to the substrate, producing intensive SERS signals with a high efficiency. Therefore, a trace amount of ctDNAs can initiate the cycled generation of residual DNA sequences for amplified SERS detection.

Scanning electronic microscopy (SEM) (Figure 1B) suggested that the gold nanoparticles were uniform in size (60 nm) on the Au NP@Si substrate with a good distribution. The reproducibility in generating SERS signals was evaluated by scanning the substrate for spectral acquisition of up to 50 randomly selected spots. As shown in Figures 1C,D, there were minimal spot-to-spot variations of the intensities in the SERS spectra, with the coefficient variation values at the characteristic peaks of Cy5 at 1,309 cm^–1 (CV < 9.88%), 1,366 cm^–1 (CV < 9.73%), and 1,509 cm^–1 (CV < 13.70%) and enhancement factor (EF) > 1.9 × 10⁶. The details of calculating the EF are listed in the Supplementary Information. We tested the SERS spectra at different laser powers by triturating the laser filters (Supplementary Figure 2A) and demonstrated the robustness of spectral data acquisition by switching laser power grades (high or medium) in five cycles (Supplementary Figure 2B). Therefore, the Au NP@Si substrate can serve as a reliable platform for SERS measurements of a trace amount of molecules.

The DNA sequences, including P, T, C, R, and W, are summarized in Table 1. Based on the literature reports (Schwartzentruber et al., 2012), the H3.3-mutated ctDNA from DIPG is different from W in the healthy samples by a single nucleotide mutation (A to T), as specified in Table 1. A theoretic analysis using the online software (NUPACK) was performed to determine the minimum free energy (MFE) secondary structures of these oligos and hybridization efficiencies between each other (Figure 2). P tended to keep hairpin structure in the solution with a relatively favorable free energy (−40.35 kJ mol^–1) for the secondary structure (Figure 2A). The addition of H3.3-mutated ctDNA (T) to the solution would hybridize a segment of P competitively, thus changing its initial hairpin structure to form a partly hybridized helix with a 3’ blunt end. This hybridization was driven by the decrease of free energy (ΔG = −61.20 kJ mol^–1). In addition, we compared the MFE changes for hybridization between W and P (ΔG = −47.09 kJ mol^–1), which suggested that P would prefer to hybridize H3.3-mutated ctDNA, even if W had only one base change. We also calculated the occupancy rates of the various oligos by NUPACK, in order to investigate their hybridization efficiency in silica at 37 ^oC. The calculation indicated that P was inclined to be hybridized with H3.3-mutated ctDNA with a high efficiency (nearly 47%); besides hybridization of 3% of W with P, the remaining W maintains its own structure (47%), leaving only 3% of the target DNA unhybridized in the solution (Figure 2J above, pie chart). We further compared the MFE changes for the hybridization scenarios between C and R (C∩R, ΔG = −57.77 kJ mol^–1) or between C and P (C∩P, ΔG = −36.67 kJ mol^–1). The results suggested that C would hybridize R, even when unreacted P was present in the solution. When the three kinds of DNA were mixed equally in mass, P would maintain its stable hairpin structure (48%), and most of the hybridization happened between C and R (45%, Figure 2J down, pie chart). This suggested that the presence of P (leftover after the cleavage reaction) would not influence the hybridization between R and C. Therefore, the oligo sequence design would promote the performance of our assay by minimizing the side reactions/hybridization of W∩P or C∩P as the sources of false-positive signals.

Agarose gel electrophoresis was employed to test DNA hybridization and Exo III enzyme–assisted DNA cleavage. As shown in Figure 3A, the bands in Lane 1, Lane 2, and Lane 5 were corresponding to P, H3.3-mutated ctDNA (T), and R in the gel electrophoresis, respectively. The hybridized product of T∩P was present as the dark band in Lane 3, with the light bands attributed to a small amount of the free oligos of P. After the addition of Exo III in Lane 4, the hybridized product of T∩P disappeared, suggesting an efficient DNA cleavage assisted by the Exo III enzyme. The residual DNA was observed after the enzymatic digestion, which highly contributed to the degradation of the hybridized product of T∩P by Exo III. GelRed usually stains double-stranded nucleic acids than single-stranded nucleic acids. It was interpreted that the band of T or R was weak because T tended to form only a short segment of double helix structure, while R was mostly single-stranded.

The feasibility of our assay was also validated with SERS measurements of a series of DNA oligo mixtures with or without Exo III enzyme, including five negative control groups and one positive experimental group (Figure 3B). There were no detectable SERS signals when only C was present on the substrate (curve A in Figure 3B); the group of C+P (curve B) and C+P+T (curve C) observed the same trend. These experiments suggested that C would not hybridize to P, which was consistent to our design and the theoretical calculation. We did not observe SERS signals for C+T+Exo III (curve D) or C+P+W+Exo III (curve E). In contrast, there was a significantly intensive SERS signal on the Au NPs@Si substrate in the positive experiment containing C, P, T, and Exo III (Curve F in Figure 3B). The signal intensity of the positive group (F) was nearly six to seven times higher than the other negative groups of B, C, D, and E at the characteristic peaks of Cy5 tag, such as 1,309, 1,366 cm, or 1,509 cm^–1 (Figure 3C). The SERS signal comparison between E and F verified that the assay was highly selective to distinguish a single-nucleotide difference between W and M sequences. In addition, we performed the reaction (C+P+target DNA+Exo III) in two separate substrates in parallel to evaluate the substrate-to-substrate variation, which suggested reproducible SERS signal acquisition (Supplementary Figure 3).

The specificity of the assay was further examined by mixing the W (1 pM) and H3.3-mutated ctDNA oligos in gradient concentrations at different ratios (1 pM, 0.1 pM, 0.01 pM, 1 fM, and 0 fM). As shown in Figure 4A, when the concentrations of H3.3-mutated ctDNA were reduced from 1 pM to 0 fM, the SERS signal intensities gradually decreased with the gradient changes, while maintaining the spectral patterns of Cy5 tag. Quantitative analysis of the characteristic peak intensities at 1,306 cm, 1,366 cm, and 1,509 cm^–1 also demonstrated this trend (Figure 4B), suggesting that the assay can specifically distinguish the samples when the H3.3-mutated ctDNA oligos were diluted by 2 or 3 orders of magnitude with W.

The LOD of the assay was evaluated by titrating the H3.3-mutated ctDNA oligo concentrations in the PBS buffer and the human serum, separately. As shown in Figures 5A,B, the SERS signal intensities of the characteristic peak at 1,366 cm^–1 decreased when the ctDNA oligo concentration was diluted from 10 to 100 nM in PBS buffer. There was a good linear relationship (R² = 0.980) between the SERS signal intensities and ctDNA oligo concentrations, with an LOD of 7.9 fM (1.0 pg mL^–1) (signal-to-noise ratio ≥ 3).

Furthermore, we performed a pilot experiment to test the assay in human serum samples. H3.3-mutated ctDNA oligos were spiked into the human serum samples from healthy donors with a concentration gradient from 10 to 100 nM. The DNA samples were then extracted from the mixture by referring to a standard DNA extraction procedure, in order to remove serum proteins or other contamination sources. As shown in Figures 5C,D, we observed a strong correlation (linearity R² = 0.991) between the SERS signal intensities and the spiked concentrations of the H3.3-mutated DNA oligos. The LOD was estimated to be 9.1 fM (1.2 pg mL^–1) according the criterion of the signal-to-noise ratio ≥3. The experiments suggested that the SERS/enzyme amplification technique allowed for sensitive detection of M in complicated samples, thus offering a useful tool for potential clinical translation.

Diffuse intrinsic pontine gliomas is a brainstem tumor of malignancy in childhood for which median survival is below 1 year (Hoffman et al., 2018). More than 90% of patients with DIPG are estimated to have the point mutation in H3.3 (65% of tumors) or H3.1 (25% of tumors), while the rest are estimated to have histone 3 wild-type tumors (Vanan and Eisenstat, 2015). Among those mutations, the averaged survival span of H3.3 mutation is the shortest (10.4 months) (Cordero et al., 2017), thus being in an urgent need of developing sensitive diagnostic tools. By using the spiked blood samples, we have demonstrated an ultra-sensitive detection of the target DNA sequence for H3.3 mutation of DIPG with an LOD of 9.1 fM (1.2 pg ml^–1), nearly 10–100 folds lower than the typical ctDNA concentration range (0.01–0.1 ng ml^–1) of the disease. This study has not tested the blood samples from patients with DIPG yet due to a limitation of the clinical resources. It is necessary to further validate this technique in carefully grouped patient blood samples with references to the clinical outcomes in the future.

In our technique, the design of P is very flexible for the introduction of any target DNA sequence, thus allowing for an easy translation to sensitive detection of the ctDNA of other types of cancers without changing the enzyme. For instance, researchers have identified a variety of clinically useful mutations in the ctDNA samples, including G protein subunit alpha Q (GNAQ) Q209L mutation (626A > T) in metastatic uveal melanoma (Madic et al., 2012), p110a catalytic subunit of the class 1A PI3K (PIK3CA) E545K mutation (1633G > A) in breast cancer (Board et al., 2010), or epidermal growth factor receptor (EGFR) T790M mutation (2369C > T) in non-small-cell (Normanno et al., 2017). The applications of the technique on the sensitive detection of these mutations are promising in facilitating the early diagnosis of the diseases and treatment with personalized medicine. The key for successful development of a new assay by the technique is careful optimization of the competitive hybridization segment of P for minimal side reactions of hybridizing W or C.