Patient characteristics and endoscopy screening outcomes

As depicted in the flow chart of Fig. 1A, the CRC surveillance study enrolled 87 LS patients with confirmed germline mutations in MMR genes and all underwent colonoscopy at baseline. Sixty-six of these patients returned at T1 and 51 at T2 for a second and third endoscopic examination. At baseline, 10 LgDs, five HgDs and three ADKs were detected by colonoscopy in 18 LS patients. At subsequent screening rounds, a lesion of each type was found in three patients at T1, while three LgDs and one HgD were detected at T2 in four patients. Except for one patient who was negative at T0 and T1 and developed an LgD at T2 (data not shown), all other lesions at T1 and T2 occurred in the 18 LS patients having a lesion at T0. None of the patients presented any other LS-related tumors during the observation period.

For the present study, all plasma samples available at the different time points from the 18 LS patients having lesions at T0 and 18 LS patients without any lesions detected during the surveillance period were considered. A total of 78 plasma samples were suitable for molecular analysis and 25 of these were collected in the presence of a colonoscopy-detected lesion: 14 LgDs, 8 HgDs, and 3 ADKs (Fig. 1B).

Overall, the median age of the patients was 56 years (interquartile range 41–62); 47% were female and 53% had germline mutations in MLH1, 36% in MSH2, and the remaining 11% in MSH6/PMS2. No differences in the distribution of sex, age, and germline mutations in MMR genes were observed between patients with and without endoscopically detected lesions at T0 (Table 1).

Table 1 Characteristics of Lynch syndrome (LS) patients with or without colorectal lesions at baseline (T0).Set-up of drop-off dPCR assays for analysis of bMSI

Plasma cfDNA was extracted from 1 to 2 mL of plasma and ranged from 0.9 to 10.1 ng/mL as measured by electrophoresis. Observing the baseline plasma samples, we did not find any significant differences in cfDNA amount related to the presence of lesions (p = 0.289), age (rs: 0.29; 95% CIBCa: -0.05; 0.56), sex (p = 0.778) and MMR gene mutations (p = 0.684; Fig. 2A–D). The molecular analysis started from a fixed volume (6.6 µL) of elution buffer containing 0.2 ng up to 1.5 ng of cfDNA. As a result, the total amount of DNA copies/µL, whether arising from WT or mutated DNA molecules, reflected the cfDNA content in plasma samples for any of the three dPCR assays adopted (Fig. 2E) with a Spearman correlation coefficient ranging from 0.59 to 0.70.

Fig. 2: Plasma cfDNA characterization at baseline.

Panels representing the distribution of cfDNA values measured by electrophoresis at the TapeStation (Agilent, Santa Clara, CA, USA) according to (A) lesion presence, (B) age, (C) sex, (D) MMR gene mutations. Each box indicates the 25th and 75th percentiles. The horizontal line inside the box indicates the median, the whiskers the extreme values measured. Individual data are represented by coloured dots according to lesion (A, C) or MMR gene mutations (D) details. B Scatter plot depicting the relationships between cfDNA (ng/mL) and age. Individual data are represented by coloured dots according to lesion details. E–G Scatter plot depicting the pairwise relationships between cfDNA (ng/mL) and each cfDNA copy assay (copies/μL) (E) rs: 0.59 (95% CIBCa: 0.32; 0.77) for Assay 1, (F) rs: 0.70 (95% CIBCa: 0.47; 0.83) for Assay 2 and (G) rs: 0.63 (95% CIBCa: 0.37; 0.79) for Assay 3.

To verify the overall performance of the assays, the LoB was first estimated using 25 WT additional samples and defined as the upper 95%CI of the mean false-positive MAF. The LoB of the five markers was 0.01% for BAT26, 0.05% for BAT25 and NR21, 0.03% for NR24, and 0.17% for Mono27 (Supplementary Fig. 1A). Afterwards, the assay reproducibility and the LoD were established by serial dilutions mixing WT and mutated DNA with expected MAFs ranging from 25% to 0.02% in triplicates. For each marker, all replicates exceeding the expected LoB had a positive signal with an rs between expected and observed MAFs higher than 0.9 (Supplementary Fig. 1B). Ultimately, the estimated LoD was 0.02% for BAT26, 0.05% for BAT25, 0.03% for NR24, 0.06% for NR21 and 0.20% for Mono27.

To further asses the reliability of the dPCR technology compared to the standard pentaplex PCR diagnostic tool, DNA extracted from two metachronous lesions occurred in a single patient diagnosed with LS was analyzed using both platforms (Supplementary Table 2). As expected, the pentaplex assay was less sensitive than dPCR in detecting MSI in tissue samples collected from precancerous and small cancerous lesions. Of note, by dPCR we were able to detected MSI in 3 and 2 loci even in the corresponding plasma sample, respectively. Representative images of the dPCR in tumor tissue and plasma samples collected from LS patients are shown in Supplementary Fig. 2.

Association of plasma microsatellite MAF values with lesion type and patient characteristics

The MAF values of the five microsatellite markers were first analyzed in relation to each other, to the total cfDNA copies (ng/mL) as well as to patient and tumor characteristics. In detail, the MAF values of BAT26, BAT25 and NR24 were significantly higher in samples from patients with endoscopically detected lesions (Fig. 3A), but did not differ significantly between lesion types (Fig. 3B). Moreover, no statistically significant associations were found (Fig. 3C, D) between the distribution of each microsatellite MAF with respect to sex and MMR gene mutations.

Fig. 3: Distribution of microsatellite MAF values according to clinico-pathological characteristics.

Panels representing the distribution of MAF values according to (A) lesion presence, (B) type of lesion, (C) sex, (D) MMR gene mutation at T0. Each box indicates the 25th and 75th percentiles. The horizontal line inside the box indicates the median, the whiskers the extreme values measured. Individual data are represented by colored dots according to lesion (A–C) or MMR gene mutations (D) details. The reported significant p-values refer to Wilcoxon test.

The correlation matrix for continuous variables represented in Fig. 4 shows that the BAT26 MAF values positively correlated with those of BAT25 (rs: 0.59; 95% CIBCa: 0.32; 0.77) and so were the Mono27 MAF values (rs: 0.37; 95% CIBCa: 0.04; 0.62). In addition, none of the MAF values correlated with age or with those of cfDNA copies detected by each assay.

Fig. 4: Correlogram summarizing relationships between continuous variables.

Pairwise relationships in terms of Spearman correlation coefficient (rs) between continuous variables such as patients’ age, microsatellite MAF and cfDNA copies (copies/μL) of each assay are reported. Colors indicate the direction of the correlation (green for positive and pink from negative correlations, respectively) and the size of the bubble the corresponding magnitude (upper triangular matrix form); single values indicate the corresponding rsp-values (lower triangular matrix form).

Figure 5 and Supplementary Figure 3 illustrates the time-trend pattern of the microsatellites on a dichotomic scale (i.e., negative vs positive), considering the 18 LS patients with lesions and the 18 without lesions, respectively. Specifically, by looking at Fig. 5, if NR24 was the marker that showed the highest frequency of positive signals at each time point, NR21 was the one with the least. Except for NR21, the remaining microsatellites displayed a higher frequency of positive rather than negative signals in samples collected in presence of lesions. Notably, whether a marker is positive or not can vary over time.

Fig. 5: Time-trend pattern of the MAF values in the 18 LS patients with lesions at T0.

A–E Sankey diagrams depicting the time trends of the five microsatellites on a dichotomized scale (i.e., negative vs positive). Colors indicates patient’s cluster profile over time according to lesion presence and markers positivity at baseline (T0).

Diagnostic value of blood MSI

To evaluate the ability of the liquid biopsy to discriminate between samples from patients with and without lesions, we estimated the AUC for both cfDNA and bMSI on their continuous scale. While the amount of cfDNA showed no discriminatory ability with an AUC of 0.61 (95% CI: 0.42; 0.80) (Fig. 6A), bMSI was able to discriminate patients according to the presence of lesions, with an AUC of 0.80 (95% CI: 0.66; 0.94) (Fig. 6B). This result was maintained after cross-validation (AUC-CV: 0.74; 95% CI: 0.58; 0.91).

Fig. 6: Evaluating the utility of blood MSI (bMSI) to discriminate lesion-bearing patients.

ROC curves of the (A) cell free DNA (cfDNA) and (B) blood microsatellite instability (bMSI) discriminating patients with and without colorectal lesion at baseline. C–F Spaghetti plots reporting the time-trend profile of the bMSI according to patient’s cluster profile over time according to lesion presence. Dots and dashed dark gray lines indicate the patients’ values and their trend over time.

For explorative purposes, we examined bMSI in terms of sensitivity and specificity, by selecting a cutoff that guaranteed at least 75% sensitivity, the specificity of bMSI was 72%. To assess the possible utility of the tool, we estimated its positive predictive value (PPV) and negative predictive value (NPV) by considering the T0 lesion prevalence registered in our overall series: this led to a PPV of 42% and NPV of 93%.

By stratifying patients according to presence of lesions, we looked at the time-trend pattern of bMSI. In LS patients with lesions at T0, but not at T1 and T2, a decrease of bMSI values compared to the baseline was observed in 8 out of 10 (80%) patients (Fig. 6C). Conversely, in the seven patients with metachronous lesions detected at T1 (Fig. 6D) or T2 (Fig. 6E), bMSI values showed a not well defined trend over time. Lastly, when LS patients who never developed lesions were considered, the range of bMSI values at T0, T1, and T2 remained comparable (Fig. 6F).