Real-life disease monitoring in follicular lymphoma patients using liquid biopsy ultra-deep sequencing and PET/CT

In the present study, we screened 84 Follicular Lymphoma patients for somatic mutations suitable as liquid biopsy MRD biomarkers using a targeted next-generation sequencing (NGS) panel. We found trackable mutations in 95% of the lymph node samples and 80% of the liquid biopsy baseline samples. Then, we used an ultra-deep sequencing approach with 2 · 10−4 sensitivity (LiqBio-MRD) to track those mutations on 151 follow-up liquid biopsy samples from 54 treated patients. Positive LiqBio-MRD at first-line therapy correlated with a higher risk of progression both at the interim evaluation (HRINT 11.0, 95% CI 2.10–57.7, p = 0.005) and at the end of treatment (HREOT, HR 19.1, 95% CI 4.10–89.4, p < 0.001). Similar results were observed by PET/CT Deauville score, with a median PFS of 19 months vs. NR (p < 0.001) at the interim and 13 months vs. NR (p < 0.001) at EOT. LiqBio-MRD and PET/CT combined identified the patients that progressed in less than two years with 88% sensitivity and 100% specificity. Our results demonstrate that LiqBio-MRD is a robust and non-invasive approach, complementary to metabolic imaging, for identifying FL patients at high risk of failure during the treatment and should be considered in future response-adapted clinical trials.


INTRODUCTION
Follicular lymphoma (FL) is the second most common non-Hodgkin lymphoma in developed countries [1]. It is genetically characterized by an upregulation of BCL2 in the progenitor B cell that transforms into a proliferating clone driven by t (14;18) translocations [2]. Nowadays, FL is considered an indolent disorder with a relatively favorable course. Long remissions are often achieved with modern day treatments, with median survival rates approaching 20 years [3,4]. However, 15-20% of patients are primary refractory or progress during the first two years after first-line therapy (POD24). These patients present a poor outcome, with 5-year overall survival (OS) probabilities between 38% and 50% [5,6].
The disease is characterized by a remitting, relapsing clinical course with progressive shortening of response duration after treatment. Moreover, high-grade transformation (HT) to aggressive lymphoma occurs in around 3% of the patients per year [7][8][9][10]. Global research efforts have been aimed at identifying patients with a high risk of progression and transformation to optimize the duration of treatment response and to ease suffering and morbidity. Many clinical, molecular, pathological, and imaging biomarkers have been described and used to stratify/group FL patients into several risk categories at diagnosis [11][12][13][14][15][16][17][18][19]. However, most of these tools remain inaccessible in daily practice and have not been adequately tested to select the best therapy.
For this reason, as the understanding of prognosis is crucial, the ultimate goal in a disease such as FL should be to develop tools and approaches to guide therapy. Current studies of risk-adapted therapy based on minimal residual disease (MRD) evaluation remain to be investigational in FL [20]. At the molecular level, several studies have shown that MRD assessment is predictive of outcome. Most of these studies have focused on evaluating the BCL2/IgH rearrangements, quantified by polymerase chain reaction (PCR) in bone marrow or peripheral blood. The levels of circulating cell-free DNA (cfDNA) fragments have also demonstrated predictive value [21,22]. However, unlike in diffuse large B cell lymphoma (DLBCL), in FL, there are no studies evaluating MRD based on tumoral cfDNA detection [23,24].
In contrast, PET/CT using the 5-point scale Deauville criteria (D5PS) is well established, regardless of the treatment used [25,26], to predict the outcome and attain complete metabolic response. Nevertheless, the use of PET/CT alone is hampered by its limited sensitivity and specificity, and the interpretation of the results is highly dependent on the evaluating radiologist [27,28]. Moreover, there is little information on interim PET/CT [26] and its combination with MRD for prognostic assessment [20,29], and there are no studies that have used liquid biopsy NGS methods. Therefore, in this study, we aim to analyze the response to therapy in FL patients using ultra-deep sequencing of cfDNA and the D5PS scale PET/CT to identify, early on, those patients who have a high risk of relapse in less than 24 months (POD24).

MATERIALS/SUBJECTS AND METHODS Patient cohort and study design
This study was designed as a prospective observational study. The cohort included 84 newly diagnosed, recurrent or transformed FL patients recruited from the routine clinical practice at the Hospital 12 de Octubre (H12O) in Madrid, Spain, and the Hospital Universitario de Toledo, Spain. Of the initial 84 patients, 58 were under first-line treatment, 15 received savage therapy and 11 did not receive treatment ( Supplementary Fig. S1). The median followup for the patients in first-line therapy was 20.5 months (rank 3-65.5), and 31 (rank 5-43.5) months for patients in savage treatment. One patient (FL5), suffered from a transformation 10 months after the first-line therapy started. After transformation, the patient was included as a transformed case (FL5t). Informed written consent of all patients was obtained according to the Declaration of Helsinki. The study inclusion criteria were; histological confirmation and the availability of enough biological material in sequential samples. Treatment was started according to the Groupe d'Etude des Lymphomes Folliculaires criteria [30], and imaging examinations were performed as ordinary clinical practice. Responsible physicians decided on the treatment regimen according to institutional and international guidelines. In all patients, DNA from lymph node biopsies and cfDNA was obtained before the treatment was started. Somatic mutations of these samples were selected as disease biomarkers for liquid biopsy MRD (LiqBio-MRD) analysis at follow-up time points. The following biological materials were analyzed: DNA from formalin-fixed paraffin-embedded (FFPE) lymph node biopsies (n = 75) and cfDNA (n = 44) before treatment start, and follow-up cfDNA samples during chemo-immunotherapy courses (n = 151).

DNA extraction
Lymph node DNA was extracted with a Qiamp gDNA FFPE kit (Qiagen, Hilden, Germany) using two to four sections from 5 to 10 microns, cut from the original paraffin block. Then, the genomic DNA (gDNA) was eluted in 35 μL ATE buffer and quantified using the Qubit BR kit (Thermo Fisher Scientific, Waltham, Massachusett, USA). For cfDNA extraction, 10 to 20 mL peripheral blood was collected in EDTA tubes and processed in less than four hours at the H12O. Samples from Toledo were collected in Roche Cell-Free DNA collection Tubes (Roche Diagnostic, Basel. Switzerland) and sent to the H12O, where plasma separation and cfDNA purification were centralized. There were no differences in cfDNA quantity or quality observed between EDTA and Streck collection tubes. The plasma was separated with two centrifugation steps at 1600 × g and 4500 × g, and stored at −80°C until further use. The purification of cfDNA was performed with a Qiamp Circulating Nucleic Acid kit (Qiagen) and quantified using a Qubit HS kit (Thermo Fisher Scientific). Fragment size and genomic DNA contamination were quantified using a Bioanalyzer 2100 fragment analysis system (Agilent, Santa Clara, California. USA).

Baseline Genotyping and LiqBio-MRD Biomarker Selection
The lymph node gDNA and plasma cfDNA baseline samples were screened for mutations with a short-length Ampliseq Custom Panel (Thermo-Fisher). The panel, established as a routinary diagnosis tool at the H12O, was designed to cover all coding regions of 56 lymphoma-specific genes in the FFPE samples (Supplementary Table S1). Samples were sequenced with an average coverage of 2,150x on an Ion S5 System platform (Life Technologies, Thermo Fisher Scientific). Variant annotation was performed using the default annotate variants single sample workflow from the Ion Reporter software (version 5.18.2.0). Mutations were called when presented more than nine mutated reads and a Variant Allele Frequency (VAF) above 2 · 10 −2 (2%). In the case of FFPE samples, deamination-related base changes were reduced by filtering out C > T/G > A changes with a VAF below 2 · 10 −1 (20%) and a transformed p-value greater than −2, unless previously described as a somatic aberration in FL (COSMIC database). To select best MRD biomarkers for each patient, we first excluded all potential SNPs, germinal variants and missense variants of unknown significance (VUS) using COSMIC, CLinVar, and dbSNP databases. Then we categorize the remaining somatic mutations based on the number of cancer patients affected and prioritized the ones more frequently involved in lymphoma until we have up to 5-10 markers per patient when possible. (Supplementary Table S2).

LiqBio-MRD Methodology and Bioinformatic Pipeline
A median of 16.14 ng per mL of plasma (range 2.63-489 ng/mL) was obtained, being the mean amount of cfDNA from the initial 10-20 mL of peripheral blood (PB) of 73.6 ng (range 15.75-1935 ng). All samples with a gDNA/cfDNA ratio greater than one, calculated using the bioanalyzer electropherogram, were excluded. The minimum quantity of cfDNA targeted for sequencing was 15 ng. Considering that a genome equivalent (GE) has a mass of 3 pg [31], 15 ng is enough to screen 5,000GE and therefore achieve a sensitivity of 2 · 10 −4 VAF (0.02%).
An amplicon-multiplexed mini-panel was defined for every patient to detect the selected MRD biomarkers identified at diagnosis. The mini-panel included molecular-tagged primer pairs (6-mer tags) to amplify every selected mutation in three biological replicates defined as P1, P2, and P3. (Supplementary Table S3 and Fig. S2). Then, the three independent tagged post-PCR products were combined in a single tube to continue with sample preparation. Final libraries were sequenced on the Ion S5 System platform (Life Technologies, Thermo Fisher Scientific Inc.) with an estimated depth of 500,000× per amplicon, as previously described [32,33]. Despite the dilution curves of 53 genetic variants indicated a potential VAF sensibility below 10 −4 ( Supplementary Fig. S3), all potential MRD biomarkers were screened with the same pipeline in triplicates of three gDNA samples obtained from healthy control donors. This permitted to define the LOD for every MRD biomarker (mean VAF plus three standard deviations) and exclude all the potential MRD markers that presented a LOD above 10 −4 ( Supplementary Fig. S4).
The FASTQ files produced after sequencing were automatically demultiplexed to separate the reads from the different amplicons and triplicates. To reduce the false positive rate, a strict bioinformatic pipeline was programmed in Python and R to eliminate low-quality reads. The specific wild-type and mutated sequences were generated for each genetic position. These sequences, obtained from the corresponding demultiplexed output file, cover the affected locus with 15 bp upstream and downstream. Only the reads that perfectly match these sequences were considered to calculate the VAF for each triplicate. The noise effects arising from PCR and sequencing were controlled by identifying and removing triplicates that overpassed the mean VAF plus one standard deviation (SD). Finally, the corrected VAF was compared with the LOD calculated for each mutation independently, using three triplicates of three healthy donors as previously explained. The LOD was computed as the mean VAF in control samples plus three times the SD. Every MRD biomarker with a corrected mean VAF below the LOD was automatically eliminated. (Supplementary Table S3) The final LiqBio-MRD value was defined by the mutation with the highest VAF at the sampling time-point, as shown in Supplementary Fig. S5.

PET/CT imaging
The PET/CT scans were performed with a General Electric Discovery MI (GEDMI) Scanner or a Siemens Biograph 6 Scanner. PET/CT and CT images were acquired in the same session after injection of 2.5-3 MBq/kg 18 F-FDG (fluorodeoxyglucose) for the GEDMI Scanner and 4-5 MBq/kg 18 F-FDG for the Siemens scanner. All follow-ups were performed in the GEDMI scanner. CT scans obtained with a low-dose protocol were used for attenuation correction of the PET/CT images. Interim and EOT 18 FDG-PET/CT scans were visually assessed according to the D5PS, with 18 FDG uptake of any residual lesion, using mediastinal blood pool and liver uptake as reference settings. PET/CT was considered to be positive when the Deauville's score was four or five, and Deauville's scores from one to three were classified as PET/CT negative.
In first-line therapy, PET/CT was performed before starting the treatment, after four cycles (n = 40), and at the end of treatment (EOT, n = 54). After finishing induction, patients were closely monitored with a physical examination and routine laboratory tests. A new scan was performed only when new symptoms or laboratory changes were detected. PET/CT was generally performed at mid-induction (n = 7) and EOT (n = 14) for patients treated in other lines. The exact time points are listed in Supplementary Table S4.

Statistical analysis
The Kruskal-Wallis test was used to determine statistically significant differences in the obtained samples' MRD values among the PET/CT-related categories. Then, a post hoc Dunn test was conducted to identify the statistically significant pairwise comparisons and the corresponding p-values. The tests were both performed using Python, the prior with the Python package SciPy (version 1.6.2) and the latter with the Python package scikitposthocs (version 0.6.7). The Pearson correlation coefficient was used to assess the linear relationship between the different variables under study. Univariable Cox proportional hazards regression models and Kaplan-Meier survival analysis were performed to test statistical associations between genetic and imaging findings and survival outcomes. Statistical calculations were conducted using SPSS 22.0 (IBM SPSS Inc, Chicago). P-values of ≤0.05 were considered to be significant.

RESULTS
Patients' characteristics and predictive features A total of 84 FL patients were included in the study (Supplementary Fig. S1). The median age was 63 years (35-90 years), and 58% of the patients were female. Eleven cases did not require treatment, 85% of the patients presented low histological grade [1,2], and 79% had advanced (III-IV) Ann-Arbor stage (Table 1). Regarding treated cases (n = 73), 58 patients received first-line treatment (39 RCHOP, 10 R-bendamustine, 4 rituximab monotherapy, and 5 radiotherapy). The other 15 patients were treated with salvage therapy after a previous relapse (Supplementary  Table S4). More important prognostic indexes in FL were analyzed. High-risk patients defined by FLIPI, FLIPI2, m7-FLIPI, and PRIMA PI did not show shorter PFS. Bulky disease, the presence of symptoms B, and a lymphocyte-to-monocyte ratio (LMR) of <2.5 were the only variables associated with a higher risk of relapse (p < 0.05). After a median follow-up of 22 months (rank 3-61.5), 19 patients relapsed after a median of 19 months (16 patients with grade I, II, or 3A and 3 patients transformed). Eight cases died after a median of 29 months (five patients had a low histological grade, and three patients were transformed). The causes of death were lymphoma in four cases (three transformed), infection (n = 2), and secondary neoplasia (n = 1). Regarding the subgroup of patients treated in first-line therapy (n = 58) 15 patients relapsed after a median of 12.6 months (3.9-44.9).
Baseline genotyping on cfDNA complements lymph node screening in follicular lymphoma Baseline genotyping with the targeted NGS panel was performed in 75 lymph node samples and 44 cfDNA plasma samples. In the lymph node samples, 510 mutations were detected with an average of 6.8 somatic mutations per patient (range 0-31) and a mean VAF of 0.31 (range 0.026-1.0). In the cfDNA plasma samples, 144 mutations were detected (average 3.3, range 0-11) with a mean VAF of 0.22 (range 0.025-0.857). Only 4 of 75 (5%) lymph node samples did not present any alteration suitable for MRD monitoring. This number increased to 20% (9 of 44) when only baseline cfDNA was considered. However, somatic mutations were detected in the cfDNA fraction for six of the eight patients without lymph node samples available.
As previously described [12], the most frequently mutated genes were KMT2D, CREBBP, BCL2, TNFRSF14, and EZH2 (Fig. 1A). Although the samples from transformed patients had a similar genetic profile, an increase of TP53 mutations was observed in this subcohort (3/10, 33% vs. 7/73, 9%, Supplementary Fig. S7). Within the 36 cases with available paired lymph node and plasma samples, 88 somatic mutations were identified in both fractions, 33 somatic mutations were only detectable in liquid biopsy, and 160 somatic mutations were only in the lymph node (Supplementary Table S2, Supplementary Fig. S6).

Clinical impact of disease monitoring by LiqBio-MRD
The dynamics of the baseline mutations were analyzed on 151 cfDNA follow-up samples from 54 of 73 patients that received treatment (Fig. 1B). Additionally, sequential samples of eleven untreated "watch and wait" patients were screened. On average, 3.6 somatic mutations per patient (range 1-12) were selected as MRD biomarkers. Nine mutations were excluded from further analyses as they presented a LOD above 1·10 −4 . These mutations mainly affected insertions or deletions of one base (Supplementary Fig. S4). On the other hand the elimination of outlier triplicates permitted to identify and correct nine false positives follow-up samples. To compare PET/CT and liquid biopsy results, we defined four groups based on the PET/CT data available when the liquid biopsy sample was collected. The first group (complete response, CR) included follow-ups with a Deauville score of 1, 2, or 3 in nodal or extranodal sites with or without a residual mass. The Non-progressive disease group included partial response with a Deauville score 4, or 5 with reduced uptake compared with baseline and residual mass(es). The Progressive disease includes a Deauville score 5 in any lesion with an increase in the intensity of FDG uptake from baseline. The Non-treatment group had samples before treatment started or from untreated patients. (Fig. 1C) The LiqBio-MRD values were significantly lower in the CR and Nonprogressive disease groups compared to the progressive group (p < 0.001 and p = 0.019, respectively). However, of 53 samples included in the CR group, four were LiqBio-MRD positive. Of note, all samples in "watch and wait" group (eight samples from eight patients) were positive by the LiqBio-MRD test.
The 54 treated patients with sequential samples available presented an average of three liquid biopsy samples (rank 1-8) after treatment started. Of them, 44 patients were under first-line treatment. To calculate the clinical impact of the LiqBio-MRD test in first-line patients, we defined three different timeframe groups: Early follow-up (n = 23) included cfDNA samples from cycles I and II; the interim group (n = 23) included samples obtained in cycles III and IV; the final or EOT group included 32 cfDNA samples obtained in cycle VI or the first sample available under maintenance (Fig. 2, left). Positive LiqBio-MRD values in the early group did not increase the risk of progression (HR EARLY 2.3, 95%, CI 0.44-11.8, p = 0.310). This tendency changed at interim monitoring (HR INT 11.0, 95% CI 2.10-57.7, p = 0.005). The differences between LiqBio-MRD-positive and negative cases were even more pronounced at EOT (HR EOT , HR 19.1, 95% CI 4.10-89.4, p < 0.001), (Fig. 2, right). The results of the entire cohort, including relapsed patients, are shown in Supplementary Fig. S8.
Interim monitoring by PET/CT of previously untreated patients predicts progression PET/CT D5PS tests also showed prognostic value at interim evaluation. PET/CT positive cases had a mPFS INT of 19 months vs. NR (p < 0.001, Fig. 3A). Comparable results were observed at EOT (mPFS EOT PET/CT 13 months vs. NR, p < 0.001, Fig. 3B), and when relapsed patients were included ( Supplementary Fig. S8). Of note, the distribution of cases according to interim and EOT TPs for PET/ CT and LiqBio-MRD presented a concordance of 76% (Kappa = 0.401).

The Combination of LiqBio-MRD and PET/CT Identifies POD24 Patients
Next, we studied the 41 first-line patients with data available on both PET/CT and LiqBio-MRD. The last TP (interim or EOT) with data for both tests was considered for this analysis. Twenty-six patients were negative by both techniques, seven were positive, and eight presented discordant results. Considering only concordant results (n = 33), the combination of both tests showed a sensitivity (SE) of 88% and a specificity (SP) of 100%, with a PPV of 100% and NPV of 96%. Strikingly, all positive patients by both tests had a 2-year PFS below 24 months (mPFS of 7 months for +/+ vs. NA for −/− cases, p < 0.001, Fig. 4). Moreover, the only case (FL25) that progressed with a negative result by both tests, was positive in a sequential cfDNA sample obtained at maintenance, five months before progression. Regarding the eight cases with discordant results (20%), six patients were incorrectly classified by PET/CT, and only two patients were incorrectly classified by LiqBio-MRD (Fig. 4). The results including relapsed patients are shown in Supplementary Fig. S9.
Other approaches for disease monitoring were also tested. Flow cytometry data was only available for eleven patients with bone marrow infiltration. Two patients were positive at follow-up and   The Non-progressive disease (excluding CR) group included partial response cases with Deauville score of 4, or 5 with reduced uptake compared with baseline and residual mass(es). The progressive disease includes a Deauville score 4 or 5* in any lesion with an increase in the intensity of FDG uptake from baseline. treatment. An additional positive liquid biopsy was received three months later, and PET/CT was performed, confirming progression but with a low tumor burden (Fig. 5A). The patient continued maintenance therapy, achieving a complete response a few months later. The opposite dynamics were observed in patient FL5. This patient did not respond to RCHOP (DS5) and received R-bendamustine as second-line therapy. The interim PET/CT showed a poor response (DS5), and the biopsy confirmed the transformation to high-grade lymphoma. After failure of rescue treatment with R-GEMOX-dexamethasone and R-polatuzumab bendamustine, the patient received radiotherapy where a reduction in the main clone was observed (Fig. 5B). Of note, the clone detected in cfDNA disappeared under first-line therapy.
Patient FL30 presented two mutations detected only in lymph node and two more only detected in cfDNA. In follow-up samples, the four mutations were undetectable. In the PET/CT evaluation, a residue was observed in the scans, complicating the interpretation of the imaging results (Fig. 5C). This patient is still in CR after two years of follow-up.
Although a solid biopsy was unavailable for patient FL31, six somatic mutations were detected in the baseline liquid biopsy sample. In follow-up cfDNA samples, despite an initial reduction during the first two cycles, a rapid increase in the disease burden after cycle III was observed. The patient progressed in cycle VI and died under rescue therapy only nine months from the start of the treatment (Fig. 5D). The dynamics of all the treated patients with follow-up samples available are shown in Supplementary Fig. S9.

DISCUSSION
This prospective study evaluates, for the first time, the usefulness of liquid biopsy MRD by ultra-deep sequencing in combination with D5PS PET/CT to identify, early on, those FL patients with a high risk of relapse in less than 24 months (POD24). Our approach is based on the use of somatic mutations as disease biomarkers, as we previously described for acute myeloid leukemia [32]. First, we screened baseline lymph node and plasma samples to identify patient-specific biomarkers. The genetic profile of our cohort mimics the one previously described by Pastore et al. [12] but with an expected increase of TP53 alterations in transformed cases [34] ( Supplementary Fig. S1). Although the custom DNA panel was initially designed for FFPE samples, the small amplicon size permitted the detection of lymphoma-specific mutations in baseline liquid biopsies (Fig. 1A). In our study, limited by the follow-up and the small and heterogeneous number of subjects, the only clinically relevant prognostic factor for PFS in newly diagnosed patients (n = 84) was the low lymphocyte-to-monocyte ratio [19]. We did not find prognostic differences by applying IPI, FLIPI, FLIPI2, m7-FLIPI, or PRIMA IPI. However, somatic mutations suitable for LiqBio-MRD monitoring were found in 95% of patients with lymph node samples and 80% of patients with liquid biopsy   samples available. Our results indicate that follicular lymphoma gene driver mutations are detectable in liquid biopsy. Still, more sensitive approaches are needed. In any case, the number of patients with potential MRD biomarkers considerably improves the applicability of MRD assessment compared to other described techniques, such as PCR of the IGH/BCL2 translocation [20,29]. Although PCR-positive is predictive of lower PFS, a considerable number of patients are t(14;18) negative (60%). Moreover, FL is a predominantly nodal disease, and the absence of t(14;18) in bone marrow does not adequately reflect the response status [35]. Further, t(14;18) can also be found in healthy individuals at low level.
In this study, LiqBio-MRD was evaluated in the plasma of 54 patients that received treatment (43 first-line therapy). The analysis, performed in 151 follow-up cfDNA samples allowed us to define patient-specific disease dynamics (Fig. 1B, Supplementary Fig. S7) and their correlation with the clinical outcome (Fig. 1C). To the best of our knowledge, there are no studies in FL evaluating early MRD assessed by liquid biopsy. Conversely, there are already some studies in DLBCL [24,36,37] and Hodgkin lymphoma [38] where it is known that levels often change rapidly after the initiation of therapy. Following this hypothesis in FL, we performed an early LiqBio-MRD evaluation (Cycle 2) on 23 patients. Although a trend to shorter PFS was observed in MRDpositive cases, 6 of 13 cases eventually became negative at later time points (Fig. 2), suggesting that FL presents a different dynamic than DLBCL [39] and treatment need longer to cleanse the tumor.
Since the Lugano classification, the criteria to determine the quality of treatment response rely on the D5PS PET/CT evaluation [39]. Several studies, including large retrospective subanalyses of randomized trials have shown that PET/CT negativity correlates solidly with PFS [25,40]. Nevertheless, the use of PET/CT alone is hampered by its limited sensitivity and specificity and the interpretation of the results being highly dependent on the evaluating radiologist [27,41]. Consistent with previous reports [23], from the 64 patients with available PET/CT scans at WOT (50 patients in first-line therapy), 70% of the patients reached a CR.
On the other hand, the LiqBio-MRD test showed an extraordinary capacity to identify patients at risk of progression after only four cycles of RCHOP (Fig. 3A). Comparable results were observed by interim PET/CT (Fig. 3B). This result suggests that LiqBio-MRD or PET/CT interim evaluation should be considered in future clinical trial settings. Regarding EOT evaluation, both tests segregated high-risk patients (Fig. 3C, D). Although similar results have already been shown in FL patients using PCR-based MRD techniques, these approaches were hampered by a lower applicability [20,29,35,42,43].
PET/CT has only been used in combination with PCR in a small exploratory analysis and MRD refined the predictive power of PET/CT [29]. However, no combination of PET/CT and MRD strategies has been reported using NGS liquid biopsy techniques. In our study, the combination of both methods identified 100% of the POD24 patients. As shown in Fig. 4, patients with concordant results (n = 28) were almost perfectly segregated. These results remained after excluding transformed cases ( Supplementary  Fig. S8). The only double negative patient (FL25, Fig. 5A) that progressed presented an increase in tumor burden detectable by LiqBio-MRD five months before progression in a sample obtained within maintenance. This suggests that sequential monitoring with a minimally invasive test such as LiqBio-MRD may be essential to identify POD24 cases and anticipate patients' relapses. Another interesting case, FL5, illustrates how LiqBio-MRD may have other possible applications (Fig. 5B). In this chemo-refractory patient, radiotherapy had an abscopal effect, confirmed by the rapid descend of ctDNA not otherwise explicated [44].
Our study has several limitations, including the limited number of patients, the heterogenous treatment administered, and the absence of available tests in all the time points. However, several factors permitted the development of a LiqBio-MRD test with extraordinary performance. First, the MRD amplicons were designed shorter than 120 bp and all MRD samples screened presented at least 15 ng of cfDNA to guaranty the amplification of enough tumor cfDNA molecules. More importantly, the use of triplicates and the definition of LOD in healthy control donors permitted to identify and correct the false positive values induced by PCR and sequencing errors or the variant intrinsic noise due to the genetic context. Although it seems mandatory to perform larger studies to confirm this preliminary data, our results demonstrate, for the first time, that NGS-based MRD quantification is feasible in liquid biopsies from FL. The achievement of negative ctDNA after treatment and in an interim analysis enhances prognostic information on the patients´outcomes, both in firstline therapy and at relapse. PET/CT and LiqBio-MRD can synergistically contribute to predicting progression and POD24 with high sensitivity and specificity. Additionally, this test better reflects intra-patient tumor heterogeneity [45,46] and could be used to detect drug resistance, high-risk transformation and guide and monitor treatment.
In conclusion, LiqBio-MRD monitoring in FL represents a promising option, complementary to metabolic imaging, to identify patients at high risk of failure early on during treatment and is a useful approach to response-adapted precision therapy to be considered in clinical trials.

DATA AVAILABILITY
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.