Integrated microfluidic platform for programmable multi-window DNA fractionation and in situ recovery

Abstract

Nucleic acid size selection underpins applications from sequencing to genome engineering, yet current methods impose trade-offs among separation breadth, recovery fidelity, and operational throughput. To address these trade-offs, we engineered a compact on-chip microfluidic field-inversion gel electrophoresis (MFIGE) platform that integrates programmable deoxyribonucleic acid (DNA) fractionation with in situ dual-membrane DNA recovery, which avoids manual gel excision in a closed, low-shear, and automation-ready format. MFIGE delivers multi-window fractionation beyond 140 kbp, with total recovery rate up to 57.9% and operates robustly across different sample types. In nanopore sequencing validation, MFIGE reshaped the read-length distribution and substantially increased long-read output. It generated more than 2900 reads exceeding 100 kbp, 976-fold higher than long-fragment accumulator (LA) reagent-based fractionation and 47.2-fold higher than the unfractionated control. It also raised the N50 to 33.58 kbp, 1.5-fold higher than LA and 3.4-fold higher than unfractionated. By programming the field strength, we concentrated over 94% of >50 kbp fragments in target wells, enabling precise, high-fidelity capture. Together, these results position MFIGE as a practical front end for long-read sequencing library preparation and other applications demanding precise DNA sizing.

---

1. Introduction

Size-based fractionation and recovery of nucleic acids is a core pre-processing step that determines data quality in modern sequencing workflows.^1,2 Current sequencing platforms exhibit a pronounced polarization in their requirements for nucleic acid fragment lengths. Third-generation sequencing technologies (e.g., Oxford Nanopore, CycloneSEQ, and PacBio) rely on ultra-long nucleic acid fragments at the kilobase-scale to resolve complex genomic structures such as chromosomal translocations and tandem repeat sequences, imposing critical demands on separation systems to span orders of magnitude in dynamic range.^3–6 Next-generation systems (e.g., Illumina NovaSeq 6000) demand precise control of 150–300 bp short fragments to maintain cluster generation efficiency and base-calling accuracy.^7–9 Consequently, rigorous size selection during library preparation is essential since excessively short fragments cause signal attenuation in nanopore sequencing, and overly long fragments induce phasing errors in sequencing-by-synthesis (SBS).^10,11 It is therefore a key technological challenge to establish a nucleic-acid fractionation system that combines a broad separation range with low-bias recovery.

Gel electrophoresis remains the mainstream technology for nucleic-acid fractionation.^12,13 Within the framework of gel electrophoresis, field-inversion gel electrophoresis (FIGE) operates through periodic electric field reorientation to achieve DNA separation in agarose gels.^14–16 Small fragments respond rapidly to field reorientation, whereas long fragments experience delayed migration through gel pores, enabling fractionation across ∼200 bp to 10 Mbp. Despite its entrenched role in conventional genomic analysis, FIGE's inherent limitations restrict its utility in precision sequencing workflows. In practice, in a standard FIGE-based fractionation-recovery, DNA recovery requires manual UV-guided gel excision,¹⁷ which physically decouples the critical steps of fragment separation and recovery and exposes long DNA molecules to damaging mechanical shear forces, causing fragmentation and loss of integrity. Moreover, poorly integrated workflows with high manual burden limit FIGE's ability to meet modern sequencing requirements for precision and throughput. In addition, substantial reagent consumption further drives up per-sample processing costs. In parallel, numerous other size-selection technologies are in common use, such as bead-based selection, ultrafiltration, and centrifugation.^18–22 However, they still suffer from narrow separation windows and high sample loss. These constraints highlight the critical need for integrated technologies capable of simultaneously delivering a broad fractionation range, high recovery fidelity, low-bias recovery, and low reagent consumption.

Microfluidics offers a route to address these bottlenecks.^23–25 As an integrated platform for reagent handling, sample selection, reaction, and detection, microfluidics chips compress processing units to square-centimeter footprints, feature high functional integration, and minimize reagent consumption.^26–28 Modular chip design enables seamless, in-channel coupling of fractionation and recovery in a closed microenvironment, avoiding damage-prone manual gel excision. Moreover, low-Reynolds-number laminar flow mitigates shear associated with macroscopic fluid handling, providing a favorable hydrodynamic environment for preserving nucleic-acid integrity.^29–31 Therefore, integrating FIGE into a microfluidic architecture preserves its broad separation range while enabling closed-channel, low-shear, in situ recovery that reduces loss, improves integrity, and supports automation.

This work presents a compact microfluidic field-inversion gel electrophoresis (MFIGE) platform that integrates programmable fractionation with in situ dual-membrane recovery in a closed, low-shear architecture. The design avoids manual UV-guided gel excision and supports parallel, on-demand selection across a broad size range. We establish predictive “design rules” that link agarose concentration, voltage, and pulse parameters to size boundaries and yield, thereby enabling either stringent depletion of short fragments or co-recovery of long- and intermediate-length fragments. In validation experiments with Escherichia coli (E. coli) and human HG002, MFIGE directed >94% of fragments >50 kbp into target wells and achieved a high total recovery rate up to 57.9%. Applied to nanopore sequencing, MFIGE increased the N50 to 33.58 kbp and yielded more than 2900 reads exceeding 100 kbp, 976-fold higher than a commercial long-fragment enrichment reagent and 47.2-fold higher than unfractionated. Together, these results position the MFIGE platform as an automation-ready solution for robust long-read library preparation and other workflows requiring precise DNA sizing.

2. Experiment section

Chip fabrication

The integrated microfluidics chip was fabricated from polymethyl methacrylate (PMMA) and consisted of three primary components: a central chamber, a recovery module, and a cover plate (Fig. 1a). The microfluidic chip was manufactured using 3D printing technology. The fabrication process began with creating a rectangular main chamber that measured 80 mm (L) × 60 mm (W) × 7 mm (D). Electrode ports were integrated into the sidewalls of the chamber. Within the main chamber, a central fractionation lane with dimensions of 43 mm (L) × 9 mm (W) × 7 mm (D) was printed, with rubber stopper ports positioned at both ends. Adjacent to the lane's right side, a recovery module containing five collection wells was fabricated, each with a volume of 90 μL. The recovery module was constructed by bonding a microfilter membrane (pore size 0.45 μm) at the front section and an ultrafiltration membrane (molecular weight cutoff 20 kDa) at the rear section, using low viscosity biocompatible double-sided tape. The open-top design of the recovery wells facilitated sample aspiration after fractionation. Corresponding access openings for rubber ports, sample loading, and the recovery module were precisely machined on the cover plate to align with components in the main chamber. Then, the assembly was secured with screws, and platinum wire electrodes were installed through the electrode ports. After assembly, the PDMS precursor (Sylgard® 184, Dow Corning) and curing agent were uniformly mixed at a 10 : 1 ratio and applied around each electrode port to prevent liquid leakage. The device was finally baked at 70 °C for 30 minutes in a vacuum oven to cure the PDMS fully.

Fig. 1 Schematic diagram of the MFIGE chip. (a) Exploded view of chip assembly. (b) Gel infusion modeling in the fractionation lane. (c) Sample loading for fractionation and recovery.

Certified Megabase agarose (C₂₄H₃₈O₁₉, Bio-Rad) served as the gel matrix material. The procedure began with preparing 0.5× TBE buffer (UltraPure™ TBE, Thermo Fisher Scientific) containing 45 mmol L⁻¹ Tris-borate and 1 mmol L⁻¹ EDTA. A predetermined quantity of agarose powder was added to the TBE buffer solution, which was microwaved to a boil. Prior to gel loading, rubber stoppers were inserted to seal both ends of the fractionation lane. Subsequently, 4 mL of the prepared agarose mixture was injected through the top loading port. Immediately after injection, a rubber stopper was inserted into the loading port to create a mold cavity. The chip was then allowed to solidify at room temperature for 30 minutes. Upon complete gelation, the stoppers were extracted. The cover plate was then sealed with medical adhesive tape, and the chamber was filled with 0.5× TBE buffer (Fig. 1b).

Experimental setup and operation

The system comprises a programmable power-supply module and an analysis module. The programmable power supply (PV7800 Pro, ITECH) controls field strength, inversion frequency, run time, step count, and time increments. The analysis module includes a pulsed-field gel electrophoresis system (PFGE, Bio-Rad) and a fluorometer (Qubit 3, Thermo Fisher Scientific). PFGE and gel imaging were used to quantify fragment sizes recovered in each well, while Qubit quantified DNA mass. Ambient conditions were maintained at 20 °C and 50% RH.

All DNA was purchased from QIAGEN Genomics. Up to 4 μg of DNA (≤70 μL) was loaded through the loading port (Fig. 1c). Platinum wires aligned with the fractionation axis were connected to a power supply, driving DNA migration into the gel under electrophoretic force for separation. To compensate for water loss due to electrolysis, deionized water was added to the reservoir approximately every 30 min to restore the initial liquid level. After fractionation, the circuit was deactivated, and the orthogonal recovery circuit was energized. Size-fractionated DNA migrated perpendicular to the fractionation axis toward the recovery module. At the gel-microfilter interface, DNA exited the gel matrix, traversed the microfilter into the recovery well, and was immobilized by the ultrafiltration membrane. The liquid-phase confinement enabled subsequent collection. At the end of the recovery step, a 3 s reverse-polarity pulse was applied and the chip was rested briefly to promote desorption of DNA from the ultrafiltration membrane into the solution. After on-chip fractionation and dual-membrane recovery, a small aliquot of the DNA solution collected from each open recovery well was withdrawn and loaded into a conventional agarose PFGE gel for size analysis.

Long-fragment accumulator for DNA fractionation

The long-fragment accumulator reagent (LA, CycloneSEQ) enriches high-molecular-weight DNA through fragment sorting, depleting short fragments to improve single-molecule read lengths. Sample DNA was adjusted to 50–100 ng μL⁻¹ with TE or nuclease-free water. A 60–110 μL aliquot of DNA was transferred to a low-bind 1.5 mL tube, mixed with 1.3× volume of LA buffer by gentle pipetting, and incubated at room temperature for 10 min. Samples were centrifuged at 10 000 rcf for 30 min to pellet DNA. The supernatant was discarded. The pellet was washed twice with freshly prepared 70% ethanol. For each wash, ethanol was added to cover the pellet, the tube was centrifuged at 10 000 rcf for 3 min, and the supernatant was removed. Residual liquid was removed by centrifugation at 10 000 rcf for 1 min, followed by careful aspiration. The pellet was then air-dried with the cap open until no visible liquid remained at room temperature. The pellet was resuspended in 50–100 μL 1× TE, briefly spun down, and incubated at 37 °C for 30 min to dissolve DNA fully. Qubit measured the sample's concentration, and pulsed-field gel electrophoresis was run to check the fragment size distribution after LA-based fractionation.

Library preparation and sequencing

Post-fractionation, DNA from recovery wells was purified and used for library preparation. Target-length DNA was transferred to a 1 mL centrifuge tube with an equal volume of magnetic bead suspension. The tube was placed on a magnetic rack after static incubation at 37 °C for 15 minutes. Supernatant was removed, and beads were washed twice with 80% ethanol. DNA was eluted in TE to 60–120 ng μL⁻¹. Then 1 μg DNA was brought to 44 μL with nuclease-free water, followed by end-repair with 3 μL enzyme and 12 μL buffer at 20 °C for 10 min, then 65 °C for 10 min. Equal-volume magnetic bead purification was performed again. DNA was eluted with 62 μL NF water. Adapter ligation was carried out by adding 5 μL adapters, 25 μL ligation buffer, and 10 μL ligase at 25 °C for 30 min. A final bead purification was performed at a 1 : 0.4 sample-to-bead ratio (v/v) with two 80% ethanol washes. DNA was eluted in 12 μL EB and incubated at 37 °C for 15 min before collection.

Nanopore sequencing (WT-02, CycloneSEQ) was used to evaluate the impact of fractionation on read lengths. After sequencing-chip quality control, 600 ng of library DNA was mixed with 6 μL of incubation buffer, 300 μL of working buffer, and nuclease-free water. The mixture was then loaded onto the qualified sequencing-chip for a 24-h run.

3. Results and discussion

The principle operation of the MFIGE platform

After loading the sample, a periodically inverted pulsed electric field was first applied along the fractionation lane. Under the combined effects of the electric field and the gel matrix, short nucleic acid chains with higher chain flexibility can rapidly adjust conformation with field reversals and continue migrating along the field direction. In contrast, long DNA reorients more slowly and can transiently hook around gel fibers during polarity switches, producing a lag behind field reversals. The difference between short and long chains in relaxation dynamics leads to marked migration-velocity differentiation under repeated field inversions: short chains maintain efficient migration, whereas long chains experience reduced net displacement due to the lag effect, forming a gradient of migration speeds among short, long, and megabase chains. Banding thus appears along the fractionation lane in a length-dependent manner (Fig. 2a). The recovery module is sealed upstream with a microfilter membrane and downstream with an ultrafiltration membrane, forming a dual-membrane confinement region. After fractionation, a DC field is applied perpendicular to the fractionation lane to drive the separated nucleic acids from the gel phase into the microchambers of the recovery module. An applied electric field generates an electrophoretic force that drives nucleic acids through the gel, enabling them to exit the gel matrix and traverse the microfilter membrane into the recovery solution. Concurrently, the ultrafiltration membrane retains all fragments. Nucleic acids are thus confined between the two membranes to complete recovery (Fig. 2b).

Fig. 2 The principle operation of the MFIGE platform. Schematic illustration of the (a) fractionation and (b) recovery process. (c) Stepwise variation of forward and reverse pulse durations per cycle and the modulation of applied voltage waveform per cycle. (d) The temperature variation of the chip during the fractionation. (e) Electrophoresis gel images of nucleic acids pre- and post-fractionation. (f) Size boundaries of nucleic acid per recovery well. (g) Recovered nucleic acid mass per well.

E. coli genomic DNA was used as the test sample to verify on-chip fractionation and recovery. The fractionation lane was filled with 0.45% (w/v) agarose. During the fractionation stage, a field-inversion pulsed mode was applied at 45 V, with an initial forward time of 1 s and an initial reverse time of 0.33 s per cycle. Forward and reverse time increased by 0.05 s and 0.03 s per step, respectively, over 10 steps per cycle, with a total runtime of 5 hours. For recovery post-fractionation, the system was switched to a unidirectional DC mode at 50 V for 1.5 hours. It is worth noting that, although the dual membranes remained functional after a run, each set was used only once per experiment to avoid cross-contamination between samples. Fig. 2c show the evolution of forward/reverse timing per cycle and the associated drive waveforms. Infrared thermography during runs indicated only a slight, spatially uniform increase in chip temperature and no pronounced local hot spots (Fig. 2d).

Under the inverted pulsed field, the distributions before and after size selection are shown in Fig. 2e. The input exhibited a continuous 8–200 kbp distribution. After fractionation, DNA in the five recovery wells showed a stepwise gradient: from left to right, the recovered fragment length decreased. As shown in Fig. 2f, the minimum length decreased from 40.0 kbp at well 1 to 14.2 kbp at well 5, while the maximum length peaked at 140.7 kbp in well 2. Because ultralong DNA exhibits length-dependent reorientation lag, its migration profile broadens and spans the lane regions aligned with recovery wells 1 and 2. Consequently, both wells 1 and well 2 recovered ultra-long DNA characterized by main bands exceeding 50 kbp, with well 2 showing the maximum intra-well length difference of 101.6 kbp. In contrast, medium-long chains experienced a relatively faster reorientation and remain relatively compact in the lane. The difference was reflected in sharply reduced size heterogeneity in wells 3–5. The maximum–minimum intra-well difference declined sharply in wells-3 to wells-5, from 30.0 kbp to 10.9 kbp. The combined mass from wells 1–2 was 1221.5 ng, accounting for 62.8% of the total, consistent with the intrinsic abundance peak of the sample in the 50–100 kbp interval (Fig. 2g).

The MFIGE platform achieves three parallel capabilities: effective removal of fragments <10 kbp, retention of fragments >50 kbp, and multi-window size fractionation. Concurrently, the system demonstrates remarkably low dependence on nucleic-acid type. Using an LA reagent to treat full-length E. coli nucleic acids can effectively remove <10 kbp fragments (Fig. 3a) but cannot capture multiple target windows in a single operation. Using identical electrical parameters (0.45% agarose, pulsed-field/DC composite field), the multi-window capability was further verified with the human sample HG002. Recovered DNA across wells displayed stable multi-window fractionation (Fig. 3b), confirming MFIGE's cross-species applicability.

Fig. 3 MFIGE demonstrates broad sample compatibility and delivers superior read-length performance. (a) Electrophoresis gel image of LA-based size fractionation. (b) Electrophoresis gel image of MFIGE-fractionated human HG002 DNA. Reads proportion distribution (c), bases distribution (d), and read-length metrics (e) across unfractionated, LA-fractionated, and MFIGE-fractionated samples.

Moreover, fragment-length-based fractionation can improve nanopore sequencing data quality in terms of read length. We quantified time-resolved read-length composition over a 24 h sequencing run. The unfractionated library contained about 25% of reads shorter than 3 kbp, with very few exceeding 10 kbp. LA fractionation reduced the fragments below 3 kbp while increasing the representation of 10–50 kbp fragments. MFIGE produced the strongest shift toward longer reads while matching the short-read depletion achieved by LA. At 12 h, reads shorter than 3 kbp were 20.5% in the unfractionated library, 6.3% with LA-fractionation, and 5.68% with MFIGE; reads longer than 50 kbp were 0.35%, 1.28%, and 7.12%, respectively, giving about 5.6-fold over LA-fractionation and about 20-fold over unfractionated under MFIGE (Fig. 3c). Bases distribution yields followed the same trend. MFIGE yielded 278.3 Mb below 3 kbp and 2448.4 Mb above 50 kbp, versus 1762.5 Mb and 394.1 Mb for the unfractionated library and 300.8 Mb and 364.6 Mb for LA-fractionation. Moreover, only samples processed by MFIGE yielded >100 kbp reads. Relative to LA-fractionation, MFIGE increased bases above 50 kbp by about 6.7-fold and reduced bases below 3 kbp by about 7.5% (Fig. 3d). It is worth noting that the <10 kbp reads observed during sequencing were not due to insufficient short-fragment removal in the fractionation. They arose because long molecules are readily sheared during library preparation. The read-length metrics across the unfractionated sample and two fractionation strategies were further compared. Furthermore, we summarized sequencing performance across the three fractionation conditions. As shown in Fig. 3e, the MFIGE-fractionated sample performed best on all metrics. Its N50 reached 33.58 kbp, 1.5× that of LA-fractionation and 3.4× that of the unfractionated control. Notably, reads >100 kbp numbered more than 2900, 976× higher than with LA-fractionation and 47.2× higher than without fractionation. The improvement fundamentally originates from the selective removal of short molecules during fractionation, which reallocates sequencing channel resources toward longer fragments and consequently elevates the effective yield of long reads. Therefore, this integrated fractionation approach alleviates limitations on read-length distribution imposed by fragment-length heterogeneity in long-read sequencing.

Gel concentration-tuned MFIGE fractionation

Gel concentration determines the pore-size distribution of the sieving matrix and thereby controls size-dependent DNA migration.^32,33 To elucidate the impact of agarose concentration on MFIGE fractionation-recovery, we systematically varied the gel from 0.30% to 0.75% (w/v) while holding all other parameters constant. As shown in Fig. 4a, DNA in each recovery well displayed a stepwise distribution at all concentrations, with fragment length decreasing monotonically along the net migration direction (left to right). Gel concentration, via its control of matrix pore-size distribution, strongly influences size-dependent migration.³⁴ At 0.30% agarose gel concentration, elevated migration rates resulted in extended migration paths, accumulating the longest fragments up to 144 kbp in distal well 3. Higher gel concentrations increase migration resistance, thereby reducing per-unit-time migration distance. When the concentration increased to 0.75%, reduced migration rates significantly shortened migration distances, shifting the long fragment recovery position leftward, and shifting the recovery position of the longest fragments leftward to well 1 (Fig. 4b and c). Crucially, while maintaining the intrinsic “short-fast-ahead, long-slow-behind” electrophoretic ordering, the compressed migration distances shift the recovery positions upstream, closer to the origin.

Fig. 4 Adjusting the gel concentration shifts fractionation boundaries and modulates size-resolved recovery yields. (a) Electrophoresis gels of fractionated nucleic acids under varied gel concentration. (b) Fragment length distribution of recovered nucleic acids per recovery well. (c) Size boundaries of nucleic acids. (d) Mass of recovered nucleic acids per recovery well. (e) Total recovery rate and mass distribution across fragment-length windows (<20 kbp, 20–50 kbp, >50 kbp).

Gel concentration–driven changes in migration rate, migration distance, and recovery position directly determine the mass distribution of recovered fragments. Fig. 4d shows that at 0.30%, well 1 recovered almost nothing, whereas at 0.75% its recovery increased sharply to 840 ng. Higher gel concentrations improve capture efficiency by slowing the migration of short fragments, preventing them from passing beyond the recovery zone by the end of electrophoresis. Recovery zone refers to the segment of the fractionation lane in the agarose gel that is spatially aligned with the recovery module, and whose contents are collected into the recovery wells during orthogonal recovery. To quantify parallel recovery across fragment-length windows, we computed the recovered mass in each window and the overall recovery yield. The total recovery rate (α) is defined as the recovered mass sum from all wells (mass_all-wells) divided by the initial input sample mass (mass_input).

Agarose concentration exerted a pronounced effect on size-resolved recovery. The recovery of fragments <20 kbp was highly sensitive to concentration changes, whereas fragments >50 kbp remained relatively stable. At 0.30%, fragments <20 kbp were utterly lost. The 20–50 kbp and >50 kbp windows yielded 365 ng and 895 ng, respectively, giving α = 31.5%. At 0.75%, the recovered masses were 213 ng (<20 kbp), 665 ng (20–50 kbp), and 1102 ng (>50 kbp), and α increased to 47.9%. Notably, as agarose concentration increased, the peak recovery of the longest fragments shifted from well 3 to well 1, yet the total mass of >50 kbp DNA remained nearly constant, indicating robust separation of this window across the concentration gradient.

Field-strength-modulation-controlled MFIGE fractionation

Electric field strength directly governs the electrophoretic driving force (F = qE), thereby modulating the distribution of recovered fragment sizes.^35,36 Using 0.45% agarose and keeping other parameters constant, we investigated the fractionation-recovery under the voltage from 35 V to 65 V. Recovered DNA consistently exhibited a stepwise size distribution along the electrophoretic axis. Optimal resolution occurred at 35 V, where the bounds of adjacent wells were tightly spaced. For example, at 35 V the adjacent wells exhibited only a narrow inter-well overlap, indicating high resolution. Well 3 extended down to 14.6 kbp, whereas well 4 reached up to 18.1 kbp, yielding an overlap of 3.5 kbp. Increasing the voltage to 55 V broadened the overlap to 18.1 kbp: the lower boundary of well 3 shifted to 60.6 kbp and the upper boundary of well 4 to 78.7 kbp, consistent with reduced resolution (Fig. 5a). Concurrently, increasing the voltage accelerated migration and shifted recoveries toward downstream wells (Fig. 5b). Quantitatively, analysis of the >50 kbp mass proportion per well at 35 V showed that fragments concentrated predominantly in well 1 at 91.1%, while adjacent well 2 contained 0%. When the voltage increased to 55 V, the enrichment peak shifted to well 3 at 94.0%, with well 4 becoming a secondary enrichment site at 44.5% (Fig. 5b).

Fig. 5 Tuning the electric field strength improves enrichment of long fragments. (a) Electrophoresis gels of fractionated nucleic acids under varied voltages. (b) Heatmap of the proportion of recovered mass >50 kbp by recovery well. (c) Total recovery rate and mass distribution across fragment-length windows (<20 kbp, 20–50 kbp, >50 kbp).

Precise voltage modulation enables size-selective DNA recovery. Given migration velocity's direct proportionality to electric field strength (v ∝ V) and inverse proportionality to molecular weight (v ∝ 1/M), Fig. 5c demonstrates that increasing voltage from 35 V to 55 V caused <20 kbp fragments to exit the recovery zone, preventing their recovery and reducing yield from 141 ng to 0 ng. Conversely, >50 kbp fragments showed enhanced mobility at higher voltage, improving their recovery from 787.8 ng to 1004.9 ng. The 20–50 kbp fragments peaked at 559.3 ng at 45 V, where migration velocity aligned optimally with the fixed fractionation duration. Raising the voltage to 65 V caused over-migration, with portions of the 20–50 kbp and >50 kbp populations moving beyond the recovery zone and escaping capture. The recovered mass declined to 436.1 ng for >50 kbp and 98.7 ng for 20–50 kbp, while <20 kbp remained fully excluded (0 ng). Together, these results show that maximizing overall recovery requires tuning the voltage so that size-dependent bands terminate within the recovery zone at the end of electrophoresis. Consistent with this, total recovery peaked at 35.3% at 45 V, which best balanced separation and capture.

Pulse-timing control

After clarifying the effects of gel concentration and voltage, the timing parameters of pulsed-field electrophoresis were investigated as key variables for optimizing fractionation and recovery. We focused on the combined effect of the number of timing steps per field-inversion cycle (N_step) and the increment in reverse-pulse duration between successive timing steps (Δb), which together determine precise control of the net forward migration time (t_net). The net forward migration time is defined by eqn (2) as the net forward-minus-reverse electrophoresis time accumulated over the entire run.where t_f and t_r represent the forward and reverse electrophoresis durations, respectively, and t_total denotes the total electrophoresis time. With the t_total fixed, programmable adjustments of N_step and Δb change the net displacement of DNA through the gel, directly governing recovery efficiency. Smaller Δb or fewer N_step prolong t_net, whereas increasing either shortens t_net and shifts fragment endpoints toward upstream wells.

When t_net is long, >20 kbp fragments migrate to more downstream positions, and short fragments pass beyond the recovery zone and are lost. When t_net is short (Δb/N_step large), >20 kbp fragments remain upstream and short fragments are retained within the recovery zone, increasing their yield (Fig. 6a and b). For example, when Δb was 10 ms, the t_net was calculated to 2.65 h, and fragments <20 kbp were essentially removed. The recovered longest and shortest fragments were ∼159.4 kbp and 25.0 kbp, respectively. When Δb increased to 70 ms, t_net decreased to 1.47 h, and the recovered longest and shortest fragments were ∼140 kbp and 8.5 kbp (Fig. 6c). The results indicate that adjusting Δb directly affects migration sensitivity. When Δb increased from 10 ms to 70 ms, the <20 kbp lower size boundary moved by 66%, versus only 12% at the upper boundary for ultralong fragments, indicating much greater sensitivity in the short-fragment regime. The sensitivity was further supported by the size-range coefficient (η_span) and by the recovery rate of >50 kbp. The size-range coefficient is defined as the difference between the average length in the well with the longest fragments (L̄_longest) and the average length in the well with the shortest fragments (L̄_shortest), normalized by the overall mean (L̄_total).

Fig. 6 Temporal parameter tuning for shifting recovery windows, and enhancing size-resolved. Electrophoresis gels of fractionated nucleic acids under (a) varied reverse-step time increment and (b) steps per cycle. (c) Recovered size range boundaries, (d) size range coefficient and the recovery rate of fragments >50 kbp, and (e) total recovery rate and mass distribution across fragment-length windows (<20 kbp, 20–50 kbp, >50 kbp) under varied reverse-step time increment. (f) Recovered size range boundaries, (g) size range coefficient and the recovery rate of fragment >50 kbp, and (h) total recovery rate and mass distribution across fragment-length windows (<20 kbp, 20–50 kbp, >50 kbp) under varied steps per cycle.

The size-range coefficient quantifies separation breadth, with higher values indicating wider fragment length ranges in fractionation. As shown in Fig. 6d, the η_span rose from 0.85 to 2.13 as Δb increased from 10 ms to 70 ms, with >50 kbp fragments recovery improving from 33.2% to 41.8%. In parallel, total recovery increased from 28.7% to 57.9%, driven by a surge in <20 kbp fragments recovery rising from 0 ng to 228.6 ng and a substantial gain in 20–50 kbp increasing from 333.0 ng to 804.4 ng, while >50 kbp rose from 817.0 ng to 1029 ng.

Increasing N_step also shortens t_net and shifts endpoints upstream, lowering the recovered lower-length boundary while leaving the upper boundary for ultra-long fragments relatively unchanged due to their intrinsically slow migration (Fig. 6f). Consequently, the overall span increased continuously with N_step, elevating η_span from 0.76 to 2.29. Importantly, >50 kbp fragment recovery showed a threshold-like, non-monotonic response, peaking at 51.7% under N_step = 40 before declining at either extreme of N_step values (Fig. 6g). Consequently, fragments above 20 kbp achieved maximal recovery at N_step = 40, with total recovery rate reaching a maximum at 52.2% (Fig. 6h). This identifies a robust operating optimum that globally maximizes recovery by calibrated control of migration distance.

Together, these results motivate a two-knob workflow for timing control. For stringent depletion of fragments <20 kbp, use a smaller Δb with low-to-mid N_step to drive shorter species beyond the recovery zone while enriching >50 kbp in target wells. For multi-window co-recovery, use a larger Δb with N_step ≈ 40 to maximize total recovery and expand η_span. Nonetheless, membrane adsorption may still contribute to residual loss, and further optimization of membrane surface chemistry and the desorption protocol will be important in future work, particularly for more complex sample matrices.

4. Conclusion

We developed an integrated MFIGE platform that seamlessly co-locates DNA fractionation with in situ recovery. Systematic tuning of gel concentration, voltage, and pulse timing delivers tunable multi-window fractionation spanning ∼8 kbp to >140 kbp. Quantitatively, voltage programming concentrated >94% of fragments >50 kbp into target wells. Pulse-timing optimization yielded a robust optimum near N_step = 40, with 52.2% total recovery and 51.7% recovery for fragments >50 kbp. Adjusting the reverse-step increment Δb from 10 ms to 70 ms increased total recovery from 28.7% to 57.9% and expanded η_span from 0.85 to 2.13 while maintaining long-fragment integrity. Size-selection validation using both human (HG002) and E. coli DNA confirmed broad applicability. Critically, sequencing validation confirmed functional impact, showing that MFIGE reshaped the read-length distribution, increased the N50 to 33.58 kbp, and produced more than 2900 reads exceeding 100 kbp, substantially surpassing both a commercial long-fragment reagent and unfractionated controls. By bridging FIGE-class fractionation resolution and closed-system recovery, this work establishes an automation-ready paradigm for long-read sequencing library preparation and other genomic applications demanding precise size selection. Building on this foundation, future work could focus on coupling this platform with automated and intelligent control systems to further reduce manual intervention and enable closed-loop operation.

Author contributions

Dongliang Li: conceptualization, writing – original draft, methodology, investigation, data curation. Chunlei Yang: methodology, investigation, data curation, visualization. Leiyang Xu: methodology, investigation, data curation, formal analysis. Tao Zeng: investigation, data curation. Xiao Shi: investigation, data curation. Quanxin Yun: writing – review & editing, supervision, funding acquisition. Yuliang Dong: supervision, writing – review & editing, conceptualization, funding acquisition. Yuning Zhang: writing – review & editing, supervision, resources, conceptualization, funding acquisition.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence this work.

Data availability

All data needed to support the conclusions in the paper are presented in the manuscript. All data were tested twice to ensure the usability of the data. Additional data related to this paper may be requested from the corresponding author upon request.

Acknowledgements

The work was supported by the National Key R&D Program of China (No. 2024YFC3406300), the Shenzhen Science and Technology Program (No. KQTD20221101093603011).

References

1. T. Ojala, A.-E. Häkkinen, E. Kankuri and M. Kankainen, Trends Genet., 2023, 39, 686–702.
2. D. J. Taylor, J. M. Eizenga, Q. Li, A. Das, K. M. Jenike, E. E. Kenny, K. H. Miga, J. Monlong, R. C. McCoy, B. Paten and M. C. Schatz, Annu. Rev. Genomics Hum. Genet., 2024, 25, 77–104.
3. J. Y. Zhang, Y. N. Zhang, L. L. Wang, F. Guo, Q. X. Yun, T. Zeng, X. Yan, L. Yu, L. Cheng, W. Wu, X. Shi, J. Y. Chen, Y. H. Sun, J. N. Yang, R. R. Guo, X. D. Zhang, L. E. Kong, Z. A. Wang, J. L. Yao, Y. S. Tan, L. X. Shi, Z. T. Zhao, Z. W. Feng, X. P. Yu, C. Li, W. Zhan, Y. L. Ren, F. Yang, Z. J. Liu, G. N. Fan, W. L. Zhong, D. C. Li, L. He, Y. W. Qi, M. Zhang, Y. N. Zhu, H. Chi, Z. Y. Zhao, Z. F. Wei, Z. Q. Song, Y. M. Ju, R. J. Guo, L. Xiao, X. M. Lin, L. Chen, C. T. Yang, Q. Y. Li, O. Wang, X. Jin, M. Ni, W. W. Zhang, L. Q. Liu, Y. Gu, J. Wang, Y. X. Li, X. Xu and Y. L. Dong, bioRxiv, 2024, preprint,  DOI:10.1101/2024.08.19.608720.
4. P. E. Warburton and R. P. Sebra, Annu. Rev. Genomics Hum. Genet., 2023, 24, 109–132.
5. S. Ardui, A. Ameur, J. R. Vermeesch and M. S. Hestand, Nucleic Acids Res., 2018, 46, 2159–2168.
6. S. L. Amarasinghe, S. A. Su, X. Y. Dong, L. Zappia, M. E. Ritchie and Q. Gouil, Genome Biol., 2020, 21, 30.
7. S. Goodwin, J. D. McPherson and W. R. McCombie, Nat. Rev. Genet., 2016, 17, 333–351.
8. T. W. Murphy, Y. P. Hsieh, B. H. Zhu, L. B. Naler and C. Lu, Anal. Chem., 2020, 92, 2519–2526.
9. Z. Y. Xu, Y. T. Wang, K. W. Sheng, R. Rosenthal, N. Liu, X. T. Hua, T. Y. Zhang, J. Y. Chen, M. D. Song, Y. X. Lv, S. J. Zhang, Y. J. Huang, Z. L. Wang, T. Cao, Y. F. Shen, Y. Y. Jiang, Y. S. Yu, Y. Chen, G. J. Guo, P. Yin, D. A. Weitz and Y. C. Wang, Nat. Commun., 2023, 14, 5130.
10. D. G. Maghini, E. L. Moss, S. E. Vance and A. S. Bhatt, Nat. Protoc., 2021, 16, 458–471.
11. M. Sereika, R. H. Kirkegaard, S. M. Karst, T. Y. Michaelsen, E. A. Sørensen, R. D. Wollenberg and M. Albertsen, Nat. Methods, 2022, 19, 823–826.
12. R. Ziraldo, M. J. Shoura, A. Z. Fire and S. D. Levene, Nucleic Acids Res., 2019, 47, e92.
13. Uzma and H. Xiong, Talanta, 2025, 293, 128029.
14. L. Qiao, K. Szuttor, C. Holm and G. W. Slater, Nano Lett., 2023, 23, 1343–1349.
15. G. F. Carle, M. Frank and M. V. Olson, Science, 1986, 232, 65–68.
16. L. Lopez-Canovas, M. B. M. Benitez, J. A. H. Isidron and E. F. Soto, Anal. Biochem., 2019, 573, 17–29.
17. H. Neoh, X. E. Tan, H. F. Sapri and T. L. Tan, Infect., Genet. Evol., 2019, 74, 103935.
18. D. L. Liu, Q. J. Li, J. Luo, Q. T. Huang and Y. B. Zhang, BMC Genomics, 2023, 24, 125.
19. J. M. Guo, D. Brassard, N. Adam, A. J. Verster, J. A. Shay, C. Miville-Godin, M. Janta-Polczynski, J. Ferreira, M. Mounier, A. V. Pilar, K. Tapp, A. Classen, M. Shiu, D. Charlebois, N. Petronella, K. Weedmark, N. Corneau and T. Veres, Lab Chip, 2024, 24, 182–196.
20. D. S. Juang, S. M. Berry, C. Li, J. M. Lang and D. J. Beebe, Anal. Chem., 2019, 91, 11848–11855.
21. P. Bruand, I. Tijunelyte, A. Castinel, C. Donnadieu, P. Joseph and A. Bancaud, Anal. Chem., 2023, 95, 18099–18106.
22. P. V. Jones, G. L. Salmon and A. Ros, Anal. Chem., 2017, 89, 1531–1539.
23. P. G. Zhu and L. Q. Wang, Chem. Rev., 2022, 122, 7010–7060.
24. S. Gimondi, H. Ferreira, R. L. Reis and N. M. Neves, ACS Nano, 2023, 17, 14205–14228.
25. S. Q. Min, T. H. Zhan, Y. Lu, D. Pan, X. Q. Chen and B. Xu, Lab Chip, 2023, 23, 4680–4689.
26. X. C. Wu, S. Q. Min, T. H. Zhan, Y. E. Huang, H. Niu and B. Xu, Lab Chip, 2024, 24, 4379–4389.
27. D. L. Li, W. Li, R. Chen, X. Zhu, D. D. Ye, Y. Yang and Q. Liao, Anal. Chem., 2023, 95, 5828–5837.
28. L. Jiao, Z. B. Pan, D. L. Li, Y. J. Hu, Q. Q. Guo, M. J. Wang, C. Ding, X. T. Luo, K. Deng and G. H. Yang, Adv. Funct. Mater., 2025, e12912.
29. X. C. Xu, H. Niu, T. H. Zhan, Y. L. Wu, P. Qiao, L. Hu and B. Xu, Anal. Chim. Acta, 2025, 1368, 344269.
30. T. H. Zhan, S. Gao and B. Xu, Sens. Actuators, B, 2023, 375, 132947.
31. B. Xu, Y. G. Huang, S. Q. Min, X. C. Wu, T. H. Zhan, J. H. Liu, F. Z. Niu and H. Niu, Analyst, 2025, 150, 1862–1871.
32. J. L. Viovy, Mol. Biotechnol., 1996, 6, 31–46.
33. L. A. Holland and L. D. Casto-Boggess, Annu. Rev. Anal. Chem., 2023, 16, 161–179.
34. A. R. Li, X. J. Chen and V. M. Ugaz, Anal. Chem., 2010, 82, 1831–1837.
35. J. Y. Wang, J. E. Butler and A. J. C. Ladd, Proc. Natl. Acad. Sci. U. S. A., 2025, 122, e2417757122.
36. M. Hügle, O. Behrmann, M. Raum, F. T. Hufert, G. A. Urban and G. Dame, Analyst, 2020, 145, 2554–2561.

---