This study was approved by the Medical Ethics Committee of the Second People’s Hospital of Lianyungang City (Permit Number: 2020005). The clinical strains were collected from 2020 to 2021 and isolated from sputum, urine, drainage fluid, or secretion samples. All the isolate samples were obtained written consent, on an institutionally approved document, from every patient.

Four PCR-amplified and sequenced ‘standard strains,’ including bla _KPC, bla _NDM, bla _OXA-48-like, and bla _IMP from Klebsiella pneumoniae were used to establish the RPA-LFS assay for carbapenemases. Seven other common pathogenic bacteria, including Klebsiella pneumoniae, Escherichia coli, Pseudomonas aeruginosa, Acinetobacter baumannii, Streptococcus pneumoniae, Staphylococcus aureus, and Enterococcus faecalis were used to validate the specificity of the RPA-LFS method. A total of 207 clinical strains with carbapenem resistance validated by the paper diffusion method were collected from 2020 to 2021 and verified that no duplicate isolates were from the same patient. These clinical isolates, including K. pneumoniae, Escherichia coli, Enterobacter cloacae, Citrobacter freundii, and Serratia marcescens, were used to validate the practical application of the RPA-LFS technology for the rapid detection of carbapenemases in Enterobacterales ( Table 1 ). All strains were collected from the microbiology laboratory of the Second People’s Hospital of Lianyungang City and identified by MALDI-TOF mass spectrometer.

For reactions using purified genomic DNA as a template, genomic DNA was extracted using the Bacterial Genomic DNA Extraction Kit (Tiangen Biochemical Technology Co., Ltd., Beijing, China) and stored at −20°C for backup. If bacterial cultures were used as templates, bacterial DNA was extracted using the heated boiling method. The individual colonies were suspended in 50 μL Tris-EDTA buffers, boiled for 10 min, and centrifuged at 12000 × g for 10 min, after which the supernatant was used as the DNA template.

For the primer design method, sequences of all the isoforms of bla _KPC, bla _NDM, bla _OXA-48-like and bla _IMP genes were downloaded from GeneBank. Five pairs of primers for each carbapenemase gene detection were designed separately using Primer Premier 5.0 to capture all the variants based on the regions conserved among the isoforms. The primer design parameters were: size setting of 30–35, product size of 100–500 bp, GC content of 20%–80%, and Tm value setting of 50–100. Default settings were used for all other parameters. Five pairs of primers were selected for testing according to their scores from the highest to the lowest.

For the RPA experiments, we used the TwistAmp Liquid DNA Amplification Kit (TwistDx Inc., Maidenhead, United Kingdom). A total of 50 μL of the reaction system was added to the tubes in the following order: 25 μL 2 × reaction buffer, 5 μL 10 × base mix, 2.5 μL 20 × core mix, 2.1 μL upstream primer (10 μM) and 2.1 μL downstream primer (10 µM; General Biosystems Co., Ltd., Anhui, China), 9.8 μL ddH₂O, and 1 μL template. To ensure that all reaction systems reacted simultaneously, 2.5 μL of 280 mM magnesium acetate was added to the PCR tube caps, and the template and 280 mM magnesium acetate were added to the reaction system simultaneously using transient centrifugation. The reaction system was vortex-centrifuged and immediately incubated in a heater at 37°C for 30 min. No genomic DNA template reaction system was used as negative control. Each sample has two tubes of reaction, one for the sample itself and the other for the control. Amplification of primers was detected using 1.5% agarose gel electrophoresis.

Primer Premier 5 software was used to design specific probes between forward and reverse primer targeting sequences, and the formation of dimers, hairpin structures, mismatches, etc. between the probe and the reverse primer should be theoretically avoided as much as possible. The design principles are: (1) The probe size is 46-51 bp, GC content is 20-80%, and Tm is 57-80°C; (2) The maximum hairpin score is 9, and the maximum primer-dimer score is set to 9. The maximum poly-X is set to 5, and other parameters are set to default values; (3) The 5’ end of the probe is labeled with FITC, and the 3’ end was blocked with SpC3, the base at the middle of the probe was replaced with tetrahydrofuran (THF), and there was at least 30 bp of base before the THF site, while at least 15 bp of base followed; (4) The 5’ end of the reverse primer was labeled with biotin.

The RPA-LFS experiment was performed using the TwistAmp DNA Amplification nfo Kit (TwistDx Inc., Maidenhead, United Kingdom) in 50 μL of the reaction system. The following systems were added sequentially to the lyophilized powder tubes that contained the enzyme components: 29.5 μL rehydration buffer, 2.1 μL forward primer (10 μM), 2.1 μL reverse primer (10 μM), 0.6 μL probe (10 μM), 12.2 μL ddH₂O and 1 μL template. To ensure that all reaction systems began simultaneously, 2.5 μL of 280 mM magnesium acetate was added to the tube cap, transiently centrifuged, and immediately incubated in a thermostatic heater at 37°C for 30 min. Then, 5 µL of amplification product was used for LFS (Ustar Biotechnologies Ltd., Hangzhou, China) visual detection within 3 min. There are two red lines displayed on the LFS, namely the control line (top) and the test line (bottom). The control line exists in every test to ensure the validity of the LFS, while the test line could only be observed for positive reactions. Each sample has two strips, one for the sample itself and the other for the control.

To evaluate the detection compliance of the RPA-LFS method, parallel PCR experiments were carried out to calculate the compliance rates of the results of the two methods. A total of 207 pre-validated CRE clinical strains were collected and bacteria were treated using the heated boiling method. One microliter of the boiled resuspension was taken and used as a template for the RPA-LFS and PCR assays. The compliance rate was calculated as:{(number of positive samples for both methods + number of negative samples for both methods)/total number of samples} × 100%.

The bla _KPC, bla _NDM, bla _OXA-48-like, and bla _IMP genes were used as target sequences, and five pairs of primer sets were designed for each gene targeting to the highly conserved area ( Supplementary Table 1 ). The genome DNA of four standard strains of K. pneumoniae were used as templates to verify the amplification of each of the five primer pairs. As shown in Figure 1 , all designed primers amplified the target bands as expected. Primer pairs with brighter target bands, fewer primer-dimers, without non-specific amplification are of better choice. Therefore, bla _KPC-4, bla _NDM-1, bla _OXA-48-like-2 and bla _IMP-4 were selected for subsequent detection.

The use of probes in the RPA reaction increases the amplification specificity and reduces primer-dependent artifacts. Specific probes were designed within the targeting fragments of the four primer pairs that were screened for better performance ( Supplementary Table 1 ). The amplification performance and false positives of the primer-probe-needle set were verified. The combination of the four primer-probe pairs provided correct positive signals (two visible red bands both on the test lines and control lines) when tested using RPA-LFS, which indicated that the four primer-probe pairs had good amplification performance. However, they also showed a visible weaken red band on the test line in the no template control, which indicated there are false-positive signals for all these five primer-probe combinations ( Figure 2 ).

RPA can tolerate some base mismatches between the primer/probe and template, which provides some flexibility in primer/probe design and screening (Daher et al., 2015). Analysis using the Primer Premier 5 software revealed that the probe had multiple consecutively matching bases to the reverse primer, which could result in a false-positive signal. Therefore, base substitutions were introduced to eliminate false-positive signals, and the principles of substitution were (1) Breaks were performed on sites with more than four contiguous matching bases or two or more contiguous matching bases at the 3 ‘ end; (2) no substitution of three bases near the 3’ end; (3) no consecutive two-base substitutions; (4) substitution of preferably no more than three bases, otherwise, the sensitivity of the assay may be affected, and (5) A-G and T-C swaps were used preferentially. The sequences of the modified reverse primer (mR) and probe (mP) are listed in Table 2 , and the replacement bases are indicated in red. Using this modified primer-probe device, false-positive signals were eliminated without affecting amplification performance ( Figure 3A ). The primer-probe sets were used for all subsequent RPA-LFS reactions. Meanwhile, analysis of the RPA amplification products using 1.5% agarose gel electrophoresis showed that the amplification products of bla _KPC, bla _IMP, and bla _OXA-48-like primer-probe sets all showed two clear bands, which represented the amplification products of the forward-reverse and probe-reverse primers, respectively, whereas the two amplification products of bla _NDM were of similar size and, thus, could not be distinguished from each other, so they were shown as a single band ( Figure 3B ).

K. pneumoniae genomes carrying bla _KPC, bla _NDM, bla _OXA-48-like, and bla _IMP genes as well as seven other common pathogenic bacteria were used as templates to detect the specificity of primer-probe combinations. Results are shown in Figure 4 , all four primer-probe sets could only detect strains containing the corresponding genes, and no bands were present at the test lines when using genomic DNA from other respiratory bacterial pathogens. Overall, these results illustrated that the established RPA-LFS detection system had good specificity towards the four classes of carbapenemases and no cross-reactivity with other pathogenic bacteria.

To assess the detection limit of RPA-LFS, purified genomes of bacteria carrying the four carbapenemases were subjected to 10-fold serial dilutions, ranging from 10⁶ to 10 fg (50 μL/reaction volume with 1 μL of diluted genome added to each reaction). Results are shown in Figure 5 . bla _KPC, bla _NDM, and bla _OXA-48-like all had a minimum detection line of 10² fg/reaction; moreover, bla _IMP had a 10³ fg/reaction. The lowest line for bla _KPC, bla _NDM, and bla _OXA-48-like was 10³/reaction by PCR and 10⁴/reaction for bla _IMP. Our established RPA-LFS reaction system was more sensitive than the PCR reaction.

The four carbapenemase families of the 207 clinically validated CRE isolates were examined using the RPA-LFS method and conventional PCR. A total of 134 CPE containing the bla _KPC gene, 69 bla _NDM, 3 bla _OXA-48-like, and 1 bla _IMP strain ( Table 3 ). There was 100% compliance with the two methods of testing. Notably, the RPA-LFS method was simple and rapid and showed a higher proficiency in detecting carbapenemase genotypes in clinical isolates compared with that of conventional PCR and subsequent sequencing.