Design and Optimization of Malaria SHERLOCK Diagnostic.

Fig. 1 illustrates the workflow of our simplified SHERLOCK diagnostic. The test combines a 10-min sample preparation step and a 60-min SHERLOCK assay prior to endpoint analysis via lateral flow strip or fluorescence measurement. CRISPR-based diagnostics utilize the programmable endonucleases (Cas enzymes) of CRISPR-associated microbial adaptive immune systems. Cas12a (also known as Cpf1) is one such RNA-guided, DNA-cleaving enzyme, which can be programmed with CRISPR guide RNAs (gRNA) to construct highly sensitive and specific nucleic acid detection platforms (15⇓⇓⇓–19). Programmed Cas12a is activated through recognition of its double-stranded DNA (dsDNA) target and exhibits indiscriminate, nonspecific DNase activity that cleaves nontarget DNAs. We exploit the nonspecific degradation of fluorophore-quencher labeled reporter single-stranded DNA (ssDNA) to detect the presence of the dsDNA target that activated Cas12a. To further decrease the LOD, a reverse-transcriptase recombinase polymerase amplification (RT-RPA) step is added before Cas12a detection to increase target DNA concentrations (Fig. 2). RPA is a powerful isothermal nucleic acid amplification tool comprised of three core enzymes: a recombinase, an ssDNA-binding protein, and a strand-displacing polymerase that coordinates DNA synthesis from primer-paired target DNA (20).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

SHERLOCK diagnostic workflow: 1) Human serum, whole blood, or DBS samples undergo a 10-min S-PREP protocol where the sample is suspended in 20% (wt/vol) Chelex-100 in TE buffer with 50 mM DTT and incubated at 95 °C for 10 min; and 2) transfer of suspended sample to lyophilized SHERLOCK pellet followed by incubation at 40 °C for 60 min prior to endpoint analysis via fluorescence or lateral flow strip.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Schematic of one-pot SHERLOCK assay. RT-RPA amplifies Plasmodium species target sequences and occurs in parallel with programmed Cas12a detection, resulting in cleavage of target sequences and collateral cleavage of spiked fluorophore-labeled ssDNA reporter detectable by fluorescent measurement or lateral flow readout using Au-NP, gold nanoparticles.

For endpoint analysis, released fluorophore from cleaved reporter ssDNA was measured by a plate reader or a handheld fluorimeter. Particularly in RLS, use of a handheld fluorimeter enables a field-applicable readout method. We did not find a significant difference in the sensitivity performance between machines and observed a similar 7- to 10-fold change in fluorescence between platforms, although they had different baselines (SI Appendix, Fig. S1). For use of the handheld fluorimeter, SHERLOCK reactions (50 μL) were performed combined in triplicate (150 μL) to increase the volume size for appropriate instrument reading.

Our assays are also adapted for endpoint detection via lateral flow strip based upon degradation of an ssDNA reporter that is labeled on opposing ends with FAM and biotin. The FAM-biotinylated reporter conjugates to anti-FAM gold nanoparticles contained within commercial lateral flow strips. If the reporter remains intact, FAM-labeled reporter/anti-FAM conjugates accumulate at the first line of the strip immobilized by streptavidin (control line). In the presence of activated Cas12a, the reporter is cleaved and freed FAM/anti-FAM conjugates are released to collect at the second line of the lateral flow strip containing anti-rabbit antibody (test line), which binds anti-FAM antibodies (Fig. 2).

There are many Cas enzymes that could have been used. We chose Cas12a [as opposed to the Cas13 family (21, 22), which also has nonspecific nuclease activity] so DNA targets could be directly detected instead of RNA, particularly in DBS where RNA may be degraded. The rapid enzymatic kinetics of Cas12a also make this nucleic acid-based technology comparable to the POC format of antigen-based lateral flow immunoassays. Cas12a bound to its dsDNA activator is capable of ∼1,250 turnovers per second with a catalytic efficiency (k_cat/K_M ∼1.7 × 10⁹ s⁻¹ M⁻¹) approaching the rate of diffusion (17). The addition of a RT enzyme further enhances the sensitivity by transcribing multiple-copy RNAs from our target sequence into DNA for detection. We optimized SHERLOCK parameters, including reaction temperature, RPA primer concentration, RT commercial brand, and ssDNA reporter concentration (SI Appendix, Fig. S2). We also lyophilized the reaction into a pellet to be resuspended with an S-PREP–treated sample for cold-chain independence in the field, and importantly, also improved the LOD by increasing sample input volume.

RPA Primer and gRNA Selection.

Our SHERLOCK assays were designed to detect four of the most common pathogenic species of malaria. We iterated a two-step design process of RPA primer screen followed by a gRNA screen. RPA primer targets were identified by reviewing the literature for the best-performing NAATs and searching for conserved and specific sequences from alignment of species-specific strains available from the National Center for Biotechnology Information (NCBI). For P. falciparum 18S rRNA, mitochondrial (cytochrome oxidase III, cytochrome B), and subtelomeric (Pfr364) targets were screened (23⇓⇓⇓⇓⇓–29). The Pfr364 target, which is a species-specific, noncoding subtelomeric repeat sequence present in 41 copies on the P. falciparum genome, had the best signal in comparison to the other targets (SI Appendix, Fig. S3). Moreover, our selected gRNA had >90% sequence homology among all assembled P. falciparum genomes available in the NCBI, as well as 86% of sequences from the Pf3k dataset (an open-access collaboration and deep-genomic sequencing database) accessed via the Integrative Genomics Viewer (IGV) (30). For P. vivax, we tested an 18S rRNA and mitochondrial target, and found that the mitochondrial target worked best (copy number per parasite can be as high as 20) (27, 31). For P. ovale and P. malariae, we tested different regions of the 18S rRNA gene known to be conserved species-specific targets (27, 32, 33) typically present in four to eight copies per genome (notably, copy number is variable and depends on the parasite life cycle stage). We mapped the sequence targets’ primers and gRNA in SI Appendix, Fig. S4 and aligned them to the corresponding regions in off-target Plasmodium species (either homologous genes, or analogous sequences identified using NCBI’s Basic Local Alignment Search Tool [BLAST] with the lowest E-values). Despite overlap in RPA primers, which can tolerate significant sequence mismatch, we found that few-nucleotide differences in gRNA sequence were sufficient to obtain discriminating species-specific detection.

We constructed five forward (F1 to F5) and five reverse primers (R1 to R5) per sequence target using guidance provided by the TwistDx manufacturer; primers were 30 to 40 nucleotides long, with goal amplicons of 100 to 200 base pairs in length. We paired forward and reverse primers for a total of 25 combinations (F1:R1-5, F2:R1-5, F3:R1-5, F4:R1-5, F5:R1-5) for each sequence target and two to three of the best-performing pairs were selected for the optimization of gRNA design (SI Appendix, Fig. S5). RPA was performed according to the manufacturer’s instructions as described in Materials and Methods. Cas12a recognizes a short nucleotide sequence (TTTN) called the protospacer adjacent motif (PAM) for generation of distal dsDNA cleavage, and two to four gRNAs based upon the TTTN PAM were designed within the RPA amplicon. The RPA reaction for each primer set was then transferred to a Cas reaction as described in Materials and Methods, and fluorescent kinetics were monitored for selection of best-performing gRNAs (Table 1).

Sample Preparation.

Accessing sample nucleic acids in a field-applicable manner involves overcoming several challenges. Preparation requires lysing the red blood cell (RBC) and parasite membrane (with the exception of the invasive merozoite form, all blood-stage parasites are intraerythrocytic), deactivating multiple inhibitory blood components, and importantly, appropriately deactivating nucleases that could shear the ssDNA reporter and lead to a false-positive signal. The requirement for simplicity and low cost ruled out commercial nucleic acid extraction kits. To test sample preparation methods, we used simulated whole-blood samples of live intraerythrocytic P. falciparum spiked into purchased EDTA-treated human blood (VWR International) to a final 1 fM (602 parasites per microliter) concentration for rehydration of our one-pot lyophilized P. falciparum SHERLOCK assay described in Materials and Methods.

One approach that did not work was HUDSON (heating unextracted diagnostic samples to obliterate nucleases), a simplified sample preparation method for viral nucleic acid extraction (16) compatible with Cas13 SHERLOCK. In HUDSON, whole-blood samples are pretreated with 100 mM TCEP [Tris(carboxyethyl)phosphine] and 1 mM EDTA to augment protein deactivation, followed by a two-step process of nuclease deactivation (heating for 5 min at 50 °C) followed by viral inactivation (heating for 5 min at 64 °C). HUDSON-treated simulated whole-blood samples produced minimal signal, likely from not accessing the intracellular parasitic nucleic acid.

We therefore assessed alternative simplified sample preparation protocols described in Materials and Methods, including various detergents, thermal lysis, and chemical deactivation protocols (Fig. 3). We discovered that treating samples with 50 mM DTT and 10 mM EGTA, followed by 95 °C incubation for 10 min resulted in a robust SHERLOCK signal, although we noticed variability in the no-template control signal that we attributed to background nucleases in different blood aliquots. However, when we tested the DTT/EGTA/95 °C sample preparation method on patient P. falciparum and P. vivax serum samples from the Dominican Republic, we found bidirectional cross-reactivity of our species-specific SHERLOCK assays. Using our P. falciparum-specific assay, P. vivax patient serum samples produced a false-positive signal (Fig. 4A). P. falciparum patient samples also produced a false-positive signal using the P. vivax-specific assay (Fig. 4B).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Sample preparation methods tested with SHERLOCK P. falciparum assay using simulated malaria samples of live intraerythrocytic P. falciparum spiked into whole blood at 1 fM (602 parasites per microliter) concentration. (A) Detergents and heating methods assessed for SHERLOCK compatibility. (B) Combinations of chelating and reducing agents tested for optimization of chemical deactivation of nucleases and inhibitors. Asterisks indicate significant differences from untreated simulated whole blood sample assessed by Student’s two-tailed t test. Bars: mean ± SD of three technical replicates. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns, not significant.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Specificity of SHERLOCK assays. (A) Using P. falciparum assay and DTT/EGTA/95 °C sample preparation, P. falciparum and P. vivax patient serum in SHERLOCK diagnostic display similar fluorescent kinetics that are eliminated when an aliquot of the same P. vivax serum undergoes nucleic acid extraction via commercial kit. (B) Using P. vivax assay and DTT/EGTA/95 °C sample preparation, P. falciparum serum demonstrates a false-positive signal that is eliminated when an aliquot of the same P. falciparum serum undergoes nucleic acid extraction via commercial kit. (C) False-positive P. vivax signal is eliminated with S-PREP. (D) False-positive P. falciparum signal is eliminated with S-PREP. (E) Performance of SHERLOCK diagnostic on clinical patient serum and whole-blood samples prepared with S-PREP: 5 P. falciparum samples (4 serum, 1 whole blood), 10 P. vivax serum samples, and 5 serum samples from healthy controls.

The false-positive signals were eliminated, however, when DNA from the same cross-reacting P. vivax and P. falciparum patient serum samples was extracted via QIAamp DNA mini kit (Qiagen), spiked into a healthy commercial serum no-template control (10-ng extracted DNA into 20 µL serum; Sigma Aldrich), and retested (Fig. 4 A and B). Furthermore, the extracted patient serum DNA maintained a robust species-specific signal with the appropriate Plasmodium species-specific assay. Extracted nucleic acid reflected combined human and parasite DNA, with numbers of human sequences dwarfing numbers of parasite sequences, and the highly sensitive and specific performance of the appropriate SHERLOCK assay on the extracted nucleic acid made cross-reactivity due to human DNA unlikely. These results were also observed on all 5 P. falciparum and all 10 P. vivax specimens, making coinfection unlikely and the specimens had all undergone species-specific qualitative PCR testing (ARUP). We hypothesized that the cross-reactivity could be secondary to nonspecific ssDNA reporter cleavage from higher concentrations of nucleases in “sick” versus “healthy” serum that resulted in incomplete deactivation of nucleases in “sick serum” by our DTT/EGTA/95 °C simplified preparation method.

This hypothesis was confirmed when we developed S-PREP using a buffer comprised of a stronger chelating agent: 20% (wt/vol) Chelex-100 (Bio-Rad) suspended in TE buffer with 50 mM DTT. Chelex-100 is a resin containing styrene divinylbenzene copolymers with paired iminodiacetate ions that act as chelating groups in binding polyvalent metal ions (34). Nucleases require metal ions as cofactors and therefore chelating agents inhibit their activity. S-PREP is a simplified sample preparation method where sample is diluted 1:3 (5 μL into 15 μL of S-PREP buffer) followed by heating to 95 °C for 10 min. We eliminated the false-positive signals of serum samples using S-PREP (Fig. 4 C and D). We conclude that higher concentrations of nucleases present in “sick” serum (patients sick with another disease but not the target disease) necessitate stronger nuclease deactivation procedures. We are unique in reporting on this cross-reactivity in nonnucleic-acid extracted clinical samples for SHERLOCK, as we are not aware of other studies comparing performance using unextracted samples against controls from patients sick with a different disease (instead of only comparing to healthy control specimens). This highlights the importance of considering baseline nuclease activity in specimen types with CRISPR-based assays, as the readout is dependent on reporter nucleic acid cleavage and contaminating nucleases are a major concern for false positives. Importantly, we demonstrated that S-PREP can deactivate high levels of nucleases and we confirmed the absence of false positives in our clinical sample set with 100% specificity (P. falciparum n = 4 serum, n = 1 whole blood; P. vivax n = 10 serum; healthy serum patient controls n = 5) (Fig. 4E). We additionally found that S-PREP was compatible with RNA-only simulated samples, despite its increased susceptibility to hydrolysis in comparison to DNA, demonstrated by detection of an RNA-only synthetic target prepared using S-PREP (SARS-CoV-2 RNA target detected in novel SHERLOCK assay) (SI Appendix, Fig. S6).

To further assess the field versatility of our work, we also tested our S-PREP/SHERLOCK diagnostic on simulated samples from multiple specimen collection types. We spiked live intraerythrocytic P. falciparum into whole blood and plasma stored in multiple different specimen collection tubes (acid-citrate dextrose, EDTA K-2, EDTA K-3, Na heparin, Na citrate, plasma heparin, and plasma EDTA) to a final 1 fM (602 parasites per microliter) concentration and prepared these samples with S-PREP for rehydration of our SHERLOCK assay, as described in Materials and Methods. Although many of these additives are known PCR inhibitors, all simulated samples were able to produce a distinguishable signal from the no-template control (SI Appendix, Fig. S7). The compatibility of SHERLOCK with unextracted samples from multiple specimen tube types emphasizes its unique robustness, versatility, and ultimately suitability for RLS.

Performance and Readout of Malaria SHERLOCK Diagnostic.

We determined the analytical sensitivity of our assays using industry standard definitions of the diagnostic LOD to guarantee a 95% probability of successful detection. We performed septet replicate testing on three different runs on simulated whole-blood samples (described in Materials and Methods) for each Plasmodium species and used probit analysis to establish: P. falciparum 0.36 parasite per microliter blood (95% confidence interval [CI] 0.23 to 1.0), P. vivax 1.2 parasites per microliter (95% CI 0.52 to 6.2), P. ovale 2.4 parasites per microliter (95% CI 0.81 to 19), and P. malariae 1.9 parasite per microliter (95% CI 1.1 to 12) (Fig. 5A and Table 2). This reaches the WHO LOD goal for low endemnicity (asymptomatic carriage) settings and is notable in that SHERLOCK was capable of attomolar to subattomolar detection in the absence of commercial kit nucleic acid extraction and sample nucleic acid concentration. This also emphasizes the ultrasensitive capacity of SHERLOCK in that our best detection level (0.36 parasite per microliter for P. falciparum) closes in on the theoretical LOD of the engineering design. Using 12.5 µL of sample input as we have, a 0.3 parasite per microliter concentration sample has a 95% probability of containing at least one parasite (and therefore being detectable) following the Poisson distribution (SI Appendix, Fig. S8).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

SHERLOCK performance. (A) Sensitivity of SHERLOCK diagnostic for detection of Plasmodium species by comparison of probit regression curves obtained from 21 replicates of 5 dilutions. (B) Fluorescence kinetics of P. falciparum SHERLOCK assay at 100 aM (60 parasites per microliter) and 2 aM (1 parasite per microliter) concentrations. (C) Specificity of SHERLOCK diagnostic using 10 fM (6,020 parasites per microliter) concentrations of parasite. (D) Comparison of performance between simulated DBS and whole-blood samples. All experiments used simulated whole-blood samples. ***P < 0.001 for Student’s t test between fluorescent output of sample type versus no-template control.

Our CRISPR diagnostic can also detect clinically relevant levels of parasitemia in 40 min or less from unextracted blood samples (10-min S-PREP followed by 30-min SHERLOCK) with better sensitivity than existing POC antigen-based RDTs, filling an important clinical diagnostic gap for hrp2 deletion P. falciparum and nonfalciparum malaria. A 0.001% parasitemia (assuming a RBC mean corpuscular volume of 80 fL and hematocrit of 45%) corresponds to ∼60 parasites per microliter (100-aM concentration), for which a 30-min detectable signal difference between the no-template control and infected blood is readily apparent (Fig. 5B). This level of parasitemia would likely be missed on RDT or light microscopy (a technician would have to view 100,000 RBCs to view an infected RBC, which is theoretically possible, but would require considerable effort). Finally, while there is no consensus definition of asymptomatic malaria, some have used parasite density cutoffs of 5,000 parasites per microliter blood (∼8.5 fM) as a threshold (vaccine trials and epidemiological studies), which is a rapidly detectable concentration with SHERLOCK (35⇓–37).

The analytical specificity of our assays was determined using simulated clinical samples at a 10-fM concentration (6,020 parasites per microliter) and demonstrated no detection of nontarget Plasmodium species, confirming high specificity (Fig. 5C). We surmise that the highly specific performance of these assays is likely attributable to a two-step target selection via RPA primer match and amplification, followed by gRNA match and Cas activation. For clinical sensitivity and specificity, we were able to detect and differentiate 5 P. falciparum (4 serum, 1 whole blood) and 10 P. vivax samples with 100% accuracy (Fig. 4E). Deidentified clinical samples were purchased from BocaBiolistics and came from symptomatic patients from the Dominican Republic. They had been previously characterized by both the BinaxNOW Malaria RDT (Alere) and species-specific qualitative PCR (ARUP), demonstrating that our diagnostic had 100% concordance with these methods in our limited clinical set.

We additionally prepared simulated DBS to a 2-aM (one parasite per microliter blood) concentration of the four Plasmodium species and tested them with our S-PREP/SHERLOCK protocol with modifications, as described in Materials and Methods. A robust fluorescence signal was demonstrated at the 1-h time point that was significantly different from the no-template control. The only notable difference in assay performance compared with whole-blood samples was a greater no-template control signal in simulated DBS samples, likely from autofluorescence from the paper substrate (Fig. 5D).

In addition to establishing the analytical LOD via fluorescent measurement, we also demonstrated a lateral flow readout given its ease of use in RLS. We found that a clearly visible band was distinguishable at 50 aM (30 parasites per microliter) for all of the Plasmodium species assays (Fig. 6); this LOD is higher than that of our fluorescent readout, but it is still lower than best-in-class contemporary RDTs (38).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

SHERLOCK lateral flow assay performance. (A–D) Detection of 1 fM (∼602 parasites per microliter), 100 aM (60 parasites per microliter), 50 aM (30 parasites per microliter), and 2 aM (1 parasite per microliter) concentrations of P. falciparum, P. vivax, P. ovale, and P. malariae, respectively, and comparison to 1-fM concentrations of off-target Plasmodium species for each assay. (E) Background-subtracted grayscale intensity averages of test line for three separate flow tests ±SD. All experiments used simulated whole-blood samples.