Trans-species microRNAs from the parasitic plant Cuscuta campestris specifically avoid loading onto self Argonautes

Abstract

The obligate parasitic plant Cuscuta campestris delivers trans-species microRNAs (miRNAs) into host plants that silence host mRNAs. Here, the genetic requirements for biogenesis, movement, and function of these miRNAs were investigated. Primary miRNA transcript accumulation precedes mature miRNA accumulation by 24 to 48 h. Trans-species miRNAs accumulate in host tissues a short distance from the site of parasite attachment. Trans-species miRNAs require C. campestris but not host Dicer-Like 1 (DCL1) for accumulation. These miRNAs specifically avoid Argonaute (AGO) loading in C. campestris tissue where they instead accumulate as miRNA/miRNA* duplexes. After arrival and short-distance spreading in host tissues, they are loaded onto host AGO proteins, including AGO1 and AGO2. This study clarifies the transcription, dicing, delivery, and function of C. campestris trans-species miRNAs. We propose that selective avoidance of self-AGO loading is a mechanism to facilitate high rates of delivery of these “export only” miRNAs to host tissues.

Graphical Abstract

Graphical Abstract.

Introduction

Cuscuta is a genus of parasitic plants with minimal chlorophyll and no functional leaves or roots, producing yellow to orange vines. As obligate parasites, they rely on host plants to complete their life cycle [1]. The Cuscuta genus contains ~200 species with a diverse host range [2]. Some species exhibit narrow host preferences, while others, like Cuscuta campestris, parasitize a broad spectrum of plants—including most eudicots and some monocots [3]. A broad host range and aggressive growth make C. campestris an agricultural nuisance that can cause substantial crop losses. Understanding the virulence mechanisms of Cuscuta is therefore critical for developing effective crop protection strategies. The haustorium is a specialized organ that anchors Cuscuta to its host and primarily functions as a vascular connection for the withdrawal of water, photosynthate, and nutrients from the host plant. The haustorium also serves as a conduit for extensive molecular exchange, including many types of RNAs [4].

Canonical plant miRNAs are usually 21 or 22 nucleotides in length and are processed from single-stranded precursors that fold into characteristic hairpin structures. These hairpins are processed by the endonuclease Dicer-Like 1 (DCL1) to produce miRNA/miRNA* duplexes. The mature miRNA is loaded onto an Argonaute (AGO) protein to form the RNA-induced silencing complex (RISC), while the miRNA* is subsequently degraded [5]. The RISC mediates sequence-specific cleavage and/or translational repression of target mRNAs, leading to gene silencing at the post-transcriptional level [5, 6]. Gene regulation by canonical plant miRNAs collectively impacts many biological functions, including development [7], organogenesis, inorganic nutrition [8], and pathogen defense [9]. Other categories of short interfering RNAs (siRNAs) can also be loaded onto AGO proteins to specify gene silencing. Plant siRNAs are often produced from double-stranded RNA precursors by one or more alternative DCL proteins (DCL2, 3, or 4).

Exchange of functional miRNAs and siRNAs between plants and their associated organisms has been reported in diverse systems, including plant–plant, plant–fungus, plant–insect, and plant–mammal interactions [10, 11]. One well-studied example is the fungal grey mold pathogen Botrytis cinerea, which delivers siRNAs into host plants to suppress immunity [12], while plants counter by exporting siRNAs and miRNAs into the fungus to reduce virulence [13]. In both cases, the small mobile RNAs are processed in the donor organism—requiring DCL1/2 in the pathogen and DCL2/3/4 in the host plant. The mobile small RNAs travel within extracellular vesicles (EVs) and incorporate into an AGO protein upon arrival [13, 14]. However, the nature of the mobile agent may differ between the two organisms. Exported Arabidopsis thaliana-derived small RNAs require RNA-binding proteins, such as AGO1, RNA helicases, and annexins, to be sorted into and stabilized within EVs [15]. These proteins are bound to the small RNAs and co-localize with the EVs, suggesting that plant sRNAs are transported to B. cinerea as sRNA–protein complexes. In contrast, for fungal-derived siRNAs, no direct evidence describes their transport form, but successful incorporation into host AGO1 suggests they may move as free duplexes.

Cuscuta campestris expresses a cohort of 96 distinct miRNA families at high levels specifically in the haustorium [16–18]. Most of the MIRNA genes for these interface-induced miRNAs have a common promoter element, which is identical to that used to drive plant small nuclear RNA (snRNA) transcription [18]. This promoter element drives transcription by RNA polymerase III, which marks the C. campestris interface-induced MIRNA genes as distinct from canonical plant MIRNA genes: Canonical MIRNA genes in plants are transcribed by RNA polymerase II [19]. At least a subset of the interface-induced miRNAs accumulate in the host, where they target host transcripts for cleavage and secondary siRNA production, potentially modulating host gene expression to facilitate parasitism [17]. We refer to this subset of interface-induced miRNAs with confirmed activity against host mRNAs as trans-species miRNAs. Throughout this report, the specific term trans-species miRNAs is used when discussing only the subset of interface-induced miRNAs with proven activity; the broader term interface-induced miRNAs encompasses both the trans-species cases and all other parasite miRNAs with interface-specific, Pol III-dependent accumulation. Despite extensive efforts, we have found no evidence that C. campestris trans-species miRNAs are functional within the parasite [17, 20], suggesting they are used for “export only”.

Here we investigate the biogenesis and genetic requirements for C. campestris trans-species miRNA function. Rapid accumulation of trans-species miRNAs in haustoria is preceded by accumulation of miRNA primary transcripts. Mature trans-species miRNAs accumulate a short distance (∼1–2 centimeters) in host tissues adjacent to the haustorium. Parasite DCL1, but not host DCL1, is required for accumulation of trans-species miRNAs. Once made, trans-species miRNAs do not associate with parasite AGO proteins and instead appear to accumulate as free miRNA/miRNA* duplexes inside the parasite. Upon arrival in the host, they become associated with host AGO proteins, including AGO1 and AGO2, depending on the 5′ nucleotide. The selective avoidance of self-AGO loading may be a mechanism to promote high rates of export of C. campestris trans-species miRNAs.

Materials and methods

Preparation of host plants and Cuscuta campestris

Nicotiana benthamiana and A. thaliana (Col-0) host plants were grown under long-day conditions (16 h light/8 h dark) and watered weekly with a fertilizer supplement for four to 5 weeks prior to parasite attachment. C. campestris seeds were scarified in sulfuric acid for 1 h, with gentle swirling every 20 min to ensure even exposure. After scarification, seeds were placed on moist tissue paper folded into a Petri dish and incubated at 30°C under long-day conditions supplemented with far-red light for two and a half to three days. For host-induced haustoria, seedlings were taped to the host stem or petiole and incubated under far-red light–enriched conditions for at least a week, until haustoria were firmly formed. In vitro haustoria were induced as previously described [21]. Briefly, C. campestris seedlings were sandwiched between 3% agar gel and six microscope slides. The sandwiches were exposed to a far-red light–enriched environment for one hour, then immediately transferred to continuous blue light. Induced haustoria were harvested after 96 hours.

Construction and agroinfiltration of artificial miRNA vectors

Artificial miRNAs (amiRs) were designed using P-SAMS [22]. A combined transcriptome database was assembled from C. campestris [23] and N. benthamiana LAB3.6 [24]. Three amiR candidates specific for CcDCL1 and another three specific for NbDCL1 were initially selected. Each candidate was predicted to be specific for the intended DCL1 and to have very poor complementarity to the other DCL1. The candidate amiR showing the strongest silencing effect in pilot experiments was used for downstream analyses. The control amiR was designed to target an irrelevant mRNA (the human erythropoietin receptor mRNA, NM_000121). Forward and reverse oligos were synthesized, annealed, and cloned into the AtMIR390a backbone [25]. The recombinant vector was transformed into Agrobacterium strain GV3101 and infiltrated as previously described [26]. In brief, a single colony was used to inoculate LB broth and incubated overnight at 28°C with shaking at 250 rpm. Cells were collected by centrifugation (4000 × g, 15 min), resuspended in infiltration buffer (MgCl₂, MES, acetosyringone, sterile water), and incubated in a cool, dark place for >4 h before infiltration. Nicotiana benthamiana leaves were infiltrated using a 1 ml needleless syringe. Three leaves per plant were treated, specifically the first and second pairs of true leaves. Agroinfiltrated areas generally covered entire leaves; if complete coverage was not possible, the area near the leaf base was infiltrated to enhance systemic movement of amiRs [27]. Oligonucleotide sequences are listed in Supplementary Table S1.

Analysis of miRNA:miRNA* ratios

Three biological replicate small RNA-seq datasets from C. campestris host-free in vitro haustoria from a previously published experiment [18] (SRA accession numbers SRR19502459, SRR19502460, and SRR19502461) were retrieved as FASTQ files from the NCBI Short Read Archive (SRA). Adapters were trimmed using ShortCut version 1.0 with default settings (https://github.com/Aez35/ShortCut). A non-redundant list of C. campestris mature miRNA and miRNA* sequences was curated based on the miRNA annotated by Cerda-Herrara et al. [16]. Occurrences of each miRNA and miRNA* sequence in each trimmed FASTQ file were tallied and converted from raw reads to reads-per-million (RPM). The data were filtered to retain only miRNA families where the mature miRNA is empirically known through prior demonstration of active targeting. RPMs were tallied by family for mature miRNAs and miRNA*s separately; these tallies were used to calculate the percentage of miRNAs for each family.

Enrichment of AGO-bound miRNAs using TraPR

Host-induced haustoria were established on A. thaliana inflorescence stems for 10 days under far-red light–enriched conditions. Each inflorescence stem supported ~10 attachments. Stems were excised above and below each attachment site, avoiding inclusion of C. campestris tissue. Each stem segment measured ~2–3 millimeters in length. For each biological replicate, 30 to 40 stem segments (∼80 milligrams total) were pooled. For in vitro haustoria, a total of 60 to 70 C. campestris seedlings were used to induce ~50 milligrams of in vitro haustorial tissue for each biological replicate. Tissues were processed through TraPR columns following the manufacturer’s instructions (TraPR Small RNA Isolation Kit, Lexogen, #128.08) with minor modifications. Briefly, tissues were homogenized in 300 µl TLB buffer using a bead beater at high intensity for 90 s (host-induced haustoria) and 30 s (in vitro haustoria). Lysates were centrifuged at 10 000 × g for 5 min at 4°C and the supernatant was transferred to a clean low-protein-binding tube (Thermo, #88379). The beads were rinsed with 50 to 150 µl nuclease-free water, centrifuged again, and the supernatant was combined. This rinse-and-centrifugation step was repeated until at least 270 µl of clarified lysate was collected. A 20 µl aliquot was reserved as the input fraction, while the remaining 250 µl was processed through columns to obtain the TraPR fraction. Total RNA was extracted from both the input and TraPR fractions, followed by RT-qPCR analysis and sRNA-seq library construction.

Immunoprecipitation of epitope-tagged AGOs

P^AGO1::3XHA-AGO1/ago1-36 (-/-) [28] and p^AGO2::3XHA-AGO2/ago2-1 (-/-) [29] plants were grown in long-day conditions (16 h light/8 h dark) until the inflorescence stems reached at least 15 cm. Three-day-old C. campestris seedlings were attached to the inflorescence stem, with 10 attachments per plant. Cuscuta campestris was allowed to establish and grow for 10 days. Cuscuta campestris tissue was then peeled away and the A. thaliana stem tissue beneath each attachment site was collected. Approximately 400 such stem segments were pooled for each biological replicate. Tissues were ground in liquid nitrogen with a mortar until a fine powder was obtained. The frozen powder was mixed with 1 ml of 4°C IP buffer [10 mM Tris–HCl, pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 0.5% (v/v) Nonidet™ P40 Substitute, RNase Out (Invitrogen™, #10777019), and 1 tablet of cOmplete EDTA-free Protease Inhibitor Cocktail/50 ml (Roche, #11873580001)]. Immunoprecipitation using HA-Trap magnetic agarose (ChromoTek, #atma) was then performed. Lysates were centrifuged at 10 000 × g, 4°C for 10 min, and the supernatant passed through a cell strainer (40 µm, Falcon, #352340). A 200 µl aliquot was reserved for the input fraction, and the remaining lysate was incubated with HA-Trap Magnetic Agarose while rotating end-over-end for 2 h at 4°C. Beads were washed five to six times with 4°C IP buffer. To elute bound HA-AGO complexes, beads were incubated in a dry bath at 95°C for 5 min in nuclease-free water to obtain IP fraction. Total RNA was then extracted from both the input and IP fractions, followed by RT-qPCR analysis and sRNA-seq library construction.

RNA extraction and RT-qPCR to quantify RNAs and miRNAs

Total RNA was extracted using the Quick-RNA Plant Kit (Zymo Research, #R2024), following the manufacturer’s protocol with minor adjustments based on downstream applications. For mRNA and miRNA detection, RNA was not subjected to DNase I treatment. Primers for mRNA detection were designed to span an exon–intron junction, and genomic DNA contamination is not a concern for the stem-loop RT primers used for miRNA detection. For primary miRNA transcript detection, RNA was treated with DNase I (Zymo Research, #E1009-A) and further purified using the RNA Clean & Concentrator Kit (Zymo Research, #R1019). For AGO-bound RNAs (17–200 nucleotides), including miRNAs and U6 snRNA recovered from TraPR and IP experiments, RNA was concentrated with the RNA Clean & Concentrator Kits using the alternative protocol (addition of 1.5 volumes of ethanol prior to column binding) to enrich low-abundance AGO-associated RNAs.

Reverse transcription was performed using the LunaScript^® RT Master Mix Kit (Primer-free, New England Biolabs, #E3025L). For mRNA detection, Random Primer Mix (New England Biolabs, #S1330S) was used as the RT primer. For primary miRNA transcripts, gene-specific RT primers were used due to lack of poly(A) tails. RT reactions followed the manufacturer’s protocol (2 min at 25°C, 10 min at 55°C, 1 min at 95°C, and then held at 4°C). For mature miRNA detection, primer design and RT conditions followed the previously described protocols [30, 31] with minor modifications: 20 min at 16°C, 60 min at 42°C, 5 min at 80°C, and then held at 4°C. qPCR was performed using Luna Universal qPCR Master Mix (New England Biolabs, #M3003). For mRNAs and primary miRNA transcripts, the program followed manufacturer’s protocol (60 s at 95°C, followed by 40 cycles of 15 s at 95°C, 30 s at 60°C, with a standard melting curve from 65°C to 95°C). For mature miRNA, qPCR was run with 30 s at 95°C, followed by 40 cycles of 5 s at 95°C, 15 s at 52°C or 55°C (depending on the Tm of the universal reverse primer), and 10 s at 70°C, with standard melting curve from 65°C to 95°C. Oligonucleotide sequences are listed in Supplementary Table S1.

sRNA-seq analyses

Small RNA libraries were made using the NEBNext Low-bias Small RNA Library Prep Kit (New England Biolabs #E3420L) with NEBNext LV Unique Dual Index Primers (New England Biolabs #E3400, #E3402, #E3404, #E3406, and #E3408). Amplified libraries were evaluated using an Agilent TapeStation device and D1000 Screen Tape. TapeStation-calculated concentrations were used to normalize library concentration during pooling. The pool of completed libraries was purified using a Blue Pippin device (Sage Biosciences) to retain DNA in the 145–195 bp size range and remove any remaining free adapters. The quality and size of the pooled libraries were then assessed with the DNA 5000 Screen Tape on the Agilent TapeStation 4150 (Agilent Technologies), and the concentration was determined via qPCR using the KAPA Library Quantification Kit for Illumina (KAPA Biosystems). The libraries were then sequenced in single-end mode on the Illumina NextSeq 2000 using a P4 50-cycle kit (Illumina) at 1 × 50 nucleotides. The pooled data were then de-multiplexed using unique dual indices to produce FASTQ files for each sample. The FASTQ files were deposited at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) under accessions GSE333156, GSE333470, and GSE333475. The GEO accessions also include the processed data used as source data for the plots in this study (see below). Library details are listed in Supplementary Table S2.

All analyses began by 3′ adapter trimming and read condensation using ShortCut version 2.0 (https://github.com/Aez35/ShortCut; “shortcut” on BioConda). Cuscuta campestris-specific siRNA loci were defined using the “Input” in vitro haustoria libraries. Because these were from samples grown in the absence of any host, all small RNAs must have come from the parasite. The trimmed sRNA-seq data were first aligned to the A. thaliana genome using bowtie [32] (-v 0), and only those reads with zero alignments were retained. The cleaned reads from “Input” in vitro haustoria libraries were used as input to analysis with ShortStack [33] version 4.1.2 using the Cerda et al. [16] genome assembly for C. campestris with default settings. The resulting small RNA cluster annotations were filtered to remove those that overlapped with known or suspected MIRNA loci. Cuscuta campestris siRNAs were defined as those with a “DicerCall” of 21, 22, or 24 and with a strand of “.”.

Analysis of sRNA-seq data from the N. benthamiana attachments began by analyzing C. campestris-specific siRNA accumulation. Trimmed sRNA-seq sequences were first aligned against a telomere-to-telomere N. benthamiana genome assembly [34] using bowtie [32] (-v 0). The sequences that did not have any alignments to the host genome were used as input to a ShortStack analysis using option “--locifile” to quantify sRNA accumulation from the C. campestris siRNA loci defined above. sRNA accumulation from these loci was quantified in units of reads per million 21–24 (RPM21-24; reads/total reads in the 21- to 24-nucleotide range). Analysis of miRNA accumulation from the N. benthamiana attachments started with a list of C. campestris mature miRNAs (no miRNA-stars) curated from the annotations of Cerda-Herrara et al. [16]. This list was classified into categories based on whether the miRNA was canonical or interface-induced and on whether the relevant miRNA family was encoded by both the host and parasite, or just the parasite alone. Trimmed sRNA-seq data were then counted for each mature miRNA, and the counts were normalized to RPM21-24. Median values for each biological replicate were plotted.

Analysis of sRNA-seq data from the TraPR experiment began by curation of a high-confidence set of A. thaliana mature miRNAs. Mature A. thaliana miRNAs from miRBase version 22 were filtered based on (a) strong literature evidence of conservation and function, and/or (b) passing previous annotation efforts [35, 36]. The resulting list was then merged with the set of annotated C. campestris miRNAs [16] to make a non-redundant set. The miRNAs were categorized by canonical versus interface-induced and by whether the miRNA family was present in one or both species. Trimmed sRNA-seq data were then counted for each mature miRNA, and the counts were normalized to RPM21-24. siRNA analysis began by aligning the trimmed sRNA-seq data to the A. thaliana genome using bowtie [32] (-v 0) and retaining only unmapped sequences. The filtered sequences were then used as input to a ShortStack analysis using option “--locifile” to quantify sRNA accumulation from the C. campestris siRNA loci defined above, followed by normalization to RPM21-24. TraPR enrichment for each miRNA or siRNA locus was defined as log₂ (TraPR/Input). Cases where both the input and TraPR RPM21-24 values were <2 were excluded and considered not computable due to low accumulation. In cases where one of the values was zero, and the other was ≥2, a minimal pseudocount was substituted for the zero value to allow log₂ transformation; this value was 0.5 * lowest observed RPM21-24 value in that dataset. Abundance of the TraPR sRNA-seq data was reported as the median RPM21-24 values for the input and TraPR fractions. The initial steps for sRNA-seq analysis of AGO1 and AGO2 immunoprecipitates were identical to those described above for TraPR. Final enrichment scores for each replicate were computed as log₂ (IP/input)^HA-AGO − log₂ (IP/input)^Col-0 to control for any non-epitope-specific precipitation. The median value of the three replicates was plotted.

Results

Trans-species miRNA accumulation extends into adjacent host stems

Cuscuta campestris accumulates trans-species miRNAs specifically at the haustorial interface [17], some of which are detectable in host tissues many centimeters away [37]. To gain quantitative estimates of trans-species miRNA accumulation in various tissues, two representative C. campestris trans-species miRNAs, ccm-miR12480b, which targets SEOR1, and ccm-miR12497f, which targets TIR1, AFB2, and AFB3 [17, 20], were examined by RT-qPCR. C. campestris seedlings were established on the petioles of N. benthamiana, followed by the collection of various specimens (Fig. 1A). As expected, both miRNAs accumulated most strongly at the haustorial interface (Fig. 1B). Notably, substantial amounts of these two miRNAs were also detected in adjacent N. benthamiana petioles, both proximal and distal to haustoria, at ∼10-fold lower levels than at the interface (Fig. 1B). Accumulation dropped sharply at the petiole base, indicating locally limited movement of C. campestris-derived trans-species miRNAs into host tissue. Within the parasite, trans-species miRNAs were detectable at low levels throughout the C. campestris shoot with a decline near the shoot tips. Minimal accumulation was observed in seedlings grown in vitro in the absence of a host (Fig. 1B). These results suggest that C. campestris trans-species miRNAs accumulate beyond the haustorial interface into adjacent host tissues.

Figure 1.

Cuscuta campestris small RNAs accumulate in hosts a short distance from haustoria. (A) Schematic diagram of experiment and sample locations. Host species: N. benthamiana. (B) RT-qPCR analysis of selected trans-species miRNAs. Each dot represents a biological replicate (three per condition, except for C. campestris seedling, which has one sample). Relative accumulation was normalized to haustorial interface by the 2^−ΔCt method. Data available in Supplementary Table S3. (C) sRNA-seq analysis of miRNAs and C. campestris-derived siRNAs. The median reads per million based on 21–24 nucleotide reads (RPM21-24) are plotted by miRNA or siRNA category and tissue sample. Nb: N. benthamiana; Cc: C. campestris. Plotted data are available at NCBI GEO accession GSE333156.

A subset of the specimens from the N. benthamiana–C. campestris experiment was used for small RNA-seq (sRNA-seq). This allowed analysis of the full set of parasite interface-induced miRNAs (which include all known cases of confirmed trans-species miRNAs) and comparison to other categories of parasite small RNAs. Consistent with the RT-qPCR study of two miRNAs, the overall set of C. campestris-derived interface-induced miRNAs was readily detectable in adjacent host petiole tissues (Fig. 1C). Nearly all canonical miRNAs encoded by C. campestris are also widely conserved among flowering plants, including the N. benthamiana host. The high accumulation of these canonical miRNAs in all tissues (Fig. 1C) precludes their use as indicators for inter-species movement. The few known C. campestris-specific canonical miRNAs [16] had very low accumulation levels and could not be reliably quantified. Therefore, C. campestris-derived siRNA loci were annotated and quantified. These loci were carefully screened to exclude any sequences that could possibly have been encoded by the host N. benthamiana genome and classified based on the predominant RNA size (21, 22, or 24 nucleotides). C. campestris siRNAs were often detectable in adjacent host petioles, albeit at very low levels compared to the interface-induced miRNAs (Fig. 1C). This suggests that short-distance movement into host tissues may not be a unique property of interface-induced miRNAs. Nonetheless, the interface-induced miRNAs were present in adjacent host tissue at much higher levels compared to other parasite-derived small RNAs.

Primary transcripts peak 24–48 h before mature miRNAs and do not accumulate in host tissues

Mature C. campestris interface-induced miRNAs were previously shown to accumulate rapidly during the adhesive phase of haustorium development regardless of host presence [18]. We investigated the mechanism controlling this precise temporal and spatial pattern using Arabidopsis thaliana as the host. Four factors together determine the net accumulation of mature miRNA: (i) transcription rate of the primary transcript, (ii) non-dicing degradation of the primary transcript, (iii) dicing rate, which processes primary transcripts into mature miRNAs, and (iv) the mature miRNA degradation rate. It is difficult to determine which factors drive net accumulation, and the observed pattern may result from a combination of all four. As a first step, we measured the levels of primary transcript across tissues and developmental stages that vary in mature miRNA abundance (Fig. 2A). Primary transcript detection used primers spanning the basal DCL1 cleavage sites, ensuring detection of unprocessed primary transcripts (Supplementary Fig. S1).

Figure 2.

Temporal and spatial accumulation of trans-species miRNAs and their primary transcripts. (A) Experimental design. Cuscuta campestris seedlings were either attached to A. thaliana under far-red light–enriched conditions (host-induced haustoria, pink line) or sandwiched with microscope slides and agar under blue light (in vitro haustoria, blue line). Cuscuta campestris seedlings and haustoria from both treatments were collected at 24, 48, 72, 96, and 168 h (black-bordered circles) and 240 h (pink-bordered circles) post-induction. (B) Temporal profiles of primary transcripts (yellow) and their mature miRNA (gray) in host-induced haustoria (left) and in vitro haustoria (right). Shaded boxes indicate the time lag between primary transcripts and mature miRNA peak expression. Each dot is a biological replicate consisting of 5–20 pooled specimens, with three replicates per condition. (C) Spatial analysis of the primary transcripts (yellow) and their mature miRNA (gray). Cuscuta campestris uncoiled stems at the interface, C. campestris cut tips, and adjacent host stems were dissected from a 240-h host-induced haustorium, alongside non-parasitized C. campestris seedlings and A. thaliana control stems. Each dot is a biological replicate consisting of 5–20 pooled specimens, with three replicates per tissue. Expression levels in panels (C) and (D) were calculated using the 2^−ΔΔCt method and 2^−ΔCt, respectively, with C. campestris snRNA U6 as the reference gene for primary transcripts and miR159 for mature miRNAs; the 24-h sample and haustorial interface were used as calibrators, respectively. Data available in Supplementary Table S4.

In addition to the previously measured miRNAs (ccm-miR12480b and ccm-miR12497f), an additional trans-species miRNA, ccm-miR12463a, which targets BIK1, was examined. We found that primary transcripts, in both host-induced haustoria and host-free in vitro haustoria, generally peak 24 to 48 h before the highest accumulation of mature miRNAs (Fig. 2B). This temporal lag suggests that primary transcripts accumulate before mature miRNAs appear. Since the lag occurs in both haustoria types, the control mechanism is likely intrinsic to the parasite’s developmental program rather than dependent on host cues (Fig. 2B). Spatially, primary transcript levels are highest at the haustorial interface, matching the peak of mature trans-species miRNA (Fig. 2C). In C. campestris tips and seedlings, where mature miRNAs were low or undetectable, primary transcripts were also relatively low but still detectable at 25%–50% of the interface level (Fig. 2C). This suggests that in tissues with the potential to form haustoria, such as seedlings and shoot tips, a basal level of primary transcripts is produced and poised for processing into mature miRNA upon haustoria onset.

Although primary transcripts and mature miRNAs are generally positively correlated, an exception occurs in the host stem adjacent to the haustoria, where only mature miRNAs—but not primary transcripts—were detected (Fig. 2C). This pattern was observed for two of the three trans-species miRNAs tested. The exception is ccm-miR12463a, the lowest-expressed of the three miRNAs examined. Mature trans-species miRNAs spread into adjacent host tissue at roughly 10-fold lower level (Fig. 1). Because ccm-miR12463a is already expressed at ∼20- and 200-fold lower levels than the other two miRNAs in the host-induced haustoria (Fig. 2B), its translocated miRNA in the adjacent host stem may have fallen below the detection threshold. Taken together, these data suggest that either duplex miRNA/miRNA* or single-stranded mature miRNAs, rather than their precursors, move into the host. These results indicate that trans-species miRNAs are both transcribed and processed within the parasite using its own machinery.

Parasite DCL1, but not host DCL1, is required for trans-species miRNA accumulation

To further investigate the contribution of parasite versus host machinery in trans-species miRNA production, we targeted a key miRNA biogenesis factor, DCL1, in either the parasite or the host. A reduction of mature miRNAs upon parasite DCL1 knockdown would indicate processing within the parasite, whereas a decrease following host DCL1 silencing would suggest that primary transcripts are translocated and processed in the host. Injection of Agrobacterium tumefaciens into N. benthamiana leaves (“agroinfiltration”) can produce functional artificial miRNAs (amiRs). Previous studies have shown that amiR can move from agroinfiltrated leaves to non-infiltrated leaves, silencing genes in both leaves, likely via the vascular continuum of the petiole and stem [27, 38]. Recognizing the potential of this mobility, we built on the concept of host-induced gene silencing by using agroinfiltration into N. benthamiana host leaves. Agroinfiltration with strains encoding amiR was predicted to result in systemic amiR migration into adjacent petioles. Attachment of C. campestris to such petioles should allow amiR delivery to the parasite.

Having observed that both C. campestris and N. benthamiana encode a single copy of DCL1 (Supplementary Fig. S2), we designed species-specific artificial miRNAs targeting parasite (amiR-CcDCL1) and host DCL1 (amiR-NbDCL1) (Fig. 3A and B) and cloned them into an A. thaliana MIR390a backbone (Fig. 3C). The amiR construct was agroinfiltrated into N. benthamiana, followed by C. campestris attachment to the petiole of the infiltrated leaf, and haustorial tissues were collected after 10 days (Fig. 3A and D). Targeting DCL1 with species-specific amiR knocked down parasite and host DCL1 by 40% and 50%, respectively, while non-targeted DCL1 was unaffected, indicating silencing without detectable off-target effects (Fig. 3E). Cuscuta campestris establishment rates were lower on amiR-CcDCL1 N. benthamiana and this effect was statistically significant in two of five trials (Supplementary Fig. S3 and  Supplementary Table S5). This suggested that the strongest CcDCL1 silencing may impact C. campestris establishment. Our goal was to evaluate the impact of trans-species miRNAs under conditions where CcDCL1 and NbDCL1 were effectively suppressed. To achieve this, we analyzed the top 10 living specimens with the strongest knockdown of CcDCL1 and NbDCL1 for miRNA quantification. As expected, a host canonical miRNA that is not encoded by the parasite genome, nbe-miR403, was dependent on NbDCL1 (Fig. 3F). The three tested trans-species miRNAs decreased significantly with parasite DCL1 knockdown but remained unchanged when host DCL1 was reduced, indicating that parasite DCL1 is required for their biogenesis, whereas host DCL1 is dispensable (Fig. 3F).

Figure 3.

Parasite DCL1, but not host DCL1, is required for trans-species miRNA biogenesis. (A) Workflow for artificial miRNA (amiR)-mediated gene silencing. (B) Alignment of amiRs with their targets. (C) Schematic of the amiR vector. (D) Application of amiR and subsequent RT-qPCR analysis. (E) Expression of CcDCL1, NbDCL1, (F) a canonical host miRNA, and three trans-species miRNAs following agroinfiltration with control amiR (gray), CcDCL1-targeting amiR (yellow), or NbDCL1-targeting amiR (green). Each dot is a biological replicate made from three to six pooled specimens. Expression values were calculated using the 2^−ΔΔCt method. Reference genes were CcRPN7 [39] and NbPP2A [40] for mRNA analysis and Cc 5.8s rRNA and Nb 5.8s rRNA for miRNA analysis. The amiR-Control samples were used as calibrators. One-tailed Mann–Whitney U test was used for pairwise comparisons. Statistical significance is indicated: ***P < .001, **P < .01, *P < .05, ns not significant. Data available in Supplementary Table S6.

Cuscuta campestris trans-species miRNAs accumulate as miRNA/miRNA* duplexes within parasite tissue

Since mature trans-species miRNAs are produced within the parasite, the key question is whether the mobile form that crosses the species barrier is single-stranded, duplex miRNA/miRNA*, and/or AGO-bound. To test this, we examined the ratio of miRNA to miRNA* present in the parasite prior to export. A higher proportion of the miRNA, as seen for most canonical miRNAs, would suggest movement as a single-stranded or AGO-bound form. In contrast, an approximately equal (∼50%) distribution of miRNA and miRNA* would implicate a duplex as the mobile agent. Three replicate small RNA-seq libraries from C. campestris host-free in vitro haustoria were previously shown to accumulate high levels of trans-species miRNAs [18]. These data were analyzed by determining mature miRNA and miRNA* accumulation for each miRNA family and calculating the percentage of mature miRNA relative to the total reads (miRNA + miRNA*) (Fig. 4A). This percentage was significantly higher for canonical miRNAs compared to trans-species miRNAs (Fig. 4B). The median value for trans-species miRNA families was around 50% of mature miRNA, with a large variance. The typical canonical miRNA family, in contrast, had >90% of the reads as mature miRNA with much less miRNA* (Fig. 4B). These findings suggest that trans-species miRNAs predominantly exist as miRNA/miRNA* duplexes within the parasite, raising the possibility that they are not loaded onto parasite AGO proteins and are instead translocated across the host–parasite interface in duplex form.

Figure 4.

Strand ratio and TraPR enrichment of C. campestris miRNAs and siRNAs. (A) Workflow to calculate percent mature miRNA from sRNA-seq libraries from host-free C. campestris in vitro haustoria. (B) Percent miRNA in three sRNA-seq libraries. Each dot represents a miRNA family. The dashed line marks 50%. Mann–Whitney test was used for pairwise comparisons. Statistical significance is indicated: ***P < .001. (C) TraPR workflow. Host species: A. thaliana. A portion of the total RNA was kept as the input fraction. The remaining RNA was incubated with positively charged resin to trap unprotected RNAs. Protein–RNA complexes escape the resin and elute into the TraPR fraction. (D) RT-qPCR analysis of selected RNAs from flanking host stems and in vitro haustoria. Each dot in host stem samples represents pooled RNA from ∼36 A. thaliana stem segments, half from above and half from below the haustorial attachment. Each dot in the in vitro haustoria samples represents pooled RNA from 60–70 C. campestris seedlings. Each condition included 3–6 biological replicates. Relative expression was calculated using the 2^−ΔΔCt (TraPR – Input fraction) method, with miR159 as housekeeping gene. (E) TraPR enrichment from sRNA-seq data. Three biological replicates per tissue are shown separately. TraPR enrichment was calculated as log₂ (TraPR RPM21-24/Input RPM21-24). (F) Accumulation from sRNA-seq data, reported as the log₁₀ of the median RPM21-24 value for each matched TraPR and Input sample. Data for panel (D) are available in Supplementary Table S7. Data for panels (E) and (F) are available at NCBI GEO GSE333470.

Trans-species miRNAs are depleted from AGO-enriched fractions from the parasite but not in host tissues

The approximately equal accumulation of miRNA and miRNA* suggests that trans-species miRNAs exist as miRNA/miRNA* duplexes within the parasite. One mechanism to explain this would be avoidance of loading onto C. campestris AGOs. To test this hypothesis, we used the TraPR (Trans-kingdom, rapid, affordable Purification of RISCs) method. This method relies on positively charged resin that captures free, unprotected RNAs, while protein-bound RNAs pass through and become enriched (Fig. 4C). Two tissue types were analyzed: A. thaliana flanking host stem segments adjacent to C. campestris attachment sites and host-free in vitro haustoria induced by blue light for 96 h. At this stage, trans-species miRNAs are already produced at detectable levels comparable to those in host-induced haustoria (Fig. 2B). These samples contain only C. campestris tissue, derived from seedlings that never encounter a host, providing a naïve environment where only parasite AGOs can interact with the trans-species miRNAs.

In both tissues, the known AGO1-bound miR159 (which is encoded by both the parasite and the host) was enriched, while non-AGO-associated RNAs such as 5.8s rRNA and U6 snRNA were depleted (Fig. 4D), confirming that TraPR purification worked as expected. The three trans-species miRNAs examined by RT-qPCR were strongly enriched in flanking host stems but depleted in in vitro haustoria (Fig. 4D). These results suggest that these three trans-species miRNAs are not bound to C. campestris AGO proteins in in vitro haustoria but do become AGO-bound after export into the host.

sRNA-seq from the TraPR experiments allowed a global view of all types of C. campestris miRNAs and siRNAs. Low levels of TraPR enrichment were unique to the entire population of interface-induced miRNAs (which include all those confirmed as trans-species) in in vitro haustoria; all other endogenous C. campestris small RNAs had much higher enrichments in in vitro haustoria (Fig. 4E). TraPR enrichment values were much higher for interface-induced miRNAs in flanking host stems compared to in vitro haustoria (Fig. 4E). Consistent with the N. benthamiana experiment (Fig. 1C), the abundance of interface-induced miRNAs was quite high in flanking host stems, while the abundance of C. campestris-derived siRNAs was very low (Fig. 4F). The low abundance prevented the computation of TraPR enrichment values for most C. campestris siRNAs in flanking host stems (Supplementary Fig. S4). For the minority of abundant siRNAs that could be analyzed, TraPR enrichment values were high in flanking host stems, suggesting that all types of C. campestris small RNAs can be AGO-associated within host tissues (Fig. 4E). Collectively, these data suggest that all the C. campestris interface-induced miRNAs are uniquely un-enriched by TraPR specifically in host-free in vitro haustoria. This suggests that they are selectively prevented from AGO association in the parasite. We conclude that C. campestris trans-species and interface-induced miRNAs are selectively prevented from loading onto parasite AGOs.

Cuscuta campestris interface-induced miRNAs can associate with host AGO1 and AGO2

After establishing that C. campestris trans-species and interface-induced miRNAs are AGO-bound only in host tissues, the next step was to determine which host AGO loads them. To test this, epitope-tagged A. thaliana AGOs were enriched by immunoprecipitation, and associated RNAs were examined. Ccm-miR12463a, ccm-miR12480b, and ccm-miR12497f have all been shown to direct slicing of A. thaliana mRNAs [17, 20], and each has a 5′-U. These features are consistent with AGO1 loading. AGO2 was also of interest because of its role in plant defense [41, 42] against viruses and bacteria. C. campestris was attached to A. thaliana lines expressing epitope-tagged AGO1 or AGO2, as well as to wild-type plants. The parasite was then removed after 10 days, and AGO immunoprecipitation was performed using specimens collected from the host stem beneath the attachment site (Fig. 5A). AGO1-bound ath-miR166 and AGO2-bound ath-miR390 and ath-miR408 were enriched in their respective lines, confirming assay robustness (Fig. 5B). All three C. campestris trans-species miRNAs were strongly enriched in AGO1 pulldowns, with much weaker signals in AGO2 pulldowns (Fig. 5B). These results indicate that these three C. campestris trans-species miRNAs preferentially loaded on host AGO1 over AGO2.

Figure 5.

Cuscuta campestris interface-induced miRNAs can be loaded on host AGO1 or AGO2. (A) Cuscuta campestris was attached to three A. thaliana backgrounds: Col-0 (WT), 3XHA-tagged AGO1, and 3XHA-tagged AGO2. After 10 days, miRNAs were quantified from total lysate (Input) and immunoprecipitated (IP) samples using RT-qPCR or sRNA-seq. (B) RT-qPCR result. Top: Control miRNAs are enriched only in their respective IPs, confirming specificity. Bottom: Trans-species miRNAs are strongly enriched in AGO1 IPs. Each dot represents a biological replicate consisting of ∼400 pooled A. thaliana stem segments beneath parasite attachment. Each condition included 3–4 biological replicates. Relative expression was calculated using the 2^−ΔΔCt (IP – Input fraction) method. 3HA-AGO1 served as the calibrator for ath-miR166 and all three trans-species miRNAs, while 3HA-AGO2 was used for ath-miR390 and ath-miR408. (C) Enrichment of miRNAs in AGO1 immunoprecipitations by 5′ nucleotide and miRNA category. (D) Enrichment of C. campestris siRNAs in AGO1 immunoprecipitations by miRNA category. (E) Enrichment of miRNAs in AGO2 immunoprecipitations by 5′ nucleotide and miRNA category. (F) Enrichment of C. campestris siRNAs in AGO2 immunoprecipitations by miRNA category. Data in panels (C–E) are median values of three biological replicates and are available from NCBI GEO GSE333475. Data from panel (B) are available in Supplementary Table S8.

sRNA-seq was performed to allow a comprehensive analysis of AGO1- and AGO2-associated small RNAs made by C. campestris that are present in infested A. thaliana stems. Interface-induced miRNAs with a 5′-U were enriched in AGO1 immunoprecipitations at levels comparable to canonical 5′-U miRNAs (Fig. 5C). In contrast, interface-induced miRNAs with a 5′-A were enriched in AGO2 immunoprecipitations (Fig. 5E). This reflects the known preferences of the 5′ nucleotide for A. thaliana AGO1 and AGO2 [43]. Similar to the TraPR results, most C. campestris siRNA loci could not be analyzed for AGO1- or AGO2-enrichment because of low accumulation in the host tissue (Supplementary Fig. S5). For the small numbers that could be analyzed, 21- and 22-nucleotide-dominated siRNA loci had very slight AGO1 enrichment and strong AGO2 enrichment (Fig. 5D and F). In contrast, there was no systematic AGO1 enrichment of C. campestris siRNAs from 24-nucleotide-dominated loci (Fig 5D) and a slight AGO2 enrichment (Fig. 5F). These results indicate that incoming C. campestris interface-induced miRNAs are loaded onto host AGO proteins according to their 5′ nucleotide sorting preferences. Loading onto host AGO proteins is not unique to interface-induced miRNAs, but the incoming C. campestris siRNAs are present in host tissues at dramatically lower levels (Supplementary Fig. S5).

Discussion

Our working model of trans-species miRNA biogenesis and function in C. campestris is summarized in Fig. 6. Transcription of C. campestris trans-species miRNA primary transcripts in haustoria precedes accumulation of mature miRNAs by 24–48 h. Primary transcripts are processed by C. campestris DCL1. The resulting trans-species miRNAs avoid AGO loading inside of C. campestris and instead remain in miRNA/miRNA* duplex form until export. Upon arriving in host cells, they are loaded onto host AGO proteins, where they function to direct mRNA slicing and secondary siRNA biogenesis of targets. This mechanism allows C. campestris to produce a class of mobile miRNAs that accumulate to high levels in host tissue that are somehow pre-programmed for export, avoiding self-targeting, and only gaining function once inside the host.

Figure 6.

Biogenesis, export, and function of C. campestris trans-species miRNAs. Schematic summarizing our current model of miRNA biogenesis, export, and functions in the parasitic plant C. campestris. Trans-species miRNAs are produced by parasite DCL1, selectively avoid loading onto parasite AGO proteins, remain as duplexes until export, and only become functional in the host using host AGO proteins.

Trans-species miRNAs accumulate most strongly in haustoria but are also present at ∼10-fold lower levels in adjacent host tissues, spreading in both directions from the haustorium. Their accumulation drops sharply at the basal petiole, suggesting limited long-distance mobility in the host. Detection of trans-species miRNAs in adjacent host tissues allowed study after export, complementing host-free in vitro haustoria where miRNAs exist before export. Both systems provided single-organism contexts for investigation: in vitro haustoria without a host and host tissue without parasite.

Primary transcripts peak 24–48 h before their corresponding mature miRNAs and accumulate most strongly in haustoria, matching the highest mature miRNA levels. This suggests transcriptional regulation in both temporal and spatial dimensions. However, miRNA accumulation also reflects processing, AGO loading, turnover, and decay rates. Thus, higher mature miRNA abundance could arise from stronger transcription, more efficient dicing, or reduced turnover of transcripts that are constitutively produced at a basal level. Indeed, primary transcripts remain detectable across stages and tissues, including seedlings and shoot tips where mature miRNAs are absent, suggesting ongoing basal transcription. The upstream sequence element (USE) identified in most trans-species MIRNA loci is very similar to the constitutive U6 snRNA promoter [18], consistent with a model where loci are transcriptionally “on” by default, while downstream regulatory layers refine mature miRNA accumulation through processing or degradation. Biologically, this default “on” mode may benefit C. campestris by maintaining a basal pool of precursors that primes haustoria-competent tissues—such as seedlings and shoot tips—for rapid maturation of miRNAs once host contact initiates haustorial development. In A. thaliana stems near the attachment, mature miRNAs are present without detectable primary transcripts, supporting parasite origin and direct transfer rather than host processing.

Cuscuta campestris trans-species miRNAs are produced by parasite DCL1 but appear to avoid loading by parasite AGOs. Incoming miRNAs with a 5′-U were enriched in host AGO1 immunoprecipitates, while interface-induced miRNAs with 5′-A were enriched in AGO2 immunoprecipitates. Based on these results, it appears that loading onto host AGOs follows the typical sorting rules by 5′ nucleotide. Many C. campestris interface-induced miRNAs have a 5′-C [18]; some but not all of these were enriched in AGO1 immunoprecipitates. Arabidopsis thaliana AGO5 primarily binds 5′-C RNAs and associates with both viral-derived siRNAs and some miRNAs [44–46]. Thus, we consider host AGO5 as another possible candidate for the binding of some C. campestris trans-species miRNAs.

Cuscuta campestris interface-induced miRNAs were found to accumulate to relatively high levels in host tissues adjacent to haustoria. However, other types of C. campestris small RNAs were also observed to accumulate in adjacent host tissues, albeit at very low levels. Cuscuta campestris-derived siRNAs were found to be TraPR-enriched and sometimes enriched in host AGO immunoprecipitates. Thus, entry into host tissues and engagement with AGO proteins once there is not a unique property of the interface-induced miRNAs. Migration of small RNAs into host tissues may be non-selective. What is unique about the interface-induced miRNAs is their very specific and high accumulation in haustorial tissue and their avoidance of self-AGO-loading. This may lead to a much more effective concentration once they enter host tissues.

The difference between canonical miRNAs and trans-species miRNAs was striking: Within host-free in vitro haustoria, canonical C. campestris miRNAs are AGO-bound, while trans-species miRNAs are not. What accounts for this selective avoidance of AGO-loading by C. campestris trans-species miRNAs? Covalent nucleotide modifications are one possibility. Pseudouridine in miRNAs and siRNAs is associated with mobility [47], while 5-methylcytosine modifications within mRNAs correlate with mRNA transport across graft junctions [48]. Similarly, tRNAs are heavily modified, and both full-length tRNAs and tRNA-derived fragments can move systemically through the phloem [49]. Addition of tRNA-like structures to non-mobile mRNAs can increase their systemic transport across grafting junctions [50] as well as into C. campestris from hosts [51]. The Cuscuta-host vasculature resembles a natural graft, and mechanisms that facilitate RNA transfer across graft junctions may also support trans-species movement. Cuscuta campestris trans-species miRNAs are transcribed from Pol III promoters [18], which are also used by tRNAs. An alternative hypothesis is that Pol III transcription itself prevents AGO loading in the nucleus of the source cells. Canonical MIRNA primary transcripts are Pol II products, and their transcription and processing are intimately connected with Pol II cofactors, such as Mediator [52]. Rapid DCL1 processing and coupled AGO-loading occur in the source nucleus for canonical plant miRNAs. AGO-loaded small RNAs are immobile and cell-autonomous [53], while miRNAs that escape nuclear AGO loading are more likely to be mobile [54]. Pol III transcription of interface-induced miRNAs may break the tight connection between MIRNA transcription, processing, and AGO loading, such that interface-induced miRNA/miRNA* duplexes escape the source nucleus unbound to AGO.

These observations invite comparison with other cases of mobile small RNAs, both within a single plant and between plants and their associated organisms. Similar to the C. campestris system, Botrytis cinerea (gray mold fungus) sRNAs (Bc-sRNAs) are also processed by parasite DCLs and, after arriving in the host, are bound to host AGO to effect gene silencing [12]. However, the B. cinerea AGO also binds mobile Bc-sRNAs and contributes directly to infection [55]. This seems to contrast with the avoidance of “self”-AGO loading we observed in C. campestris. Long- and short-distance mobility of siRNA-mediating gene silencing within a single plant has been tied to the mobility of siRNA duplexes [53]. In this model, AGO proteins are not mobile, and once a given siRNA duplex is disassembled for AGO-loading, mobility ceases. This paradigm closely parallels our observations in C. campestris, where trans-species miRNAs appear to accumulate as AGO-free miRNA/miRNA* duplexes within parasite tissues. Whatever the AGO-avoidance mechanism is, it may enable the C. campestris trans-species miRNA/miRNA* duplexes to easily enter the host at much higher concentrations compared to other types of small RNAs. Importantly, we cannot rule out the hypothesis that movement of trans-species miRNAs is occurring in association with some other non-AGO proteins; the hypothesis that movement occurs as free miRNA/miRNA* duplexes remains inferential at this point. If they are migrating as free duplexes, this could enable future crop protection strategies based on specific sequestration of short duplexes.

Cuscuta campestris trans-species miRNAs largely avoid parasite AGOs and are likely exported as duplexes for selective loading onto host AGOs. The biological benefit seems clear—avoiding self-targeting, which is particularly important given that parasitic plants and their plant hosts share greater sequence similarity than fungal pathogens and their host. A previous study [20] demonstrated sequence dissimilarity between Cuscuta and host transcripts as one safeguard, but the complementarity scores are not always beyond the threshold of functionality, perhaps leaving some residual risk of self-targeting. Selective exclusion from parasite AGOs could therefore provide an additional layer of protection while at the same time encouraging movement into adjacent host tissues. The mechanism of selective AGO avoidance is a key unresolved question: it may involve differences between parasite and host AGO proteins, RNA modifications, differences in co-transcriptional AGO-loading between Pol II and Pol III transcripts, or other undiscovered mechanisms. These questions define clear directions for future investigation.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by an award from the United States National Science Foundation [grant number 2003315] to M.J.A. Funding to pay the Open Access publication charges for this article was provided by The Pennsylvania State University. 

Data availability

Small RNA seq data (both raw reads and the processed data used to generate the figures) are available at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) under accessions GSE333156, GSE333470, and GSE333475. All other plotted data are incorporated into the article and its online supplementary material.