WSMV colonies were maintained on Karl 92 wheat. New wheat, at the 2-leaf stage and Feeke’s scale 2–3, was infested using a snip of wheat containing at least 20 wheat curl mites from an established plant.

TriMV colonies were maintained on RonL wheat, but otherwise treated the same as the WSMV.

BYDV was not cultivated and was only found in naturally-infected field-grown wheat plants.

Naturally infected wheat was grown at the USDA Conservation and Production Research Laboratory in Bushland, TX. These fields were planted to hard red winter wheat cv. Karl 92 in late September, at a seed rate of 31 kg/acre. Urea was applied preplant at a rate of 45 kg/acre, and Ally herbicide was applied pre-plant at 28 g/acre. Fields were irrigated as needed during the season with a center pivot sprinkler system. Wheat was harvested in late June with a John Deere 9500 combine mid- to late July. Samples 8–44 were grown in the field.

Positive controls for WSMV and TriMV (Samples 1–7) infected wheat came from greenhouse grown plants used to maintain the colonies described previously. Because these plants are used to maintain the vectors, no information about the infection stage of the samples was maintained.

Healthy control plants were grown in the same greenhouse in a Bugdorm to keep insects out.

A representative leaf of each infection status can be found in Figure S1.

Copy number was determined by conducting a Standard Curve Absolute Quantification test. This test requires cDNAs derived from the viral RNAs. Reverse transcription was first performed using Qiagen’s Omiscript RT kit and prescribed manufacturer’s protocol.

For the WSMV test, each reaction mix contained 1x Taqman Fast Advanced Master Mix (Applied Biosystems), 0.3 µM forward primer Wsm F 9075, 0.3µM reverse primer Wsm R 9196, and 0.25µM Wsm P 9142 FAM-Probe (Tatineni et al., 2010). 20 ng of template cDNA was added to each reaction for all three virus assays. For the TriMV test, reaction mix concentrations were the same as for the WSMV test. Forward primer Tri F 9754, reverse primer Tri R 9875, and Trip 9802 FAM-Probe (Tatineni et al., 2010) were used. The BYDV primers and probe were designed in our lab with the Thermofisher Custom Taqman^® Assay Design Tool, using sequence accession number EF521836-1: forward primer BYDV F2910, sequence 5’GAAGGAGGCGCCGTAGAA 3’, reverse primer, BYDV R2949, sequence 5’CCTGCTCGATTGGGTTGGA 3’ and probe, BYDV2928, sequence FAM 5’ACCACTGGCCGAACTG 3’ MGB-NFQ. The BYDV primers and probe can be requested from ABI, Assay ID APXGRYP, 20x Custom Taqman^® Gene Expression Assay BYDV and used 1x per reaction. The Real-time PCR thermal cycling profile was the default program: 95°C for 20 s followed by 40 cycles of denaturing at 95°C for 1 s then annealing at 60°C for 20 s.

WSMV, TriMV, and BYDV virus copy numbers in leaf samples were determined using plasmid DNA standards. Plasmid DNA for all three viruses were synthesized by Invitrogen (Thermo Fisher, Pleasanton, CA). A 300 bp WSMV insert was synthesized from accession number NC001886 at position 9001-9300. A 300bp TriMV insert was synthesized from accession number FJ669487 at position 9661-9960. A 300bp BYDV insert was synthesized from accession number EF521836-1 at position 2801-3100. The vector was the commercial pMA-T plasmid. The plasmid size was 2674 bp. Sequence identity within the insertion sites was 100% as checked by Invitrogen. Approximately 5 µg plasmid DNA was synthesized. The website “DNA Copy Number and Dilution Calculator” (Thermo-Fisher) was used to calculate copy number dilutions. A stock 0.5 ng/µl dilution was made with RNAse free water, then standard serial dilutions of 10⁷, 10⁶, 10⁵ 10⁴, 10³, 10², and 10 copies/µl were made. Standard curve quantification real-time tests were run on prepared samples.

The results of virus quantification are reported in Table S1. A threshold copy number of 17 was set to identify samples as being positive for a viral infection. We observed substantial variation in copy number across our samples. In some cases, we had below 1000 copies of WSMV (samples 26–32), whereas others showed over 1 billion copies (samples 15 and 18). For the purposes of this proof-of-concept manuscript, we only accounted for the binary positive-or-negative assignment in assigning our spectra to categories for the PLS-DA and ANOVA analyses. Further studies should explore how copy number influences the Raman spectra.

We sought to determine whether RS coupled to statistical methods could be used to differentiate wheat plants infected with different combinations of viruses. We first employed partial least squares discriminant analysis (PLS-DA), a supervised classification method, to distinguish between healthy, W and B-infected plants. We built a model, Healthy-Single-Single (HSS), using 58 spectra from the healthy, B and W datasets. Spectra were truncated to the range of 500–1700 cm^-1, area-normalized to the 1440 cm^-1, then interval partial least squares (iPLS) was used to select the 228 wavenumbers that resulted in the highest cross-validation performance. HSS, containing nine latent variables (LVs), was found to be on average 90% accurate in differentiating among healthy and single infection wheat as shown in the cross-validation confusion table (Table 3). The model explained 99.4, 0.27, 0.07% of the variation in the first three latent variables, respectively. This suggests that RS in combination with chemometrics can enable a highly accurate prediction of wheat disease status in single virus scenarios.

Next, to identify potential markers of disease, we conducted ANOVAs on the intensities at bands associated with the metabolites described above. First, we searched for bands that could uniquely identify healthy wheat. We found that the bands at 747 and 1218 cm^-1 were significantly more intense in the healthy wheat spectra than spectra of any infected type. Additionally, we identified several additional bands, 1186, 1326, 1455, 1601, 1630, and 1678 cm^-1, where the healthy spectra show overlap with only one type of infection. These sets of Raman bands could potentially be useful markers to identify healthy plants among diseased ones.

We then sought to determine whether Raman markers exist for specific disease combinations in terms of WSMV (shortened to W). We found that while W could not be identified uniquely from any ANOVA, bands at 747, 1051, 1218, 1455 cm^-1 showed a significant difference between the W infection and one of its coinfections. The two double coinfections, WT and WB, could be distinguished from each other by almost all bands analyzed, including 747, 1000, 1051, 1186, 1218, 1288, 1326, 1455, 1601, and 1630 cm^-1. In all except 1455, 1601 and 1633 cm^-1, WT is significantly more intense than WB. Finally, the triple infection, WBT, could be uniquely identified by the 1186 cm^-1 band and shows overlap with only one other combination at 1218 cm^-1. These results indicate that RS can identify biochemical markers of different WSMV-associated wheat diseases.

BYDV (shortened to B) alone, like W, could not be uniquely identified using ANOVA. However, at 1051 cm^-1, B, WB and WBT clearly separated from healthy, W and WT, suggesting this band could be a generic marker of infections that include B. Additionally, the bands at 1455, 1601, and 1630 show significant differences between the B combinations and WT. While no band enables differentiation of B and WB, the single infection could be differentiated from WBT by bands at 1156, 1186, 1218, and 1525 cm^-1. Using this large number of marker bands across different infections, RS can be used to enable differentiation of WSMV, BYDV, and TriMV-associated wheat diseases.

While these observed differences could be used for disease diagnostics, they are also indicative of biochemical changes under different infecting conditions. We observed that in general, WT was significantly more intense than WB. As described previously, TriMV is known to enhance the symptoms of WSMV. Our results suggest that perhaps BYDV weakens the symptoms of WSMV or suppresses them with its own symptoms. The fact that we do not observe WSMV coinfections changing in tandem as is seen in BYDV coinfections (1051 cm^-1) provides further evidence that BYDV may alter the biochemical responses elicited by WSMV. Additionally, we found that across most bands examined with ANOVA, WBT showed the lowest intensity. This may indicate that the combination of all three viruses places greater strain on the host metabolism, lowering all relevant metabolite concentrations. The only bands that do not show this trend are 1601 and 1630 cm^-1, the phenylpropanoid bands. The buildup of the compound in the infected system is reasonable, as it is associated with cell wall hardening. These observations show the potential of RS for examining coinfecting pathogens on a pathogen-to-pathogen level.

Through RS, we identified marker bands that are associated with major players in the infection cycle. 1186 and 1218 cm^-1 are associated with xylan, a major cell wall component in grasses such as wheat. Changes observed in this band may suggest strengthening or weakening of the cell wall. We observed that this band was more intense in the spectra of healthy plants than all infected classes except B, though WT was very close. This observation suggests that the cell wall (through xylan degradation) is weakened in the presence of many of these viral infections. Additionally, we found that in the spectra collected from WBT-infected wheat, these bands were less intense than the corresponding bands in all other classes (WT, WB…), suggesting that triple infection may put a greater burden on xylan maintenance than other combinations of viruses.