Exemption for review by the ethical committee system and informed consent was given by the Committee on Biomedical Research Ethics-Capital region in accordance with Danish law on assay development projects.

All authors comply with the data availability policy.

Within the European Network for Diagnostics of Imported Viral Diseases (ENIVD) we gathered a wide range of emerging viruses, as inactivated culture supernatants or as purified viral DNA or RNA (Table 1). Viruses were inactivated by heat and/or gamma-irradiation, or by suspension in an RNA-extraction reagent (TRIzol, Life Technologies; TriFast, Peqlab; AVL buffer, Qiagen) [27]. The majority of viruses were grown in Vero E6 cell-cultures (kidney epithelial cell line derived from African green gonkey) (ATCC CRL-1586), but poliovirus (PV) was grown in L20B cells (a murine recombinant cell line) [28]. We also used six control samples from the QCMD EQA programme for 2010 and 2013 (WNV10-01, WNV10-07, WNV13-01, WNV13-10, WNV13-11 and DENV13-01). The WNV13-01 sample contained West Nile virus (WNV) at a concentration of 1.0×10⁷ copies/ml and the DENV13-01 sample contained Dengue virus (DENV) type 1 at a concentration of 1.0×10⁶ copies/ml. The WNV10-01 and WNV13-10 samples contained a mixture of flaviviruses (DENV type 1, 2 and 4, and Japanese encephalitis virus (JEV)). The WNV10-07 and WNV13-11 samples contained a mixture of DENV type 3, tick-borne encephalitis virus (TBEV) and yellow fever virus (YFV), each at a concentration of 1.0×10⁶ copies/ml.

We used clinical samples received for routine diagnostic analysis at Statens Serum Institut (SSI), Copenhagen, Denmark (Danish National reference laboratory (ISO 17025; 2005)), from the CCH Fever Project bio-bank at the Swedish Institute for Communicable Disease Control (Sweden), and from the Department of Microbiology, Aristotle University of Thessaloniki (Greece). The samples were (Table 2): i) One parapoxvirus-positive skin lesion (blister) sample from the hands of a shepherd; ii) One Chikungunya virus-positive serum sample from a traveller hospitalized for Dengue-like symptoms (high fever, joint pain, rash) after visiting Thailand; iii) Eight DENV-positive serum samples from travellers experiencing mosquito bites in the jungle of Thailand, iv) One CCHFV-positive serum sample (from the CCH Fever program); v) One sandfly fever Toscana virus-positive cerebrospinal fluid (CSF) sample from a traveller hospitalized with meningitis after visiting Toscana, Italy; vi) Six WNV-positive urine samples from patients hospitalized with West Nile fever (two of them with encephalitis). In addition, we used six hepatitis C virus (HCV)-positive serum samples and five HCV-positive plasma samples. One of the HCV-positive serum samples had a known viral concentration (1.2×10⁶ IU/ml) determined by standardisation against the WHO control. As negative controls we used virus-negative clinical samples (urine, CSF and serum).

As previously described [19] we centrifuged 230 µl of sample at 17,000 g for 10 min, filtered the supernatant through a 0.22 µm Spin-X spin filter (Costar) and treated it with DNase (Invitrogen or New England Biolabs) for 30 min-1½ h. The viral nucleic acid (NA) was extracted using the PureLink Viral RNA/DNA kit (Invitrogen), without the addition of carrier RNA. All samples were treated with this protocol with the exception of the QCMD panel samples, CSF, urine, and plasma samples, which were not DNase treated. Virus-positive supernatants suspended in RNA-extraction reagent were purified according to the manufacturer’s instructions (TRIzol, Life Technologies; TriFast, Peqlab; AVL buffer, Qiagen). The resulting RNA was further purified using the QIAamp RNA viral Mini kit (Qiagen). The extracted viral NA was eluted with 30–50 µl DNase/RNase-free water, and stored at −20°C or immediately used.

Reverse transcription (RT) on purified viral RNA was performed with three different methods: i) The P-N6/SSIII method, which uses the Superscript III Reverse Transcription kit (Invitrogen), combined with 5′-phosphorylated random hexamers (P-N₆) (Eurofins MWG Operon). Briefly, 11–12 µl viral RNA was mixed with 1 µl 10 mM dNTP mix and 1 µl 250 ng/µl P-N₆, incubated at 85°C for 5 min, and cooled on ice. Next, 4 µl 5x first strand buffer, 1 µl 0.1 M DTT, 1 µl RNaseOUT (40 U/µl) (optional) and 1 µl Superscript III RT enzyme (200 U/µl) was added, and the sample mixed and incubated at 25°C for 10 min, 42°C for 60 min and 95°C for 5 min. ii) The RT-reaction included in the WTA kit (Qiagen), which uses T-Script reverse transcriptase combined with random and oligo-dT primers. RT was performed according to the manufacturer’s instructions. iii) The VILO method, which uses a cDNA Synthesis kit (Invitrogen) containing Superscript III reverse transcriptase combined with random primers. The method was performed as previously described [19], [29]. The samples were stored at −20°C or immediately used.

For viral RNA amplification we used the WTA method [29] with the QuantiTect WTA kit (Qiagen), except for the reverse transcription step that was replaced by one of the three RT methods described above. We also modified the protocol by performing amplification at 30°C for 2–8 h. We purified Repli-g amplified DNA according to the supplementary protocol, using the QIAamp DNA Mini Kit (Qiagen), and validated its purity and concentration using a NanoDrop spectrophotometer (Thermo Scientific). The DNA was stored at −80°C or immediately used. To avoid contamination between samples, we adopted precautions normally used during routine viral diagnostic PCR analysis at SSI, where extraction, amplification and analyses are physically separated and negative samples are included in all steps.

The technique used for routine diagnostic virus analysis at SSI is quality-assured real-time PCR (ISO 17025; 2005, SSI). To confirm presence of virus in the samples and quantify the virus before and after WTA, we performed virus-specific real-time PCR. We used in-house assays for DENV, WNV, orthopoxvirus, parapoxvirus, Usutu virus, Hantaan virus, Toscana virus, BK virus (BKV) and rotavirus A; and previously published assays for JC virus (JCV) [30], cowpox and monkeypox viruses [31], Chikungunya virus [32], Eastern equine encephalitis virus (EEEV) [33], JEV [34], TBEV [35], YFV [36], Lassa virus and CCHFV [37], Dobrava-Belgrade virus (DOBV) [38], Puumala virus [39], Rift Valley fever virus (RVFV) [36] and Marburg virus [40]. PCR was performed using an Mx3005P (Stratagene) thermal cycler. We calculated the fold difference in concentration from the ΔC_t obtained from real-time PCR before and after WTA, combined with dilution factors. Here we made the assumption that 1 cycle change in C_t-value was equivalent to a doubling of target DNA. We estimated the sample concentrations of the HCV-positive, DENV-positive and WNV-positive clinical samples by performing a series of 10-fold dilutions of the HCV-positive serum sample (1.2×10⁶ IU/ml), the DENV13-01 QCMD sample (1.0×10⁶ copies/ml) and the WNV13-01 QCMD sample (1.0×10⁷ copies/ml), under the assumption that no viral NA was lost during purification.

We analysed samples with the LLMDAv2 microarray, developed at the Lawrence Livermore National Laboratory (LLNL), USA and described elsewhere [14], [19], [41]. The LLMDAv2 contains 388,000 oligonucleotides probes designed from all sequenced viruses and bacteria [14]. Labelling and microarray hybridization was performed according to manufacturer protocols (Gene expression analysis, Roche NimbleGen) with the exception that 8 µg, instead of 2 µg, of labelled material was used for hybridization. Microarray data was analysed using a simple Excel-based data analysis method developed at SSI (SSI analysis) as described previously [19]. Since the SSI analysis is not optimized for bacteria, any bacterial hits were excluded from the results. Non-human, non-zoonotic pathogens were also excluded since they are assumed to be clinically irrelevant in a diagnostic setting. Additional data analyses were performed on the samples using the CLiMax software developed at LLNL and described elsewhere [14], [41].

Microarray data were submitted to the Gene Expression Omnibus (GEO) database http://ncbi.nlm.nih.gov/geo/with the accession number GSE55576. All microarray data used in this study are MIAME compliant.

To enable successful microarray identification of virus in clinical samples, we have previously used the Phi29 polymerase-based WTA method (Qiagen) [19], [29]. The WTA protocol includes three sequential reactions: a reverse transcription reaction to generate cDNA, ligation of cDNA fragments into large linear chains, and amplification by the Phi29 polymerase [29]. To assure an efficient ligation, we replaced the included RT reaction with Superscript III and 5′-phosphorylated random hexamers (P-N₆) hereafter called P-N6/SSIII. This was done in order to phosphorylate the 5′-end of the cDNA fragments so that new phosphodiester bonds could be formed during the ligation step [42], [43]. We compared this method to the manufacturers RT reaction (T-Script using random and oligo-dT primers) and to RT using Superscript VILO cDNA kit [19], [29]. Prior to RT and amplification, samples were pre-treated according to a previously described protocol [19]. The different RT protocols were tested in parallel on 10-fold serial dilutions of an HCV-positive serum sample (1.2×10⁶ IU/ml) (Figure 1A), on two supernatants containing the hantaviruses Puumala virus and DOBV, respectively (Figure 1B), and on 10 HCV-positive clinical samples with varying viral concentrations (Figure 1C–1D). For all samples tested, whole transcriptome (WT) amplification of cDNA generated by P-N6/SSIII was more efficient than VILO or T-Script. Therefore, the P-N6/SSIII RT-reaction was used for all further WT amplifications.

WT amplification using the P-N6/SSIII RT-method was tested for its ability to amplify emerging viruses. Due to difficulty in getting access to clinical samples positive for a diverse set of emerging viruses, we initially tested the method on a wide range of virus-positive cell culture supernatants (SN), purified viral NA or QCMD panel samples (Table S1). The WT amplification was analysed using virus specific real-time PCRs before and after amplification (Table S1 and Figure 2). For all samples tested, amplification of the emerging virus was observed (Figure 2A). For EEEV (Alphavirus), Usutu virus (Flavivirus), WNV (Flavivirus), PV (Enterovirus), Hantaan virus (Hantavirus), RVFV (Phlebovirus) and Toscana virus (Phlebovirus) the amplification was relatively small with a fold increase between 25–500 (Table S1 and Figure 2A). However, for other samples much larger fold increases were observed, such as JEV (Flavivirus) with a fold increase of 1.5×10⁶, DOBV (Hantavirus) with a fold increase of 4.5×10⁶ and Puumala virus (Hantavirus) with a fold increase of 1.4×10⁴ (Table S1 and Figure 2A). When we examined the relationship between amplification (fold increase) and viral content (C_t-values before WT amplification) (Figure 2B), we observed a significant correlation between WT amplification and viral content. Samples containing a high viral content were amplified to a lesser extent than samples containing a lower viral content, which could reflect that for samples with a high concentration of NA, primers and nucleotides are depleted quickly, resulting in a lower WT amplification.

To investigate whether the presence of several viruses in a sample would interfere with WT amplification, we tested two QCMD panels of samples (WNV10-01 and -07) containing mixtures of four and three different flaviviruses, respectively, each at a concentration of 1.0×10⁶ copies/ml (Table S1 and Figure 2A). All seven flaviviruses were WT amplified; however DENV-1 was amplified to a lower degree than to the other DENV subtypes or flaviviruses (Figure 2A). This most likely reflect a difference in the sensitivity of the Dengue subtype-specific primers used to analyse the WT amplification rather than virus subtype-specific variation in the WT amplification. In summary, the modified WT amplification method was able to amplify emerging viruses in 21 different non-clinical samples.

To test the sensitivity of the previously described LLMDA microarray [14], [19], [26] for RNA viruses, we performed microarray analysis on a 10-fold dilution series of a HCV-positive serum sample (1.2×10⁶ IU/ml), a RVFV-positive supernatant (3.3×10⁶ copies/ml), a WNV-positive QCMD panel sample (WNV13-01) (1.2×10⁷ copies/ml), a DENV-positive QCMD panel sample (DENV13-01) (1.0×10⁶ copies/ml) and two QCMD panel samples (WNV13-10 and WNV13-11) containing mixtures of JEV, DENV-1, DENV-2, DENV-4 and YFV, DENV-3, TBEV respectively (1.0×10⁶ copies/ml) (Table 1). Dilutions ranging from 10⁶ to 10² copies/ml were WT amplified, labelled and hybridised to the LLMDAv2 microarray. Microarray analysis was performed using the SSI [19] and CLiMax data analysis methods [14], [41] (data not shown).

For the HCV, RVFV and WNV samples, dilutions of 10⁶ to 10³ copies/ml yielded sufficient viral material for successful identification by the LLMDAv2, while dilutions of 10² copies/ml were not detected by the microarray (Table 1). Dilutions of 10² copies/ml will theoretically result in an input of 24 copies to the RT-reaction and 12 copies to the WTA-reaction. Analyses of the viral concentrations after WT amplification of the non-detectable 10² copies/ml dilutions showed that RVFV, HCV and WNV were amplified to 1.7×10⁵, 4.3×10⁵ and 4.4×10⁵ copies/ml, respectively (Table 1).

The detection limit for DENV, JEV and TBEV was 10⁴ copies/ml (Table 1) and the detection limit for YFV was higher (10⁵ copies/ml) than the rest of the flaviviruses analysed. The WNV13-11 sample was documented as containing YFV, DENV-3 and TBEV, each at 1.0×10⁶ copies/ml; however, analysis of the C_t-values of YFV and TBEV before amplification showed a higher value for YFV (C_t = 27) compared to TBEV (C_t = 23) (Table 1), which could indicate a lower viral content of YFV in the WNV13-11 sample than was documented. Analysis of the viral concentration after WT amplification of the non-detectable 10³ copies/ml dilutions showed that DENV, JEV and TBEV were amplified to 3.1×10⁵, 1.0×10⁴ and 2.1×10⁴ copies/ml, respectively (Table 1). From this we conclude that at least 10³ copies/ml is needed for a successful amplification with the modified WTA method and at least 10⁵ copies/ml is needed after WT amplification in order to reliably identify viruses with the LLMDAv2. This concentration is equivalent to 0.17 femtomolar, demonstrating exquisite sensitivity of the LLMDAv2 platform.

The LLMDA microarray [14], [19], [26] was tested for its ability to correctly identify a wide range of virus-positive cell culture supernatants (SN), purified viral NA or QCMD panel samples containing emerging viruses (Table 2). The WT amplified samples previously described (Table S1) were labelled and hybridised to the LLMDAv2 microarray. Microarray analysis was performed using the SSI [19] and CLiMax data analysis methods [14], [41].

In all 21 samples analysed, both methods identified the correct virus (Table 2). In more than half of the samples, human endogenous retroviruses (HERV) were also found (Table 2), consistent with the presence of human host DNA. The CLiMax method identified additional retroviruses such as baboon endogenous virus (BEV), simian retrovirus 1 (SRV-1), Mus dunni endogeneous virus (MDEV), murine leukemia virus (MuLV) and mouse mammary tumour virus (MMTV). These additional viruses were not identified by the SSI method because non-human, non-zoonotic pathogens were considered clinically irrelevant and excluded in the SSI data analysis.

For four of the samples (cowpox virus, monkeypox virus, PV-1/PV-2 and Usutu virus), the SSI method had difficulties in distinguishing between different genus-members and subtypes. In the Usutu virus sample (Flavivirus), the SSI method identified both Usutu virus and JEV, another Flavivirus species, as being present, while the CliMax analysis correctly identified Usutu virus only. In the PV sample, the CLiMax analysis correctly identified PV-1 and PV-2, while the SSI analysis made an additional false-positive detection of PV subtype 3 (Table 2). In the samples of cowpox virus and monkeypox virus, both methods identified additional members of the Orthopoxviridae family as present. The SSI analysis identified both samples as mixtures of cowpox, monkeypox, vaccinia and variola viruses, while the CliMax analysis identified the correct cowpox or monkeypox virus together with the variola minor virus (Table 2), which belongs to the same genus. Detailed examination of the probes with positive signals (greater than the 99^th percentile of the negative control intensities) showed that all such probes with alignments to the variola minor virus genome had strong matches in the cowpox and monkeypox genomes; so that the identification of variola minor virus in the CliMax analysis in these samples is most likely due to cross-hybridization of these probes.

For five samples (Hantaan virus, Puumala virus, RVFV, Naples virus and Marburg virus) the CLiMax analysis identified additional viruses that were not observed using the SSI analysis (Table 2). To better understand the source of these additional predictions, we used the CliMax software to generate sequence-probe alignment plots, where the intensity of each probe is plotted against its alignment position in the viral genome. These plots clarify whether identification of a virus is based on presence of the whole genome or may be due to cross-hybridization from probes matching sub-regions of other genomes present in the sample. For example, the sequence-probe alignment plots for the Puumala virus sample show the positive probes to be uniformly distributed across all three Puumala virus genome segments, indicating the presence of the whole viral genome (Figure 3B, top). Probe hits for the bovine viral diarrhea virus 1 (BVDV-1) genome show a different pattern, landing in only a narrow region suggesting nonspecific- or cross-hybridisation (Figure 3B, bottom). We refer to this pattern as a fragmented alignment plot.

We observed a similar fragmented alignment pattern for RVFV segment S in the Marburg sample, indicating non-specific cross-hybridisation (data not shown). In contrast, we obtained uniform positive probe distributions for mammalian orthoreovirus 1 and 3 (MRV) genomes in the Hantaan virus sample, for CCHFV and Simian virus 5 (SV5) genomes in the RVFV sample, and for BVDV-1 in the Naples virus sample, indicating that these complete viral genomes were truly present (data not shown). CCHFV specific PCR could not confirm the presence of CCHFV in the RVFV sample (data not shown). The other additional findings were all considered clinically irrelevant and therefore not further investigated by PCR.

The presence of several viruses in a sample did not interfere with identification, as can be seen by the microarray analysis of the two panels of samples (WNV10-01 and -07) containing mixtures of different flaviviruses (Table 2). Microarray analysis correctly identified all four viruses present in WNV10-01 and all three viruses present in WNV10-07 (Table 2). However, the individual DENV subtypes were difficult to distinguish completely. In the WNV10-10 sample the CLiMax analysis identified DENV type 3, and in the WNV10-07 sample both analysis methods detected DENV type 1 and 2. These extra DENV findings were later confirmed as false-positives by Dengue subtype-specific PCR (data not shown). In addition, the SSI analysis of the WNV10-07 sample identified Omsk haemorrhagic fever virus (OHFV), which also belongs to the Flavivirus genus [44]. This finding was not observed using the CLiMax analysis and hence not checked by PCR. The CLiMax analysis also found HERV, rotavirus A and porcine rotavirus C in both samples as well as BVDV-1 in WNV10-01 and SV5 in WNV10-07. The presence of rotavirus A was confirmed by rotavirus A-specific PCR (data not shown). BVDV, SV5 and porcine rotavirus C were considered clinically irrelevant and therefore not confirmed by PCR.

In summary, the LLMDAv2 correctly identified single and multiple viruses present in non-clinical samples with a very low level of false positive signals. The CLiMax analysis method identified every virus present in the samples whereas the simpler SSI analysis method only identified clinically relevant human pathogens.

We tested the LLMDAv2 microarray on 18 clinical samples previously identified as positive by real-time PCR for emerging viruses. The correct virus was identified in 17 samples using both the SSI (Table 3) and CLiMax analyses (data not shown). The sample identified only as a parapoxvirus was determined to be Orf virus, a member of the Parapoxvirus genus. Seven of the eight DENV-positive samples were clearly determined by the microarray analysis to be positive for DENV type 2, DENV type 1 or DENV type 3. DENV type 4 was not identified in any of the clinical samples. Additional DENV subtypes were detected in four of the samples, but at very low probe signal intensities compared to the correct DENV subtype probe signal (Figure 4A+4B). These were confirmed as negative by Dengue subtype-specific PCR (data not shown). One DENV-positive sample was also positive for hepatitis GB virus C (GBV-C). One DENV-positive sample was not identified by the microarray. Six urine samples were positive for WNV and two of these samples were identified as having additional viruses (Table 3). One WNV sample was also positive for the polyomaviruses JCV and BKV (Figure 4C), which later were confirmed as present by PCR (data not shown). Another WNV sample was positive for JEV (Figure 4D), but this finding could not be confirmed by PCR (data not shown). In addition, the microarray detected HERV in almost all samples, consistent with the presence of human DNA, and the common Torque Teno virus (TTV) [19], [23], [45] in the CCHFV and two DENV samples. Virus-negative urine, CSF and serum were also analysed and confirmed to be negative for virus (Table 3), except for HERV found in the CSF sample. In summary, the LLMDAv2 correctly identified emerging viruses present in 17 of the 18 clinical samples analysed. The only sample not identified was a DENV-positive sample, in which the viral concentration was determined to be below the detection limit, as described below.

To assess viral concentration in clinical samples, we performed specific real-time PCR before and after WT amplification. We estimated the viral concentration of 6 WNV-positive urine samples and 6 DENV-positive serum samples by comparison to PCR results for the series of 10-fold dilutions of the QCMD panel WNV and DENV samples (Table 2 and Table 3). The WNV-positive urine samples were determined to have concentrations between 3.7×10⁴ and 4.9×10⁵ copies/ml before WTA and concentrations between 3.0×10⁵ and 2.8×10⁷ copies/ml after WT amplification (Table 3). These samples all had concentrations above the detection limit (10³ copies/ml) determined for dilutions of the WNV-positive QCMD sample (WNV13-01) (Table 1). Analysis of the WT amplification showed that WNV from urine samples was not amplified as efficiently as WNV from the QCMD sample (Table 1 and Table 3), however the concentration after WTA was still above 10⁵ copies/ml and hence detectable by the LLMDAv2.

The DENV-positive serum samples were determined to have concentrations between 1.1×10⁴ and 1.8×10⁷ copies/ml before WT amplification and between 1.4×10⁵ and 2.5×10¹² copies/ml after WTA (Table 3). The DENV-positive sample which was not detected by the microarray had an estimated concentration of 1.1×10⁴ copies/ml, which was near the pre-amplification detection limit seen for dilutions of the QCMD DENV sample (10⁴ copies/ml) (Table 1 and Table 3); and a concentration after WTA of 1.4×10⁵ copies/ml, which is near the post-WTA limit of detection (10⁵ copies/ml). This sample was also near the limit of detection with real-time PCR, with a C_t value of 37 before amplification. In summary, 11 out of 12 clinical samples analysed had viral concentrations above the detection limit of the LLMDAv2.