Microarray profiling predicts early neurological and immune phenotypic traits in advance of CNS disease during disease progression in Trypanosoma. b. brucei infected CD1 mouse brains

Human African trypanosomiasis (HAT), also known as sleeping sickness, is a major cause of mortality and morbidity in sub-Saharan Africa. We hypothesised that recent findings of neurological features and parasite brain infiltration occurring at much earlier stages in HAT than previously thought could be explained by early activation of host genetic programmes controlling CNS disease. Accordingly, a transcriptomal analysis was performed on brain tissue at 0, 7, 14, 21 and 28dpi from the HAT CD1/GVR35 mouse model. Up to 21dpi, most parasites are restricted to the blood and lymphatic system. Thereafter the trypanosomes enter the brain initiating the encephalitic stage. Analysis of ten different time point Comparison pairings, revealed a dynamic transcriptome comprising four message populations. All 7dpi Comparisons had by far more differentially expressed genes compared to all others. Prior to invasion of the parenchyma, by 7dpi, ~2,000 genes were up-regulated, denoted [7dpi↑] in contrast to a down regulated population [7dpi↓] also numbering ~2,000. However, by 14dpi both patterns had returned to around the pre-infected levels. The third, [28dpi↑] featured over three hundred transcripts which had increased modestly up to14dpi, thereafter were significantly up-regulated and peaked at 28dpi. The fourth, a minor population, [7dpi↑-28dpi↑], had similar elevated levels at 7dpi and 28dpi. KEGG and GO enrichment analysis predicted a diverse phenotype by 7dpi with changes to innate and adaptive immunity, a Type I interferon response, neurotransmission, synaptic plasticity, pleiotropic signalling, circadian activity and vascular permeability without disruption of the blood brain barrier. This key observation is consistent with recent rodent model neuroinvasion studies and clinical reports of Stage 1 HAT patients exhibiting CNS symptoms. Together, these findings challenge the strict Stage1/Stage2 phenotypic demarcation in HAT and show that that significant neurological, and immune changes can be detected prior to the onset of CNS disease.


Introduction
At least 70 million people in 36 countries throughout sub-Saharan Africa are at risk from the neglected tropical disease human African trypanosomiasis (HAT) commonly termed sleeping sickness. [1]. The condition results from infection of the morphologically indistinguishable extracellular haemoflagellate protozoan parasites, the West African form Trypanosoma brucei gambiense and the East Africa type T. brucei rhodesiense transmitted by the bite of the teste fly (Glossina sp). The disease progresses through two distinct clinical stages [2]. During the haemolymphatic Stage1, the trypanosomes invade and replicate in the blood, lymphatics and peripheral organs leading to a systemic inflammation with a variety of non-specific clinical symptoms. After a variable period, which is more prolonged in the gambiense form, trypanosomes invade the CNS initiating the encephalitic Stage 2 of the disease characterised clinically by protean neurological features and pathologically by the presence in the brain of trypanosomes, macrophages, lymphocytes, cytokines and chemokines in the CSF [3] and activation of microglia and astrocytes [4]. This exacerbates the cytokine mediated blood brain barrier (BBB) breakdown thereby amplifying the inflammatory response with a worsening debilitating neuropsychiatric impairment and the disruption of circadian rhythm control giving rise to major alterations of sleep structure, the hallmark CNS presentation [5]. If not diagnosed accurately, which is essentially reliant on CSF analysis due to the lack of robust clinical staging biomarkers, sleeping sickness is invariably fatal. As clinical investigation studies are severely hampered for numerous logistical reasons including ethical considerations and the very limited availability of post-mortem material, unravelling the neuropathogenesis of HAT has been and remains essentially dependent on a range of animal models, particularly the rodent paradigm [6,7]. The well characterized human clinical phenotype is readily recapitulated in a range of rodent models infected with variety of trypanosome stabilates including the human resistant species T. b. brucei. The model employing the CD1 outbred mouse strain infected with the T. b. brucei GVR35 stabilate, has been widely used for over 30 years in molecular pathogenesis studies and provides the focus of the current investigation [8,9]. In these various studies, the consensus has been that up to 21dpi, most parasites are essentially restricted to the haemolymphatic system, beyond which, the infection can no longer be effectively treated with Stage-1 drugs. If not then treated with Stage-2 drugs, the mice succumb to a combination of neurological and immunosuppressive complications resulting in death.
As recently reviewed [10], it had been thought that parenchymal invasion was preceded by the initial infection of the choroid plexus and the circumventricular organs via the fenestrated endothelial layer of the blood vessels in these regions. However, a series of experiments has challenged this strict Stage1/Stage2 demarcation. Fluorescently labelled T. b. brucei and T.b. rhodesiense parasites were detected in the parenchyma just a few hours post-infection [11] while in a rat model study [12] parasites and T cells were observed in the parenchyma at 9dpi. An updated re-examination of the CD1/GVR35 model was undertaken using qPCR, Contrast Enhanced Magnetic Resonance Imaging (CE-MRI) and histopathology. The study confirmed the presence of trypanosome DNA in brain homogenate at 7dpi, and cellular neuropathology and a significantly amplified CE-MRI signal at 14dpi, each criterion increasing incrementally at later time points [13].
These rodent model findings in conjunction with reports of CNS symptoms in Stage1 rhodesiense HAT patients suggest that BBB impairment may be more progressive and begin earlier than had been previously envisaged [14]. However, it should be emphasized that when and by what route(s) trypanosomes invade the CNS remains an ongoing and contentious issue. In fact, the presence of parasites in the parenchyma has been challenged with the suggested alternative of a restricted location in the pial space of the meninges following breakdown of the blood-CSF barrier [15] although at a later stage of infection, the trypanosomes may enter the glia limitans leading to encephalitis and death. Despite the wealth of genetic information accrued from targeted gene studies our understanding of the molecular mechanisms of the two critical events in HAT neuropathogenesis, the host's dysregulated immune response and parenchymal invasion are poorly understood.
As reviewed [3,[16][17][18], the products of the Tlr9, Tlr2 and Myd88 genes have been shown to be critical determinants in the initiation of the innate response via activation the NFκB complex and subsequent expression of genes encoding the inflammatory cytokine mediators TNFα, IFN α/β and IFN-γ which in turn induces the expression of a panel of the chemokine genes most notably Cxcl9, Cxcl10 Cxcr3 and Ccl5 the encoded products of which are integral to trans-endothelial leucocyte migration. This pro-inflammatory assault is counter balanced by the activity of several Th-2 anti-inflammatory molecules most notably IL6, IL10 and TGFβ-1and an ensuing immune suppression phenotype with devastating neurological consequences.
Due to the inherent structural and interconnected complexities of the endothelial neurovascular unit (NVU) [19], the adult quiescent BBB, though highly regulated is also fragile and susceptible to a range of insults leading to a variable degree of vascular leakage [20]. Despite an early dye study reporting BBB damage [21], to date, there is no convincing evidence of any permanent structural disruption of the paracellular complex of the BBB to facilitate trypanosome migration in rodent models [22,23]. Other proteins mechanistically involved in trypanosome migration in rodent models are laminins 4 and 5 [24] and a small cohort of ECM metalloproteinases [25]. A leucocyte-trypanosome coupled neuroinvasion model mediated by the establishment of an IFNγ induced CXCL10 gradient in the astroglial endfeet of the NVU has been proposed [26] while reduced migration was observed for both trypanosomes and T cells into the parenchyma of TNF-α and IFN α/β null mice [18]. Treatment of cultured endothelial cells with the trypanosome encoded cysteine protease brucipain can challenge the molecular integrity of the parenchymal basement membrane resulting in transient perturbation of calcium signalling [27,28].
The undoubted success of targeted studies has expanded our knowledge of the molecular genetics underpinning the HAT clinical phenotype including the assembly of the African trypanosomiasis ID 5143 KEGG pathway (S1 Fig) [29] and its 38 participating genes. However, the full complement of differentially expressed genes that drive phenotypic change await identification. Notwithstanding well documented limitations, a microarray approach offers an unbiased experimental stratagem to monitor simultaneous differential gene expression during disease progression. Somewhat surprisingly, microprofiling of T.b.brucei infected CNS material has been restricted hitherto to two investigations by Amin and colleagues [26,30] using the C57BL/6 inbred mouse infected with the T.b brucei stabilate AnTat 1.1E utilizing an older glass slide based microarray technology. The first [26], was essentially a targeted study on the expression of a panel of chemokine genes between the two pairings (6-15dpi) and (6-28dpi). The latter study was more global in approach [30], quantifying the total number of differentially expressed genes with a GO annotation from the three temporal pairings (6-15dpi), (6-28dpi) and (15-28dpi).
We report here the outcome of a microarray transcriptome analysis performed on whole brain tissue collected at 0, 7, 14, 21 and 28dpi from the CD1/GVR35 HAT mouse model using the commercial Illumina MouseWG6_V2_R3-11278593_A chip. Over 4000 differentially expressed genes were subjected to a combined KEGG pathway and GO term functional enrichment analysis. In addition, the transcriptome database was interrogated to assess the expression profile over the 28day timeline of documented and candidate trypanosomiasis genes. We surmise that by whatever neuroinvasion route(s) and mechanism(s) used, the presence of trypanosomes, wherever they reside in the CNS compartment, elicits changes in the host's transcriptome providing genetic correlates with clinical phenotypic traits. We hypothesise here that the recent findings [11][12][13] of neurological features and parasite brain infiltration occurring at much earlier stages in HAT rodent models than previously thought might be explained by early activation of host genetic programmes controlling CNS disease. Our data shown below suggest that this is indeed the case.

Ethics statement
All animal procedures were authorised under the Animals (Scientific Procedures) Act 1986, Home Office Licence PPL60/4414 and approved by the University of Glasgow Ethical Review Committee

Mice and infections
Innocula of 2 x 10 4 of T.b.brucei of the cloned stabilate GVR35 were intraperitoneally injected into adult female CD1 mice (Charles River Laboratories). Infected CD1/GVR35 mice develop a fluctuating parasitaemia and by 21dpi the mice cannot be treated with Stage 1 drugs. Mice chosen for this study were all healthy littermates of similar age and weight. Three replicate infected mice groups were euthanized and perfused at 7, 14, 21 and 28dpi in addition to a group of four non-infected control mice equivalent to 0dpi as previously described [8]. At 7dpi the animals were screened for the presence of trypanosomes by examination of wet blood films. All mice were found to be parasitaemic.

Microarray analysis
Excised brains weighing around 500mg were immediately homogenized in 5ml RNABee (amsbio) and stored in 1ml aliquots at 80˚C for subsequent microarray and end-point RT.PCR analyses. Total cellular RNA was prepared from infected and control brain (0dpi) replicates according to manufacturer's guidelines. Briefly, the RNA samples were DNase treated (Ambion) and column purified using a RNeasy MiniEluate Kit (Qiagen). The integrity of the samples was assessed on an Agilent 2100 Bioanalyzer which provides an entire electrophoretic trace of the RNA sample that evaluates the extent of degraded ribosomal species expressed as a RIN value between 0 and 10. Multiple RNA preps were performed for each sample to ensure (3 x 7dpi), (3 x 14dpi), (3 x 21dpi), (3 x 28dpi) and (4 x non-infected controls (0dpi)) with RIN values � 8.5 were available for custom microarray analysis outsourced to FIOS (Fios Genomics UK). Using the Ambion Illumina Total RNA Amplification Kit, the RNA samples were converted into cDNAs and subjected to T7 in vitro synthesis to generate biotinylated cRNA and hybridized to 50 oligomer probes printed on the MouseWG6_V2_0_R3_11278593_A Bead-Chip and stained with streptavidin-Cy3. A single WG6 Chip contains six whole genome microarrays each encoding 45,281 probes representing the 22,000 genes or so that comprise the mouse genome as estimated by UniProt as from August 2020. Raw data were transformed using a variance stabilizing transformation (VST) method prior to normalisation across all arrays using the robust spline normalisation (RSN) method. Relative transcriptional activity was converted to log 2 expression units. From the full complement of the 45,281probeset, 23,213 were detected at least once across the arrays, with an expression hybridization signal intensity above the fluorescence background corresponding to a log 2 value � 6.3, were retained for subsequent differential gene expression analysis.
Sixteen arrays to accommodate (3 x 7dpi), (3 x 14dpi), (3 x 21dpi), (3 x 28dpi) and (4 x 0dpi) samples underwent a QC check using the QualityMetrics Bioconductor package [31]. Arrays were scored using the four parameters, maplot, boxplot, heatmap and manual inspection where 15 arrays passed QC and a non-infected control (0dpi) failed. A total of 10 single and/or multi-factor Comparisons were analysed using linear modelling. Subsequently, empirical Bayesian analysis was applied using the Bioconductor package limma [32] and corrected for multiple testing using the Benjamini and Hochberg method [33]. For each Comparison, the null hypothesis was that there was no difference between the groups being compared. Comparisons were examined to identify expression differences at a strict statistical level (adjusted pvalue <0.001). Based on this stringent threshold, significant probes in each Comparison were assessed using a hypergeometric test for functional enrichment (p<0.05) of KEGG pathways and GO terms (p<0.05) based on their annotation information.

End-point RT.PCR analysis
End-point RT.PCR was performed on the same source of total cellular RNA samples that were earmarked for microarray experimentation. Reactions were carried out in the linear amplification range and message levels corresponding to the genes of interest were expressed relative to the activity of the mouse housekeeping gene Cyclophilin. In brief, and as detailed elsewhere [34], total cellular RNA was converted to cDNA using SuperScript III (Thermo Fisher Scientific-UK). PCR reactions were performed on 5ng of cDNA using RedTaq Ready Mix (Sigma-UK) against gene of interest primer sets. Thermal cycling parameters were, an initial denaturation step of 94˚C/5 mins, a core cycle comprising (94˚C/1 min)/ (55˚C-65˚C/1 min)/ (72˚C/1 min) for 25-35 cycles was followed by a final extension of (72˚C/10 mins. PCR products were separated by TAE gel electrophoresis, visualised by ethidium bromide staining and quantified by densitometry using an UVIdocD55XD documentation system (Uvitec UK).
CD1 mouse brain RNAs collected in triplicate at 0, 7, 14, 21 and 28dpi. Biotin labelled cRNA was hybridized to the Bead Chip and probed with streptavidin-Cy3. After a series of quality control checks, only probes detected at least once across the arrays were retained for subsequent analysis. Following normalization to minimise variation gene expression measurement, 23,213 probes were designated as transcriptionally active, corresponding to over 11,000 genes. This figure represents a brain message complexity of around 50% similar to that reported from other mouse brain microarray studies but significantly less than the true mRNA complexity considered to be around 80% [35]. This disparity arises in the sensitivity of microarray detection as many CNS genes are expressed at low levels and the use of whole brain homogenate as the source of mRNA can masquerade regional brain expression differences. S1 Table lists this collection of normalized transcripts presented in alphabetical order alongside the log 2 expression unit value for each of the four time points for each replicate group and the three noninfected control samples equivalent to 0dpi. Ten pair-wise Comparisons were performed on this active probeset. A Comparison nomenclature that specifies the two different time point combinations was adopted. The Infected Series compared the non-infected control equivalent at 0dpi against the four infected time points denoted (0-7dpi) 1 , (0-14dpi) 2 , (0-21dpi) 3 and (0-28dpi) 4 . The Temporal Series measured activity changes between six different infected time point combinations viz (7-14dpi) 5 , (7-21dpi) 6 , (7-28dpi) 7 , (14-21dpi) 8 , (14-28dpi) 9 and (21-28dpi) 10 . Expression difference for all Comparisons was set at a strict statistically significant adjusted p value <0.001 with an attendant adaptive threshold fold change with associated significance statistics.
The resultant transcriptome database (S2 Table) comprises a dataset for each of the ten Comparison is set out in order of decreasing adjusted p values and corresponding fold changes were subjected to i) differential gene expression analysis, ii) KEGG (S3 Table) and GO (S4 Table) functional enrichment analysis and iii) the database (S2 Table) was interrogated to assess expression profiles of candidate trypanosomiasis genes (S9-S12 Tables)

Differential gene expression analysis
Quantitative transcriptome analysis. Inspection of the volcano plots (Fig 2A and 2B) which chart differential gene expression against fold change, with the numbers of up-regulated transcripts depicted in red and down-regulated samples in blue, reveal a dynamic quantitative transcriptome over the 28 day timeline unveiling two key observations. First, the 7dpi Comparisons in both Infected (Fig 2A) and Temporal Series (Fig 2B) had the greatest activity differences relative to all other Comparisons. Secondly, there was no significant differential gene expression in the (0-14dpi) 2 and (14-21dpi) 8 pairings. Transcript data for the Infection Comparison (0-7dpi) 1 (Fig 2A) confirmed that by 7dpi the levels of 8145 mRNAs of the normalized probeset of 23217 were altered approximating to around 4000 genes. 4360 transcripts had increased and 3785 were down-regulated. The absence of any significant differential gene expression activity in the (0-14dpi) 2 Comparison suggests that by 7dpi or earlier, one message population was up-regulated while the activity of the other had decreased, but by 14dpi or before, mRNA levels of both populations had readjusted to approximately those of the noninfected control (0dpi). This clearly demonstrates that the interval between 0 and 14dpi marks a period of intense transcriptional activity. However, it should be recognized that the host's response may in part, be a consequence of a transitory systemic response to the initial trypanosome infection. Message activity levels in the (0-21dpi) 3 and (0-28dpi) 4 pairings with corresponding values of (37/5) and (324/30) are indicative of a post-14dpi period of increased transcription (Fig 2A). This cyclical activity between 0 and 14dpi followed by a post-14dpi induction was also evident in the Temporal Series profiles (Fig 2B).
When total transcript numbers were plotted against FC � 2 (Table 1), only around 30% of the total transcript population of the 7dpi pairings (highlighted in shade), had a ± FC � 2 compared to the higher values of the post-14dpi Comparisons such as 65.86% for (14-28dpi) 9 . This infers that transcriptional activity between 0 and 14dpi is less expansive when measured by FC. This besots high-throughput transcriptome studies as to the weighting to be given between p values and corresponding FC as the true index of biological significance. This was highlighted in a zebrafish heart tissue microarray study where the application of different FC reported different hypoxia outcomes [36]. Accordingly, we adopted a strict statistical significance of an adjusted p value of <0.001 and an adaptive FC threshold with its own associated significance statistics. This implies that although around 60% have a FC < 2.0 in (0-7dpi) 1 their activity may well be biologically significant.  4 ] and the Temporal Series [(7-14dpi) 5 (7-21dpi) 6 (7-28dpi) 7 (14-21dpi) 8 (14-28dpi) 9 (21-28dpi) 10 ] generated ten transcriptome datasets which were subjected to three levels of analysis viz i) Differential gene expression analysis, ii) KEGG/GO Functional enrichment analysis and iii) Candidate gene interrogation. Qualitative transcriptome analysis. Comparisons (0-7dpi) 1 , (7-14dpi) 5 , (0-28dpi) 4 and (14-28dpi) 9 that capture key significant quantitative transcriptome changes over the 28 day timeline (Fig 2) were selected for a limited qualitative analysis of the statistically most significant 20 genes for each Comparison (S5 Table) and of the 25 transcripts with the highest FCs across all Comparisons ( Table 2).
As depicted in shade (S5 Table), 15 of the 20 (0-7dpi) 1 down-regulated genes were up-regulated in the (7-14dpi) 5 Comparison. The shared (0-7dpi) 1 and (7-14dpi) 5 qualitative profiles, although phenotypically disparate, could be grouped into a range of functional categories including modulation of expression (Luc7l3, Prpf18, Pdcd4, and Gpbp1), neuronal activity  Table 1. Total transcript population plotted against the number of statistically significant mRNAs with � 2 FC reveals variation in the level of gene activity when measured by p value differential gene expression and � 2FC. (Gria2, Cdkl2, Ggnbp2 and Ccdc174) pleiotropic signalling cascades (Cacnb4, Ccpg and Pak3) and mitochondrial activity (Dnm1l and Mrpl50).). This contrasts with the exclusive immune content of (0-28dpi) 4 and (14-28dpi) 9 sharing eight common genes shown in shade (S5 Table) comprises the innate response (Saa3), IFN induced Interferon Stimulated Genes (ISGs) (Ifitm1, Ifitm3, Gm4951and Gbp3), chemokines (Cxcl9, Ccl4, Ccl5, Ccr5 and Cxcl16), complement activity (C3), the adaptive response Cd3g including antigen presentation (Cd74, H2-K2 and H2-Q8), dysfunctional immunity (Aif1) and the bacteriolytic enzymatic activity (Lyz1 and Lyz2). Examination of the 25 transcripts with the highest FCs across all Comparisons (Table 2) unveiled a similar demarcation between an early neuronal dominated phenotype and the post-14dpi immune dominated phenotype. All down-regulated genes were (0-7dpi) 1 entries. This included changes to the gene regulatory machinery (Zranb and Eif5,), mitochondrial activity (Dnm1), signalling cascades (Camk4, Dgkb and Mfap3l), neurotransmission (Gria2, Kif5c, Slc17a6, Selt and Unc13c) and neuronal activity (Slc24a2 and Myo5a). The inclusion of the Plp1 gene (corroborated by end-point RT.PCR) is difficult to explain as it encodes the major protein of the CNS myelin sheath. In addition, the significance of the imprinting entries Copg2os2 and Meg3 is difficult to evaluate. Except for Csrp1 (neuronal development) and Cop-g2os2 (imprinting) all other up-regulated mRNAs were found in (0-28dpi) 4 and immune exclusive. This featured the antigen presenting genes (Cd74, H2-Ab1and H2-K2). Additional functional groupings comprised complement genes (C3 and C4a), IFNγ ISGs genes (Igtp and Gbp2), the IFNα,β ISG (Irf1) and the chemokine genes (Cxcl9, Cxcl13 and Ccl5) in addition to Cxcl10 which has been strongly implicated in trypanosome invasion [26]. Increased levels of Fcgr4 mRNA encoding a Fc gamma receptor is indicative of IgG activity. Temporal gene expression patterns. Most of the significantly active transcripts could be grouped into one of, or variations of, four broad generic temporal gene expression patterns denoted [7dpi"], eg Snhg11, [7dpi#], eg Sfxn3, ([28dpi"], eg Cxcl9) and [7dpi"-28dpi"], eg Laptm5 (Fig 3). The first two, define the functionally disparate phenotype between 0 and 14dpi with respective peak and minimum expression at 7dpi. However, it should be appreciated that in many cases and in contrast to the highly stylized patterns represented by Snhg11 and Sfxn3, many examples showed more gradation in the message levels between 14dpi and 28dpi particularly in the [7dpi"] population. The [28dpi"] profile represented by the chemokine gene Cxcl9, is characterized by an initial and variable change in message levels between 0 and 7dpi followed by a distinct post 14dpi inducible period, typically with attendant high fold changes describes the immune dominated Comparisons (0-28dpi) 4 and (14-28dpi) 9 (Tables 2 and S5). The fourth, [7dpi"-28dpi"] represent mainly immune genes, have a characteristic twin peak pattern of similar levels as represented by the lysosomal-associated protein transmembrane 5 gene (Laptm5). As the profiles are generated from quantified message abundance at five fixed time points, they cannot measure the extent or otherwise of the temporal length of up-regulated plateaus and repressed troughs. For example, peak levels in [7dpi"] could feasibly plateau from 3dpi to 8dpi or equally from 6dpi to 11dpi. Accordingly, these patterns should be viewed more as useful expression trends.

Functional enrichment analysis
To gain insight of the phenotype predicted from the qualitative transcriptome database (S2 Table), a KEGG [29] and GO [37] functional enrichment analysis of all Comparisons was undertaken as detailed in S3 and S4 Tables respectively. Due to the GO annotation selection criteria, the number of GO terms far exceeds KEGG pathways entries (Table 3) confirming the dominant down-regulated pattern of (0-7dpi) 1 followed by the subsequent up-regulation in (7-14dpi) 5 and the post-14dpi induction in the (0-28dpi) 4 and (14-28dpi) 9 pairings.
Enrichment profiles for all Comparisons collated in heatmaps (S2-S5 Figs) catalogue the statistically most significant 75 KEGG and GO entries. This confirmed two distinct phenotypes which are boxed in each heatmap between the functionally diverse neuronal dominated phenotype represented in (0-7dpi) 1 and (7-14dpi) 5 and the post-14dpi immune dominated Comparisons (0-28dpi) 4 and (14-28dpi) 9 share an almost identical up-regulated KEGG and GO heatmap fingerprint (S2 and S4 Figs). There was no significant KEGG or GO enrichment of down regulated genes outwith the 7dpi Comparisons. The S2 Fig pathways, Oocyte meiosis ID 4114 , Prostate cancer ID 5215 and Pancreatic cancer ID 5212 do not have a CNS phenotype. This, an inherent issue of enrichment analysis termed "crosstalk" by Donato [38] arises from common genes participating independently in different pathways and GO terms. A more detailed KEGG analysis supplemented when appropriate with GO data, was performed to compare the phenotypes between (0 and 14dpi) and (14 and 28dpi). KEGG pathway identifiers, sourced from S3 Table were grouped into a range of functional categories reflecting phenotypic heterogeneity over the 28 day timeline (S6-S8 Tables) are summarized in the KEGG bar graphs (Figs 4 and 5) plotting ten key pathways against p values for each Comparison.
Phenotypic analysis between 0 and 14dpi. As the up-regulated profile of (0-7dpi) 1 mirrors the contents of down-regulated pathways of (7-14dpi) 5 and vice versa, KEGG analysis between 0 and 14dpi was restricted to the contents of (0-7dpi) 1 (Fig 4 and S5 Table). The 12 up-regulated (0-7dpi) 1 pathways (S6 Table) were classified into five broad functional categories. Selected pathways to illustrate this phenotypic diversity are listed in  Forty of the 67 selected down-regulated (0-7dpi) 1 KEGG identifiers sourced from S3 Table were categorised into 11 broad functional groupings (S6 Table). As depicted in Fig 4 this revealed a neuronal dominated phenotype characterised by a down-regulation in Transcription (Ribosome biogenesis ID 3008 ) and Protein processing (Protein processing in ER ID 4141 ) supported by an array of down-regulated GO terms including RNA Regulation of Gene expression GO: 0010468 and Translation initiation GO: 0006413 . Disturbance of the neurotransmission network was predicted by the inclusion of eight pathways including Serotonergic synapse ID 4726 , and Dopaminergic synapse ID 4728 in addition to modulation in the level of synaptic plasticity (Long-term depression ID 4730 and Long-term potentiation ID 4720 . Although Circadian Entrainment ID 4713 was down-regulated, the activity levels of the three GO terms, Entrainment of circadian clock GO: 0009649 , Entrainment of circadian clock by photoperiod GO: 0043153 and Photoperiodism GO: 0009648 had increased. Changes to a set of pleiotropic signal transduction pathways [39] integral to CNS homeostasis were recorded including MAPK signalling ID 4010 , cAMP signaling ID 4024 and ErbB signalling ID 4012 in addition to the neural specific pathway Neurotrophin signalling ID 4722 and Calcium signalling ID 4020 with critical roles in neurotransmission, excitability and long-term plasticity (S6 Table). There was a decrease in B and T cell signalling in addition to Rig1-like receptor signalling ID 4622 and Nod like receptor signalling ID 4621 involved in viral and bacterial pathogen pattern recognition respectively (S6 Table). In contrast five Type I interferon GO terms represented by Type I interferon signalling pathway GO: 0060337 and two innate terms including Regulation of innate immune response GO: 0045088 were up-regulated.
Despite CE-MRI evidence of an increase in barrier permeability in GVR35/CD1 brains [13], there was no significant change in the activity of the Tight junction ID 04530 (S3 Table) or in any of the GO "tight junctions" or "blood brain barrier" GO terms (S4 Table). However, the up-regulated activities of Blood vessel development GO: 0001568 , Vasculature development GO: 0001944 and Blood microparticle GO: 0072562 terms in (0-28dpi) 4 may be indicative of a physiological disturbance of vasculature homeostasis (S4A Table).

Expression profiling of candidate genes
The transcriptome database (S2 Table) was interrogated to determine the 28 day expression patterns of the 38 African trypanosomiasis ID 5143 genes and a panel of proposed Stage 2 biomarker genes (S9 Table) in addition to candidates associated with pathogenic change (S10 Table), the host's immune response (S11 Table) and BBB integrity/impairment (S12 Table). Although default expression parameters were initially set to an adjusted p value of < 0.001 and FC around 1.5, the significance of these parameters was also assessed within the context of individual gene function and expression pattern. The format of these S Tables depicts the Comparison number with the maximum fold change denoted (Max FC # ), its corresponding adjusted p value and best fit transcription expression pattern (Fig 3) for each gene.
The human Slc super-family with over 400 transporter genes organized into 65 functional families [48] have crucial neurologic roles including synaptic activity and in conjunction with the Abc gene family, regulate the efflux and influx of small molecules across the BBB. Changes to this expansive transport network can disrupt CNS homeostasis as documented by the 71 Slc genes associated with human brain disorders [49]. In the mouse, over 250 genes were expressed in the brain [50]. By 7dpi around 100 were up-regulated while the activity of over 50 were decreased by 7dpi (S2 Table). The 26 selected represent the breadth of CNS function (S10 Table) including nine members were associated with neurotransmission including eg Slc6a9 [7dpi"] and Slc17a6 [7dpi#], six were enriched in the BBB [51] eg Slc2a13 [7dpi#] and Slc1a3 [7dpi#] encoding the glucose transporter GLUT13 and the glutamate transporter EATT respectively and six Circadian Gene Data Base (CGDB) genes [52]. In contrast to the vast majority of Slc genes which have a 7dpi expression pattern, SlcO2b1 (anion transport) Slc15a3 (proton-dependent transport of histidine) and Slc11a1 had a [7dpi"-28dpi"] profile.
A scan of the (0-7dpi) 1 dataset identified around 50 circadian gene candidates, the majority, entries in the CGDB [52]. Included in the selected 26 (S10 Table) were a subset of SCN enriched core "clock" genes [53] was the eponymous Clock gene [7dpi#] and Per1 [7dpi"] both of which have been reported to be altered in rodent models [54,55]. The remainder of candidates were classified as SCN enriched non-clock [56] including four high FC genes (S10 Table) with established roles in signalling or rhythmicity were the down regulated AMPA glutamate receptor genes Gria1 and Gria3, Itpr1 (Calcium signalling) and Scg2 encodes a member of the secretogranin family of neuroendocrine secretory proteins. An earlier study [57] reported a reduction in GRIA2 and GRIA3 levels in SCN slice preparations consistent with our finding on the reduced levels of the corresponding message levels in (0-7dpi) 1  Changes occur in aromatic amino acid metabolism during trypanosomiasis including a decrease in tryptophan levels [58]. The activity of three metabolic tryptophan genes, Ccbl1 [7dpi"], Inmt [7dpi"] and Tph2 [7dpi#] was altered (S10 Table). Tryptophan levels, in part are regulated by indoleamine 2,3-dioxygenase, encoded by the African trypanosomiasis ID 5143 Ido1gene (S1 Fig) was incrementally expressed [28dpi"]. This is a rate limiting enzyme in the kynurenine pathway, several pathway catabolites are associated with the inflammatory response [59]. Despite the importance of the pro-inflammatory neuropeptide Substance P [60] in addition to prostaglandins D2 and E2 and nitric oxide as depicted in the African trypanosomiasis ID 5143 no significant changes were detectable in any of the corresponding genes.
There is no documented evidence of a hypoxic episode during trypanosome infection. Despite subthreshold p values, four African trypanosomiasis ID 5143 globin genes shared a [7dpi"-28dpi"] pattern. The failure to detect erythroid enriched Gata1 and Eraf mRNAs may reduce the likelihood of blood contamination as a source of Hb signal. Globin genes in nonerythroid tissue including neurons and glia are constitutively expressed or are induced following a cellular insult eg hypoxia [61]. There was an up-regulation of the neuronal specific Ngb [7dpi"] and the ubiquitously expressed Cygb [7dpi"] encoding neuroglobin and cytoglobin respectively that may provide neuroprotection against hypoxic attack [62], however, the physiological role of these neural globins remains equivocal [63]. Two participants of hypoxia HIF signalling ID 4066 pathway, the endothelial specific transferrin receptor gene Tfrc necessary for iron uptake and the Pdha1 gene encoding the key enzyme pyruvate dehydrogenase were down-regulated with attendant high FCs.
The activities of a small panel of myelin genes including the three high FC CNS myelin structural genes Plp1, Mobp and Omg, Nsmaf, a sphingomyelin enzymatic gene, a myelin transcription factor gene Myt1l and Mal1 involved in myelin sheath maintenance were decreased by 7dpi (S10 Table). The length of this transient dysregulation period may be sufficient to challenge the structural integrity of the CNS myelin sheath. Under physiological conditions, neurons are reliant on oligodendroglial support in ensuring the correct propagation of action potentials which may impact synaptic activity.

Host immunity genes
Over 300 immune genes, the majority described either by the [28dpi"] or the minor [7dpi"-28dpi"] expression profile were expressed during infection (S2 Table). Selected genes to represent the spectrum of the host's response categorised into nine functional groupings are detailed in S11 Table. This revealed a panel of markers for both innate immunity (microglia (Tmem119 [7dpi"-28dpi"] and macrophages Mpeg1 [28dpi"]) and adaptive immunity (the immunoproteasome gene Psmb8 [7dpi"-28dpi"] and the two high FC genes MHC class II Cd74 and the MHC I beta-2-microglobulin (β2M) (S11 Table). Genetic evidence of early diapedesis activity was inferred by the expression patterns of Selpg [7dpi"-28dpi"] encoding the E-selectin ligand and Icam2 [7dpi"-28dpi"] in addition to the Cam integrin genes Itgb4 and Itga7 [7dpi"] (S11 Table). Of note was the activity and diverse function of three other CAM genes sharing the [28dpi"] pattern. Cd274, with a high FC expresses the programmed death ligand-1 (PD-L1) leading to T cell suppression, Cd40 is a member of the Tnfr superfamily and Cd86 involved in T lymphocyte proliferation (S11 Table). The high activity level (26.4 FC) of the extrahepatic gene Saa3, encoding the serum amyloid acute phase protein SAA3 which is secreted into the CSF had a classic [28dpi"] pattern may justify biomarker investigation as Stage2 biomarker as is the Cd74 gene (44.94 FC) (S11 Table).
Raised levels of IgM and IgG are recognised in African trypanosomiasis ID 5143 (S1 Fig). Although 13 (14-28dpi) 9 "immunoglobulin" GO: terms were up-regulated, no annotations specific to IgM or IgG activity were recorded. Indirect evidence of IgM activity was inferred from the expression kinetics of the Mzb1 gene [28dpi"] which is involved with IgM assembly and secretion, but not IgG [64] while Igj [28dpi"] encoding a polymerizing protein that links IgM monomers. There was an increase in the levels of the Fc gamma receptors genes (Fcγr), particularly Fcgr4 and Fcgr3 based on FC values and to a lesser extent Fcgr2b. These multifunctional Fcγr genes that share a [28dpi"] pattern may be indicative of late IgG activity (S11 Table).
Expression of these 300 genes or so, underpinning both innate and adaptive immunity are in part orchestrated by a panel of documented cytokines and a network of chemokines as outlined below.
The cytokine gene expression profile. As noted previously, five GO terms were up-regulated by 7dpi (S4 Table) consistent with the common [7dpi"] pattern for Ifn α/β, Ifnβ1 and the cognate transcription factor gene Irf3 (S11 Table). An earlier study reported a role for Type I IFNs in early resistance [65] while the Ifn α/β receptor KO mouse had a reduced number of trypanosomes in the parenchyma compared to wild type [18]. In addition to Ifn α/β, the proinflammatory Th1 response comprised Ifnγ, Tnfα, Il1α, Il1β, and Il18, all with a [28dpi"] profile. The inflammasome, a pathogen sensor mediates the activation of caspases-1 and 4 that cleave the pre-IL1 β and IL18 polypeptides to release the highly pro-inflammatory cytokines [66]. Both caspase genes, in addition to Il1β, and Il18 shared the incremental [28dpi"] pattern. There was no evidence of a Th17 mediated pro-inflammatory response. Th2 anti-inflammatory activity was limited to Tgfb-1 [7dpi"-28dpi"] in addition to Il6 and Il10, both with a [28dpi"] profile as had their corresponding receptor genes. Two other active interleukin genes were Il15 [28dpi"] associated with GABA and serotonin transmission [67] and Il34 [7dpi"] encoding a ligand for the microglia receptor CSF1R [68] had a [7dpi"-28dpi"] profile. Unlike the afore described cytokines with modest FCs, the Ik cytokine gene with a [7dpi#] pattern had a high FC. The encoded nuclear spliceosomal RED protein represses the expression of (MHC) class II induced antigens by IFNγ [69]. A comparison of the cytokine gene expression profile described here with African trypanosomiasis ID 5143 and a rodent brain compilation study [3] identified a string of common genes are highlighted in in shade (Table 4). Notable differences were the absence Il12a and Il12b from the rodent brain profiles and the absence of IlIa from African trypanosomiasis ID 5143 .
JAK-STAT activity initiates interferon cytokine ligand binding leading to the transcription of a myriad of key IFNα/β and IFNγ families of ISGs most with a [28dpi"] or [7dpi"-28dpi"] pattern (S11 Table), several with a high FC. Prominent IFNα/β ISGs included members of the Interferon Regulatoryu Factor (Irf) family [70], the IFN-induced transmembrane protein (Ifitms) grouping [71] and the IFN-induced protein with tetratricopeptide repeats (Ifit) family [72]. The guanylate-binding protein GTPases families the small 47 kDa Immunity-Related GTPases and 65-67Kda family play a pivotal role in the type 2 interferon response. In addition, Gvin a member of a very high MW GTPase family of unknown function was active [73].

PLOS NEGLECTED TROPICAL DISEASES
which can lead to suppression of cytokine signalling by inhibition of the JAK-STAT pathway. The forkhead transcription factor gene Foxp3 [7dpi"] has a critical role in Treg mediated suppression [75]. The chemokine gene expression profile. The main drivers of the chemokine gene response as measured by elevated FCs (S11 Table) were the Ccl members, Ccl4, Ccl5, Ccl7 and Ccr5 [28dpi"], the Cxcl species, Cxcl9, Cxcl10, Cxcl11, Cxcr3 Cxcl13 [28dpi"], Cxcl16 [7dpi"-28dpi"] and Cxcl14 [7dpi"]. The Cxcl10 post-14dpi inducible pattern and to lesser extent Cxcl9 is consistent with the predicted high CXCL10 gradient in the astrocytic endfeet of the NVU purported to facilitate the coupled traversal of leukocytes and trypanosomes across the BBB [26]. Using the same end-point RT.PCR stratagem, we have reported a similar expression profile for the Ifnγ and inducible ligand genes Cxcl9, Cxcl10 and Cxcl11 and its shared receptor gene Cxcr3 to that reported in the infected brains of a rat model of HAT [34]. In contrast to the other Cxcl mRNAs, Cxcl14 had a Stage 1 [7dpi"] signature while a recent study found that the pleiotropic CXCL14 specifically binds to CpG DNA and activates TLR9 on macrophages, thereby inducing inflammatory cytokines [76]. This expression signature was compared (Table 5) with a compilation profile of Ccl and Cxcl ligand genes expressed in infected rodent brains [3]. Common genes are depicted in shade, with the exception Cxcl11 which harbours a mutation in the C57/BL6 mouse other differences between the profiles was the presence of Cxcl1 and Cxcl5 in addition to the grouping of Ccl genes 9, 12, 19, 21 and 28 in the Masocha profile [3]. Although a four-fold increase in Ccl28 expression in the (6-28dpi) pairing was reported [30], no message activity was recorded. Other active chemokine genes not featured in the Masocha profile [3], were Cx3cl1 [7dpi"] originally termed fractalkine in man, which unlike other chemokines has both chemoattractant and cell adhesion properties. Its cognate receptor gene Cx3cr1 had a [28dpi"] pattern.
As detailed earlier, although most of the cytokine and chemokine genes, in addition to many innate and adaptive immune genes had a [28dpi"] or a [7dpi"-28dpi"] profiles, several transcripts increased incrementally between 0 and 14dpi evidence of a Stage1 immune response. This selected panel included the innate mRNAs, Cxcl9, Cxcl13 and Ccl5 and H2-K2, H2-DMb1and B2m as representative of adaptive immunity genes (S11 Table). This is consistent with the mixed neuronal-immune phenotype predicted from the functional enrichment analysis describing the period between 0 and 14dpi.
Brain barrier integrity/impairment genes. As laid out in S12 Table, candidate genes associated with barrier impairment can be broadly classified whether the genetic changes are (a) "disruptive", involving alterations in the macromolecular architecture of the endothelia paracellular barrier complex or (b) non-disruptive or "functional" leading to an increase in vascular permeability caused by well documented and candidate mediators [77]. A useful

PLOS NEGLECTED TROPICAL DISEASES
Microprofiling the mouse CNS response to trypanosome infection resource reference for this screen was a "molecular atlas" [51] of brain enriched mRNAs assembled from a compilation study of control and disease BBB expression databases comprising 17 paracellular genes, eight members of the Abc transporter family and 39 genes belonging to the Slc transporter super family. (a) Disruptive changes. Excluding the endothelial gene Esam [7dpi"] encoding the endothelial specific adhesion JAM protein, this screen failed to detect significant changes in the activities of any of the key paracellular barrier structural genes consistent with the subthreshold values for the Tight Junction ID 4530 and Tight junction GO: 0005923 . However, a set of ten ancillary including Tight Junction ID 4530 genes eg Ctnna1 [7dpi"], Tjap1 [7dpi"] and Exoc4 [7dpi#] were altered by 7dpi (S12 Table). A late event during the neuroinvasion process is the enzymatic degradation of the basement membrane by the metalloproteinases, most notably MMP2 and MMP9 [25]. However, the significance of this to the corresponding expression patterns Mmp2 [7dpi"-28dpi"] and Mmp9 [7dpi"] is equivocal as the expression parameters were subthreshold. The African trypanosomiasis ID 5143 (S1 Fig) schematic highlights the importance of the laminin composition of the endothelial perivascular space during neuroinvasion. An immunofluorescence study observed that the ECM protein laminin 4 is permissive for parasite migration from the parenchymal basement while laminin 5 restricts entry into the parenchyma [24]. Despite having different expression profiles Lama4 [7dpi#]and Lama5 [7dpi"] (S12 Table), subthreshold expression parameters tempers comparisons between both studies.
(b) Functional changes. African trypanosomiasis ID 5143 (S1 Fig) incorporates a non-disruptive functional neuroinvasion route based on an in vitro human endothelial layer BBB chamber model study [27]. It has been proposed that the trypanosome encoded cysteine protease brucipain binds to the membrane bound PAR2 receptor disrupting the host's calcium signalling network facilitating the transendothelial migration underpinned by nine denoted genes, six of which were differentially expressed at 7dpi including the signal transduction gene Gnaq [7dpi#] two phospholipase genes eg Plbc3 [7dpi"] and three protein kinase C genes eg Prkca [7dpi#]. Further evidence of this critical relationship between maintenance of calcium homeostasis and BBB integrity [78] was the early down-regulation of around fifty calcium signalling genes. Of the 11 selected entries (S12 Table), Cacnb4, Camk, Ednrb and Itpr1 had high FCs in addition to Slc24a2 and Slc8a1 encoding calcium exchangers. This was reflected in a decrease in Calcium signalling ID 4020 activity albeit at a marginal subthreshold value in addition to ten (0-7dpi) 1 GO terms eg Calcium ion transmembrane transport GO: 0070588 and the up-regulation of 18 (0-28dpi) 4 terms possibly reflecting a different role for calcium post-14dpi.
Barrier homeostasis is regulated in part, by a cohort of pro-and anti-angiogenic mediators several of which were differentially expressed albeit with modest FCs by 7dpi (S12 Table). This included the up-regulation of three members of the VEGF/PDGF family [79], and the angiogenic genes Angptl 4 [7dpi"], Angptl6 [7dpi"], the pro-angiogenic endothelial specific Aggf1 [7dpi#] [80] and Eng [81] [7dpi"]. In contrast to these mRNAs, an increase in the activities of, Angpt1 and the highly endothelial enriched Angpt2 increased post-14dpi [28dpi"]. Dysfunctional signalling of the Wnt/beta catenin pathway can lead to BBB breakdown [82]. The four ligand genes Wnt2, Wnt4, Wnt7a and Wnt7b and its cognate receptor Fzd2 gene shared a [7dpi"] pattern.
Perturbation in the expression of the Abc and Slc transporter gene families, particularly the latter, as alluded to elsewhere may challenge the functionality of the BBB. In addition, an increase in the low rate of endothelial transcytosis network during CNS disease could also impact on vascular permeability consistent with a modest up-regulation of three well characterized endothelial transcytosis genes, viz Cav1 [83] [28dpi"], Mfsd2 [84] [7dpi"] and Plvap [85] [7dpi"]. The activity of the highly endothelial enriched Ly6a message [28dpi"] with an attendant high FC may be of interest. The gene encodes a GPI-anchored protein that facilitates the neuroinvasion of the adeno serotype AAV-PHP.B in a mouse strain specific manner, including CD1 [86]. The Aquaporin-4 water channel encoded by the Aqp4 gene [7dpi#] was abundantly expressed in the astroglial endfeet has a central role in CNS water regulation linked to BBB permeability [87]. Both gene and the Vasopressin-regulated water reabsorption ID 94962 pathway were down-regulated by 7dpi.

Discussion
Disease progression over the 28 day timeline of the CD1/GVR35 HAT mouse has been well characterised in contrast to the underlying transcription profile. Based on this knowledge gulf, the twin aims of this microprofiling study were to i): track changes to the quantitative and qualitative transcriptomes of infected CD1 mice brains leading to the identification of message populations and predicted phenotypic traits ii): examine the expression profiles of candidate genes with documented or potential association with trypanosomiasis. Within the general context that the infected transcriptome could be broadly divided into four message population each with a distinct expression pattern, one critical and novel observation in addition to three others define this study. First, the period between 0 and 14dpi was marked by the coincident up-and down-regulation of over 8,000 transcripts approximating to around 4,000 genes. Functional enrichment analysis of this population identified many of the active genetic programmes to be associated with changes in neuronal behaviour and an early immune reaction. Second, the levels of over 300 mRNAs associated with various aspects of the immune response were induced post-14dpi. Third, although markers indicative of early changes in vascular permeability were identified by 7dpi, there was no genetic evidence of any structural disruption to the paracellular BBB. Fourth, candidate screening validated the expression profile of many documented immune response genes but failed to corroborate the activity of others associated neuropathogenesis and identified a small cohort of novel transcripts considered for further study.
It should be emphasised that before discussing the significance of these four findings, the interpretation of the microarray data per se should be tempered by two technical constraints. The first, is the inherent issue of the weighting to be given between differential expression p values and the corresponding fold change [36] as illustrated with the (0-7dpi) 1 Comparison, where only 2370 transcripts had a FC± �2 from an active message population of 8145 mRNAs. The second, concerns the detection sensitivity when using whole brain homogenates where changes in message abundance in infected small regional loci eg BBB or SCN could be masked by the combined expression levels in the rest of the non-infected brain. This scenario may explain the failure to validate published data such as the endothelial basement laminin protein genes [24] and ECM metalloproteinase genes [25].
In contrast to that reported here, there was no evidence of an equivalent period of differential gene expression activity between 0 and 14dpi in C57/BL6 brains infected with the AnTat 1.1E T.b brucei stabilate [29]. Notwithstanding these inconsistencies during Stage 1 between the murine models, there was a similarity in gene numbers between the Amin study pairing (6-28dpi) [29] and our equivalent Comparison (7-28dpi) 4 in addition to qualitative Stage2 similarities between the chemokine profiles [26] and in a cohort of non-cytokine/chemokine secreted proteins [30]. These transcriptomic differences may in part arise from the variation in the complex interplay between different combinations of host and parasite genetics. This is consistent with the variability in trypanotolerance between a series of commonly used inbred mouse model strains [6]. Gross differences were reported in spleen and liver pathology following infection of two T. b. brucei genetically distinct stabilates TREU927 and STIB247 in BALB/ c mice [88]. Variation in trypanotolerance in the cattle breeds N'Dama and Boran infected with T. congolense [89] as well as differences in the disease profiles between T.b brucei and T.b rhodesiense patients, the development of CNS symptoms in rhodesiense Stage1 patients [14] and the presence of asymptomatic patients [90] all highlight the spectrum of responses elicited by this parasite. Moreover, an intermediate stage in HAT has been proposed following reported variation in the levels of the Stage 2 biomarkers CXCL13 and Neopterin in T b gambiense patients [91].
Functional enrichment analysis up to 14dpi identified genetic networks that predicted a myriad of transient changes to metabolism, mitochondrial activity, pleiotropic signalling, transporter activity, calcium signalling, neurotransmission including synaptic plasticity, circadian activity and vascular permeability. There was also evidence of a type I interferon response, activation of the innate system triggering both a B-and T-cell responses against trypanosome VSG coat antigens and complement activation. These phenotypic traits can be envisaged as the precursor for the initiation of a string of neuronal and immune changes that precede progression to Stage 2. These observations share similarities with the outcome of a T.b.brucei rat model collaborative study reporting the detection of trypanosome DNA at 6dpi, the presence of parasites and T cells in the parenchyma at 9dpi and significantly, early signs of circadian disruption [12]. Data from our CD1/GVR35 model [13] also detected trypanosome DNA in brain homogenates by 7dpi and observed significant barrier impairment on CE-MRI and neuroinflammation at 14dpi. Moreover, intravital microscopy observations claimed the presence of fluorescently tagged GVR35 trypanosomes within the parenchyma of infected mice as early as five hours after infection [11]. However, this whole issue of parenchymal invasion has been fundamentally challenged in a rat model that failed to detect trypanosomes in the parenchyma and contended that the parasites were restricted to the pial space following breakdown of the blood CSF barrier [15]. These combined studies purporting the early presence of parenchymal trypanosomes may be sufficient to elicit an early expansive genetic change consistent with our transcriptome and phenotypic observations up to 14dpi.
This period of early initiation of innate and adaptive immunity was followed by a secondary wave of post-14dpi activity marked by an up-regulation of over 300 immune genes including a combined grouping of immune-regulatory cytokine genes, ISG's and chemokine genes (S11 Table) that control the innate response. The pro-and anti-inflammatory cytokine gene signatures are similar with a rodent model consensus profile [3]. There was no evidence of a proinflammatory Th17 response. A notable feature was an expansive increase in ISG activity comprising members the IFNα/β induced Irf, Ifitm and Ifit gene families and the IFNγ induced large 65-to 67-kDa GTPases and p47GTPase gene families. Message levels of ISGs Irgm3, Ifi47 and Gbp2 were also increased during T.gondi infections [73]. Excluding the Cxcl14 gene expression that peaked at 7dpi, the post-14dpi chemokine gene expression signature closely resembled a rodent model compilation profile [3]. This post-14dpi period was marked by the activity of many adaptive genes including histocompatibility H2 immunoproteasome genes supplemented by a large grouping of "Adaptive" KEGG and GO terms. Specificity to protozoan infection was confirmed by the up-regulation of six protozoan response KEGG pathways including the eponymous African trypanosomiasis ID 5143 . There was evidence of shift the immune response towards an increase in Type 1 IFN suppression inferred by the post 14dpi activity of the suppression marker genes Cd274, Parp14 Trim21 and Usp18 [74].
By CE-MRI, we were able to quantify an increase in contrast agent leakage signal at 14dpi [13]. However, it is unknown whether this change in permeability was "disruptive" or "functional" in origin [77]. There was no evidence of expression differences in any of the major paracellular complex genes or pathways. This is consistent with a rat study where trypanosomes were able to cross the BBB without any loss of tight junction proteins [22] although dysregulated tight junction activity was claimed to account for an increase in fluorescent dye uptake in a rat model [21]. Functional impairment could arise from the perturbed expression between 0 and 14dpi of any of the two BBB enriched Abc and nine Slc genes [51]. During this period there were fluctuations in genes that regulate calcium homeostasis [78] and angiogenesis [80][81][82] both processes being critical to barrier integrity. However, it remains equivocal whether any of these proposed early transient changes could impact on functional barrier activity to account for leakage of contrast agent from the BBB at 14dpi [13].
A recent RNASeq study [92] on transcriptomes from peripheral blood and the CSF of T.b. rhodesiense late Stage 2 patients reported a robust innate response in the blood transcriptome sharing genes and pathways catalogued in the (0-7dpi) 1 and (7-14dpi) 7 pairings. In contrast, the late CSF transcriptome had a different expression landscape reflected in the enrichment of anti-inflammatory and neuro-degeneration signalling pathways indicative of a more severe and advanced neurological phenotype than the 28dpi CD1 brain homogenate. Of additional interest were the differences between the Cxcl ligand gene profiles highlighted by the absence Cxcl9, 10, 11, 14 and 16 and the presence of Cxcl3, 5 and 13 in the CSF.
Candidate trypanosomiasis genes with high FCs include Cd74. Generally regarded a MHC class II chaperone, it has been shown to be multi-functional in pathological situations [93] and was expressed at high levels in the Toxoplasma gondi infected mouse brain [94]. Others include Csrp1 (neuronal development), Cd274 (T cell inhibition), Ly6a (neuroinvasion), three serpin genes Serpina3g, Serpina3h and Serping1 (multifunctional protease inhibitors), Cxcl14 (7dpi peak expression) and Saa3 as a Stage 2 biomarker. Others based on functionality are members of the multifunctional transporter Slc gene family involved in neurotransmission and BBB transport, several circadian genes eg Gria2, Gria3 and Itpr1 and the three BBB transcytosis genes Cav1, Plvap and Mfsd2 in addition to the Ly6a.
These changes to the neurotransmission network, the innate response, circadian activity and vascular permeability between 0 and 14dpi merit further investigation. Using the current experimental paradigm, with the inclusion of a series of additional time points either side of 7 and 14dpi, would provide more detailed temporal mapping of these key phenotypic changes and define the temporal lengths of the up-and down-regulated plateaus. A similar approach might be considered for more accurate immune mapping. This may also unravel late CNS disease genes and pathways identified in the late Stage 2 rhodesiense transcriptome study [92].
Future high throughput studies would be severely compromised using whole brain tissue. Further investigation requires a more ambitious experimental paradigm using advanced but established and readily accessible technologies such as Laser Capture Microdissection [95] or a single cell-omics approach [96]. This would facilitate access to distinct functional anatomical brain regions to generate homogenous transcriptomes and corresponding trypanosome DNA loads. This has recently been performed on different endothelia transcriptomes [97], in addition to the SCN [53] and the choroid plexus [98]. Adopting a RNASeq experiment paradigm to quantify homogenous cell populations coupled to the anticipated sensitivity advances in cell imaging and CE-MRI will extend our knowledge of when and by what route(s) the trypanosomes enter the CNS as well as reveal some of the molecular complexities of the phenotypic traits that describe neuroinvasion, the immune response and neurological dysfunction including disruption of the circadian network.
In conclusion, this study has reinforced the growing consensus that early neuronal and immuno phenotypic changes precede CNS disease. This justifies the case for further investigation of the up-down-regulation wave of activity up to 14dpi and has also provided a valuable reference resource, documenting the expression patterns of over 8000 altered transcripts.
Supporting information S1 Fig. African trypanosomiasis ID 5143 KEGG pathway. The schematic is a combined gene and phenotypic flow chart highlighting the IFNγ mediated initiation of the innate response involving the TLR9-MYD88 cascade and the adaptive B cell attack on the trypanosome VSG coat responses during Stage 1 followed by cytokine mediated breakdown of the BBB and the role for laminins in this process. An alternative neuroinvasion route is recognised involving disruption of calcium homoeostasis following endothelial binding of the trypanosome cysteine protease brucipain. Parasite invasion activates a series of Stage 2 neurological pathological changes proposing critical roles for apoptosis, NO, prostaglandins and tryptophan metabolism. These key events in HAT disease progression are depicted within the framework of the 28 day timeline of the CD1/GVR35 mouse model. KEGG and GO functional enrichment analyses data are depicted in the four heatmaps (S2-S5 Figs). Up-regulated KEGG pathways and GO terms are in red while blue depicts the down-regulated data. The ten Comparisons are arranged along the X axis and KEGG pathways and GO terms lie on the Y axis. The number of displayed KEGG pathways and GO terms was restricted to 75, presented in decreasing order of p values. Phenotypic demarcation in both KEGG and GO profiles are boxed. Hierarchical clustering was applied to pathway and GO terms. (TIF) S1 Table. List of normalized transcripts. Only probes detected at least once across the 45,281 probeset spotted on the Illumina MouseWG6_V2_0_R3_11278593_A Bead Chip. were retained for subsequent analysis. Following normalization to minimise variation gene expression measurement, 23,213 probes were designated as transcriptionally active. This collection of normalized transcripts is presented in alphabetical order alongside the corresponding log 2 expression unit value for each probe at the four time points viz 7dpi, 14dpi, 21dpi and 28dpi for each of the three replicates denoted R1, R2 and R3 and three non-infected control samples equivalent to 0dpi. (XLSX) S2 Table. Generation of transcriptome datasets. Ten pair wise Comparisons were performed on this active probeset. Expression difference between each sample pair was set at a strict statistically significant adjusted p value <0.001 with an attendant adaptive threshold fold change with associated significance statistics. The Infected Series compared the non-infected control equivalent to 0dpi against the four infected time points denoted (0-7dpi) 1 , (0-14dpi) 2 , (0-21dpi) 3 and (0-28dpi) 4 . The Temporal Series measured activity changes between six different infected time point combinations viz (7-14dpi) 5 , (7-21dpi) 6 , (7-28dpi) 7 , (14-21dpi) 8 , (14-28dpi) 9 and (21-28dpi) 10 . The resultant transcriptome database (S2 Table) set out for each pairing in order of decreasing p values and corresponding fold changes. (XLSX) S3 Table. KEGG functional enrichment analysis of (A) up-regulated and (B), down-regulated pathways. Up-regulated (A) and down-regulated (B) KEGG functional enrichment analysis was performed on Infected Series Comparisons (0-7dpi) 1 , (0-21dpi) 3 and (0-28dpi) 4 and the Temporal Series Comparisons (7-14dpi) 5 , (7-21dpi) 6 , (7-28dpi) 7 and (14-28dpi) 9 . Analysis of pairings (0-14dpi) 2 , (14-21dpi) 8 and (21-28dpi) 10 were excluded as there was no significant gene expression activity in theses Comparisons. KEGG pathways/ID number are set out or each Comparison in order of decreasing order. (XLSX) S4 Table. GO Functional enrichment analysis of (A) up-regulated GO terms and (B), down-regulated GO terms. Up-regulated (A) and down-regulated (B) KEGG functional enrichment analysis was performed on Infected Series Comparisons (0-7dpi) 1 , (0-21dpi) 3 and (0-28dpi) 4 and the Temporal Series Comparisons (7-14dpi) 5 , (7-21dpi) 6 , (7-28dpi) 7 and (14-28dpi) 9 . Analysis of pairings (0-14dpi) 2 , (14-21dpi) 8 and (21-28dpi) 10 Table. KEGG functional enrichment analysis of up-regulated genes in Temporal Comparisons (7-14dpi) 5 , (7-21dpi) 6 Table. a) African trypanosomiasis ID 5143 and b) Stage 2 Biomarker gene expression profiles. The expression profiles of the 38 genes that define the African trypanosomiasis ID 5143 KEGG pathway (a) and a panel of published HAT biomarker genes (b). Each gene was matched against its Comparison # with the maximum fold change (Max FC # ), adj p value and expression pattern (Fig 3.). (DOCX) S10 Table. Candidate genes associated with pathogenesis. Pathogenic candidate genes, were selected based on a high FC, inferred biological significance or both, were grouped into 13 functional categories. Each gene was matched against its Comparison # with the maximum fold change (Max FC # ), adj p value and expression pattern (Fig 3.). (DOCX) S11 Table. Host's immune gene response. Immune genes were selected based on a high FC, inferred biological significance or both, were grouped into nine immune functional categories. Each gene was matched against its Comparison # with the maximum fold change (Max FC # ), adj p value and expression pattern (Fig 3.). (DOCX) S12 Table. Candidate genes associated with BBB integrity/impairment. Candidate genes were designated a) as "disruptive" involving structural changes to the paracellular barrier complex or b) "functional" where a change in the activity of a range of non-cytokine mediators. Each gene was matched against its Comparison # with the maximum fold change (Max FC # ), adj p value and expression pattern (Fig 3.). (DOCX)