VPDA protocol

The raw reads in a metagenomic sample are mapped by MetaPhlAn 3 to a database of 1.1M markers using bowtie2 (Langmead and Salzberg, 2012 (link)). The default bowtie2 mapping parameters are those of the ‘very-sensitive’ preset but are customizable via the MetaPhlAn 3 settings. In MetaPhlAn 3, the input can be provided as a single FASTQ file (optionally compressed), multiple FASTQs in a single archive, or as a pre-performed mapping. Internally, MetaPhlAn 3 estimates the coverage of each marker and computes the clade’s coverage as the robust average of the coverage across the markers of the same clade. The clade’s coverages are then normalized across all detected clades to obtain the relative abundance of each taxon as previously described (Segata et al., 2012 (link); Truong et al., 2015 (link)).
In version 3, we further optimized the parameter of the robust average which excludes the top and bottom quantiles of the marker abundances (‘stat_q’ parameter). This is now set by default to 0.2 (i.e. excludes the 20% of markers with the highest abundance as well as the 20% of markers with the lowest abundance). To further improve the quality of the read mapping, we adopted quality controls before and after mapping by discarding low-quality sequences and alignments (reads shorter than 70 bp and alignment with a MAPQ value less than 5).
We also introduced a new feature for estimating the ‘unknown’ portion of the taxonomic profile that would correspond with taxa not present in current databases; this is computed by subtracting from the total number of reads the average read depth of each taxon normalized by its taxon-specific average genome length. Additionally, the new output format for MetaPhlAn 3 by default includes the NCBI taxonomy ID of each profiled clade, allowing for better comparisons between tools and tracking of the species name in case of taxonomic reassignment.
Finally, alongside the default MetaPhlAn output format, profiles can be now reported using the CAMI output format defined by Belmann et al., 2015 (link); BioBoxes, 2020 that can be used for performing benchmarks with the OPAL framework (Meyer et al., 2019 (link)). To support post-profiling analyses, a convenience R script for computing weighted and unweighted UniFrac distances (Lozupone and Knight, 2005 (link)) from MetaPhlAn profiles is now available in the software repository (metaphlan/utils/calculate_unifrac.R), alongside the phylogeny (in Newick format) comprising all MetaPhlAn 3 taxa. The improvements and addition in MetaPhlAn 3 compared to the previous MetaPhlAn two version are summarized in Supplementary file 2.

In order to test VarElect’s performance against Phenolyzer [23 (link)], we generated 34 benchmark queries, composed of a disease causing gene (the probe gene) and its relevant disease/phenotype/symptom terms. Of these, 13 queries were based on in-house exome analyses, 9 were based on exomes analyzed at Toldot (http://www.toldot-dna.com/content/about-us), 12 from published literature, including 4 recorded in MalaCards. In the VarElect runs, each probe gene was accompanied by background genes, some consisting of filtered gene lists from the original experiment, and others including groups of 500 randomly sampled genes from the Illumina TruSight whole exome gene panel (http://www.illumina.com/products/trusight-one-sequencing-panel.html) (see Additional file 2: Table S1 for more details).
For comparison of VarElect’s against Exomiser [24 (link)], 10 real pathological variants from solved in-house exomes were spiked (each one separately) into the demo VCF file that is distributed with Exomiser (“Pfeiffer.vcf”, after removal of the originally spiked variant). These VCF files were analyzed using hiPHIVE - an algorithm for cross-species phenotype analysis in whole-exome candidate gene prioritization [24 (link)], using the command line tool of Exomiser (V 7.1). The following parameters were used: prioritiser = hiPHIVE -F 2 -Q 30 –P true. As routinely done when using Exomiser, a restricted vocabulary tool, the submitted disease-related terms were translated to Human Phenotype Ontology (HPO) terms using Phenomizer (http://compbio.charite.de/phenomizer/). The analysis output included 715 variants from 607 genes. The same genes were submitted in parallel to VarElect using the original (untranslated) disease-related terms for gene prioritization (submission details in Additional file 3: Table S2).
For comparison of VarElect against Phevor2 [25 (link)], the same data used for Exomiser were submitted to Omicia Opal 4.16.1, using the VAAST3 filtering and prioritization pipeline [26 (link)]. Spiked VCF files were uploaded into Omicia commercial cloud application, the variants were then annotated, filtered and scored by VAAST tool version 3.0.3.6. VAAST results were further submitted for gene phenotype prioritization by Phevor 2.0.0 (with the same phenotypes as used by VarElect). The genes that passed the VAAST procedure were also submitted to VarElect. Spiked genes that were filtered out by VAAST (3 cases) were subsequently added manually to the gene list submitted to VarElect.
For comparison of VarElect against Ingenuity (IVA), the same data used for Exomiser were submitted to analysis in IVA. The IVA default filtration pipeline was used. Biological context setting was first selected as 0-hops from the relevant phenotype terms as translated by Ingenuity Pathway Analysis (IPA). If the spiked gene was not included in the list of kept genes, then the biological context filter was reset to “within 1 hop”. The gene list which included the spiked gene was submitted in parallel to VarElect using the original disease-related terms for gene prioritization in both the direct and indirect mode.

IHC was performed as described (Sorrelle et al., 2019 (link)). Briefly, slides were warmed in a 60°C oven for 10 min before deparaffinization and rehydration. Slides were fixed in 10% neutral buffered formalin for 30 min followed by a PBS wash. Antigen retrieval was performed in antigen retrieval buffer (10 mM Tris-HCl, 1 mM EDTA with 10% glycerol) at 110°C for 18 min (∼4–5ψ). Tissue sections were blocked with 2.5% horse serum (S-2012-50; Vector Laboratories) or 2.5% goat serum (S-1012-50; Vector Laboratories) followed by overnight incubation with primary antibody. The next day, slides were washed and incubated with HRP-conjugated secondary antibody for 30 min on a shaker. Slides were then washed three times for 5 min in PBST before developing signal. For developing chromogen signal, Bentazoid 3, 3′ diaminobenzidine (DAB; BDB2004L) was used. Slides were counterstained with hematoxylin and then coverslipped using VectaMount (H-5501; Vector Laboratories) and scanned at 20× using the Hamamatsu NanoZoomer 2.0-HT. For developing fluorescence signal, opal substrates (SKUFP1488001KT, SKU FP1487001KT, SKUFP1497001KT; Akoya Biosciences) were used. When performing multiplex IHC, slides were stripped off of antibodies by incubating in 10 mM citrate buffer (pH 6.2, 10% glycerol) in a pressure cooker at 110°C for 2 min. Sequential staining was performed by stripping after every round of staining before probing for the next antibody. Slides were counterstained with DAPI and then coverslipped using Prolong Gold (#P36931; Life Technologies). Slides were scanned at 40× using the Zeiss Axioscan.Z1 in Whole Brain Microscopy Facility of UT Southwestern.

Tissue sections of 4 μm, serially taken after the H&E stained sections, were
stained with mIHC using the Opal^TM 7 solid Tumor Immunology Kit
(PerkinElmer, Waltham MA, USA). In order to optimize the incubation time and
concentration of antibodies the staining method was modified from the
manufacturer’s instructions, as previously done for colorectal cancer.^{26 (link)} For
optimization, PCa tissue sections from the actual cohort were used with the aim
of allowing exposure times of 30–200 ms and a signal range of 5–30 ms. The
sections were dried overnight, heated at 60°C for two hours, deparaffinized, and
rehydrated. They were then sequentially stained using specific antibodies
directed against T-box expressed T-cells (T-bet) also known as Tbx21 expressed
on Th1-cells, CD8, CD20, FOXP3, CD68, and pan-cytokeratin. The nuclear staining
was performed with DAPI and visualization of specific antibody binding together
with different Opal fluorophores (OF) from the Opal ^TM 7 solid Tumor
Immunology Kit. The specific antibodies directed against T-bet were used with
OF520 (green), those against CD8 with OF570 (red), those against CD20 with OF540
(yellow), those against FOXP3 with OF620 (orange), those against CD68 with OF650
(cyan), and those against cytokeratin with OF690 (magenta). The antibody working
concentration and clones were as follows: 4 μg/ml anti-T-bet (clone 4B10:
sc-21749, Santa Cruz Biotechnology, Inc, Dallas, Texas, US), 0.12 μg/ml anti-CD8
(clone C8/144B, Dako Agilent, Santa Clara, CA, US), 4 μg/ml anti-CD20 (clone L26
ab9475, Abcam, Cambridge, UK), 0.33 μg/ml anti-FOXP3 (Tregs), 0.24 μg/ml
anti-CD68 (clone KP1 M0814, Dako Agilent), and 3.6 μg/ml anti-cytokeratin
(pan-CK) for identification of tumor epithelial cells (clone AE1/AE3 M3515, Dako
Agilent). Slides were mounted using ProLong Diamond Antifade Mountant (Thermo
Fisher, Waltham, MA, USA).

During the observation phase, monthly recording of spontaneous leg movements will begin at Month 1 and continue through Month 4. Infants will wear wireless inertial sensors on their ankles (Opal, APDM Inc, Portland, Oregon, United States.) for at least eight hours for two consecutive days, secured by custom leg warmers with a pocket to hold the sensor in place (see Figure 2). Two consecutive days is a sufficient and optimal amount of time to demonstrate an infant's typical daily performance while balancing burden on the infant/caregiver (36 (link)). If the infant is home, research staff will visit the home on day 1 of data collection to teach the caregivers how to place the sensors in the morning, remove them at bedtime, and charge them overnight. If the infant is still in the hospital, research staff will place and remove the sensors and coordinate with nursing staff to avoid interfering with clinical care.