PneumoTRACE logo

PneumoTRACE

Enter the access password to continue.

Incorrect password. Please try again.

Methods

Data sources

Our primary data source comes from the analysis of remnant clinical nasopharyngeal samples obtained from Yale New Haven Hospital. These samples are typically collected for the purposes of viral testing in the emergency department or outpatient clinics in the Yale New Haven Health system. We aim to collect 3000 samples per year, including individuals <18 and 50+ years of age. This data source is unique in its comprehensive age coverage, in comparison to most carriage surveillance systems which focus exclusively on children <5 years of age. For each sample, we have information on:

  • Pneumococcal positivity (based on qPCR lytA and piaB)
  • Pneumococcal serotype (based on sequence-based determination)
  • Date of sample collection
  • Age group
  • Race/ethnicity
  • PCV history
  • Viral positivity status

The carriage data are supplemented by a public dataset made available by the CDC’s Active Bacterial Core Surveillance (ABCs) system, stratified by age group, year, and state. Connecticut is one of the surveillance sites for the ABCs system. These data, which are delayed by 2–3 years, are combined with the latest carriage data to obtain statistical estimates of current IPD patterns (“Nowcasts”). We combine the IPD counts with population size estimates for the ABCs catchment areas, obtained from the Census Bureau, to calculate rates of IPD. As of May 28, 2026, the latest data available were through 2024.

Lab methods

All nasophayngeal swabs are preserved in Viral Transport Media (VTM) after collection and stored at the Yale School of Medicine’s Clinical Microbiology Laboratory. Frozen samples are transported to the Yale School of Public Health for nucleic acid isolation, qPCR, and serotyping.

Automated DNA Isolation

DNA was isolated from all samples using a Monarch® Mag Viral DNA/RNA Extraction Kit (New England Biolabs, #T4010), adapted for use on the Thermo Scientific Kingfisher Flex System for high throughput processing, and omitting use of the kit-provided Carrier RNA. Each DNA isolation was run with up to 92 samples to leave space for controls in subsequent steps and included a negative extraction control of nuclease-free water.

In brief: 200µL of each sample was first mixed with 5µL Proteinase K and incubated at room temperature for 15 minutes. Proteinase-K-treated samples were then combined with 420µL lysis buffer/bead mixture and loaded into the KingFisher Flex system. In the Kingfisher Flex, samples were further mixed with the lysis buffer/bead mixture, washed in 500µL Monarch Buffer BX, followed by two washes in 500µL freshly mixed 80% Ethanol. DNA was eluted in 75µL nuclease-free water and either stored at 4°C for immediate downstream processing, or at -80°C for long-term storage.

Screening for pneumococcus using qPCR

Samples are screened using qPCR to detect the lytA (autolysin protein) and piaB (iron transporter lipoprotein) genes present in S. pneumoniae. Because non-pneumococcal streptococci are known to have the LytA gene and may give a false positive signal, we rely on detection of PiaB alone as the most conservative estimate of positivity. To ensure that the breadth of viable samples are included, all LytA+ and PiaB+ samples are carried over to serotyping analysis, in addition to negative isolation controls and a small selection of qPCR-negative samples included for quality assurance.

Each qPCR reaction mixture (25µL total volume) consisted of:

  • 12.5µL Luna® Universal Probe qPCR Master Mix (New England Biolabs, # M3004)
  • 0.5µL LytA-Forward (10uM),
  • 0.5µL LytA-Reverse (10uM)
  • 0.5µL LytA-Probe (10uM),
  • 0.5µL PiaB-Forward (10uM)
  • 0.5µL PiaB-Reverse (10uM),
  • 0.5µL PiaB-Probe (10uM),
  • 8.5µL nuclease-free water,
  • 2.5µL DNA template.

All PCR plates included a nuclease-free water negative template control and two synthetic dsDNA gBlocks (Integrated DNA Technologies) as positive controls, each designed to amplify one of the assay targets. qPCR results were interpreted using the CFX Maestro Software for CFX Real-Time PCR Instruments (Bio-Rad), using a single threshold Cq Determination Mode, a Baseline Subtracted Curved Fit Setting, and applying the Fluorescence Drift Correction option to remove any abnormal curvatures. Results were visually inspected in-software to confirm appropriate exponential curvature in each amplified sample.

qPCR cycling conditions were as follows:

  • 1 cycle of: 95°C for 3 minutes
  • 40 cycles of: 98°C for 15 seconds, 60°C for 30 seconds, and a plate read

Amplification of S. pneumoniae Using Targeted Multiplexed PCR

To enrich S. pneumoniae DNA, selected samples and controls were amplified via multiplexed PCR using primers developed previously Downs et al. , with slight modifications. The reference paper utilized a Fluidigm microfluidics system to perform a qPCR capable of differentiating between 92 Streptococcus serotypes, in addition to 14 other bacterial species. To achieve this level of granularity without access to a microfluidics system, the same primer set was instead utilized to generate PCR amplicons for downstream sequencing.

DNA were first amplified by PCR using 94 primer pairs separated into three pools, to prevent oligo cross-amplification. All primers were ordered at 100µM and resuspended in 200uL IDTE, pH 8.0 (10 mM Tris-HCl/0.1 mM EDTA) (Integrated DNA Technologies) and subsequently diluted to 10µM in the same IDTE buffer solution. Primers were combined at equal volumes into their respective pools (A, B, or C), each containing 30 - 32 primer pairs.

For each sample, three PCR mixes were prepared as follows: * 12.5µL of Q5® High-Fidelity 2X Master Mix (New England Biolabs, #M0492) * 4µL primer pool (A, B, or C) * 3.5µL nuclease-free water * 5µL template (DNA or nuclease-free water)

Post-amplification, products from each sample were combined and purified using Selective Phase Reversible Immobilization (SPRI) beads (Mag-Bind® TotalPure NGS, Omega Bio-Tek), using a 1.8x bead to sample ratio. During purification, 30µL immobilized DNA were washed twice with 80% EtOH before being eluted in 20µL nuclease-free water.

DNA Library Preparation and Sequencing

DNA libraries were prepared for sequencing using Oxford Nanopore Technology’s Native Barcoding Kit 96 V14 (SQK-NBD114.96), referencing their protocol for ligation sequencing amplicons[5]. As anticipated PCR amplicon sizes were all below 200bp, the recommended options and reagents for a “very short” (<1kb) library length were used through the entirety of the process. Prepared DNA libraries were loaded into a MinION/GridION Flow Cell (DNA) (FLO-MIN114) (Oxford Nanopore Technologies), following the manufacturer’s instructions[5]. Flow cells were run on a MinION Mk1B, using the MinKNOW software, including all reads with a Q score >=1.

Sequence analysis

Sequencing reads are mapped to reference sequences from the SeroBA database, and serotype is determined based on the combination of targets that are detected, based on the approach described by Downs. For some amplicons, it is not possible to resolve the serotype based on the short sequence alone, leading to ambiguity within the serotype. To resolve this ambiguity, we used information on the expected prevalence of serotypes, based on IPD data from 2024, combined with published invasiveness data. This allows us to up or downweight certain serotypes–if a serotype is never seen in IPD, it is unlikely to be a major contributor to carriage.

Statistical methods

Nowcasting

The latest IPD data for the US lags 2–3 years behind the current time. To address this gap, we use two complementary approaches for determining the current distribution of serotypes causing IPD (“Nowcasting”). As more colonization data accrue, the goal is to eventually combine these two methods to obtain a single consensus estimate. For now, the two methods generate independent estimates, which are presented side-by-side.

The first method uses a traditional time series forecasting methodology. IPD data for 1998–the latest timepoint (e.g., 2024) are used to fit the model. For each serotype, a negative binomial regression is fit, with fixed effects for age group, and linear splines to capture vaccine-associated changes, with knots in 2000, 2005, 2010, and 2015, and an autoregressive, order 1 (AR(1)) term that varies by age group to capture residual variation. The denominator (offset) for these analyses is the total number of IPD cases. This helps to smooth over differences due to reporting or behavior, for example during the pandemic. This method was validated and evaluated against alternative model structures by using historical data (e.g., fitting the model through 2013 and predicting IPD in 2014 and 2015). For each serotype and age group, we obtain a median estimate for the nowcast as well as uncertainty intervals for these estimates. These models are fit using a Bayesian approximation method (INLA). Minimally informative priors (Normal(0,1)) are applied to fixed effects and help to stabilize the estimates against extreme values.

The second method for nowcasting relies solely on the age-stratified colonization data. Some serotypes are more or less likely to cause IPD. The relative ratio between colonization and IPD for each serotype is similar across populations. In other words, the serotypes that are most invasive in one population also tend to be the most invasive in other populations. These case-carrier ratios are also somewhat stable over time, though certain bacterial lineages can also influence invasiveness in addition to serotype, and the introduction of vaccines reduces the observed invasiveness in immunized age groups. By combining the case-carrier ratios from published studies (meta-analyzed by Lochen and Croucher) with our serotype-specific prevalence data, we can obtain predictions of the distribution of serotypes causing IPD. Uncertainty around these projections are obtained using Monte Carlo resampling of the invasiveness estimates and the carriage data.

These two methods each have important limitations. The time series forecasting model relies on trends continuing over a 20+ year period. The second method relies heavily on accurate detection of colonization. Some serotypes (e.g., serotype 1) are not commonly carried and might not be represented. Likewise, other serotypes, like serotype 4, are mostly found in high-risk populations, like among people who are homeless. Therefore, our carriage surveillance might miss these serotypes. By looking at the estimates from both methods together, we can obtain a more holistic picture of the currently circulating serotypes and their likely impact on rates of IPD.