About This Calculator
To measure the frequency of HIV-1 latency, The Siliciano Lab
developed an assay some years ago that uses uninfected donor cells to grow HIV-1 out of cells from infected patients at various dilutions1, 2
. This calculator, written by Daniel Rosenbloom
and made into a web app by Oliver Elliott
, computes the frequency of latent infection based on which dilutions of patient cells are positive for HIV-1 growth.
Interpreting the Output Statistics
- The maximum likelihood estimate and confidence interval provide a good estimate of the infection frequency, assuming that (1) many cells (>100s) are used in the assay, and (2) each cell in each well has the same probability of being infected.
- The p-value can be used to detect very strange, or unlikely, results in the limiting dilution assay. For instance, if several large wells are negative but several small wells are positive, the resulting p-value will be very low, indicating that there may have been cross-contamination, improper labeling of wells, insufficiently mixed cells, or another failure of the assay. (Frequent users of this calculator should note: Every 100 assays that you run, you should expect to see a p-value below 0.01, just by chance!)
- The log-likelihood of the data is of mostly technical interest.
- For all-negative well outcomes, the maximum likelihood estimate is formally zero. To provide more useful information, a posterior median estimate is given instead. This value reports the median of the Bayesian posterior distribution, using a uniform prior. If you had a reasonable prior belief that your assay could turn up at least one positive well, then this is a decent rough estimate. A higher bound is also provided, at the 95th percentile of the posterior distribution; this is a good upper bound. The p-value and log-likelihood are not meaningful outputs in this case, and so they are not shown.
- If you have all-positive assay results, then close this app and use smaller wells!
This calculator features three ways to input data. The "Custom Well Sizes" pane allows you to input data with maximum flexibility. The "Constant Limiting Dilution" pane, on the other hand, will automatically compute your well sizes given the starting well size and a dilution factor. While these are both useful, regular users may prefer the convenience of pasting data in bulk or uploading it from a text file. To use this feature, format your text such that each row has three numbers: the well size (# cells), the number of replicates for this size, and the number of positive outcomes for this size. The upload/paste feature also allows you input data for multiple
patients. In this mode, the app will produce a graph showing the infection frequencies (IUPM) across the set of patients.
A Layman's Guide to the Assay
When HIV infects a patient, the virus actively reproduces in some cells, while in others it lies dormant after integration into the host genome. Because the virus betrays no sign of itself in these cells, such latent viral reservoirs present a major obstacle to treatment. A cell harboring quiescent HIV can turn on at any moment. How can we query the size of the latent viral reservoir in a quantitative way? Suppose, for the sake of argument, a patient undergoing HIV treatment has 10^12 T-cells of which 10^6 have latent, proviral HIV. One can imagine putting the T-cells into a high-throughput sequencing machine and making an estimation based on coverage depth of the viral genome. But here there is another layer of nuance: just because a cell has provirus, it is not necessarily replication-competent. The virus could have integrated into a non-transcribed region of the genome or, perhaps, has too many mutations to be viable. So what we want to know is, how many cells in the latent reservoir are actually capable of kickstarting a new infection? This, after all, is the number that the patient has to worry about.
In the assay, there are wells each of which has a certain number of chambers (confusingly called "cells") which contain patient blood. If any single cell has blood with replication-competent HIV in it, the whole well registers a binary positive signal. For example, here's a schematic with unrealistically low numbers (3 wells, 9 cells per well, 1 positive):
Returning to our hypothetial example, suppose that the million cells in the reservoir can all make competent, replicating virus. If there were 100 wells each with a million cells in it, chances are that all wells would be positive. In this case, the calculator would not be able to tell you anything interesting. However, if your wells were smaller than a million cells and not all of them turned positive, the calculator would predict the proportion of latent, replication-competent HIV.
Chun, T.-W., Carruth, L., Finzi, D., Shen, X., DiGiuseppe, J. A., Taylor, H., Hermankova, M., Chadwick, K., Margolick, J., Quinn, T. C., Kuo, Y.-H., Brookmeyer, R., Zeiger, M. A., Barditch-Crovo, P., Siliciano, R. F.
Quantification of latent tissue reservoirs and total body viral load in HIV-1 infection
, 183–188 (1997).
Laird G. M., Rosenbloom D. I. S., Lai J., Siliciano R. F., Siliciano J. D. Measuring the frequency of latent HIV-1 in resting CD4+ T cells using a limiting dilution co-culture assay. To appear in HIV Protocols