# PX1/PIN and ex-vivo antimalarial susceptibility: partial reanalysis

**Outcome:** Original experiment partially reproduced. This is a retrospective reanalysis of author-released summary-level source data, not a new cohort analysis, a reproduction of the adjusted model, or evidence of clinical treatment failure.

## Research question

In the released Ugandan *Plasmodium falciparum* source table, are ex-vivo susceptibility measurements different between source-labelled PX1 PIN and LMD haplotypes, including among samples labelled K13 wild type?

The target paper describes PIN as three PX1 substitutions plus two deletions. PX1 is the phosphoinositide-binding protein locus `PF3D7_0720700`. Here, genotype labels are used as supplied; this analysis does not call or phase the full haplotype from reads.

## Data and selection

Inputs were the 2026 article's `Source_dataset1` workbook and WHO's Version 2.0 antimalarial-marker compendium. The two downloaded workbooks, retrieval metadata and checksums are documented in [`data_manifest.json`](../data_manifest.json). Raw data stay in the ignored local `.cache` directory and are not emitted in outputs.

The phenotype workbook contains 1,346 rows with unique sample identifiers. Applying the available source fields and the PIN/LMD, WSAF > 0.9, and K13 categories used by the authors' analysis code leaves 772 rows. This count is not the paper's depth-qualified 465-sample analytic cohort: the workbook has no numeric per-sample sequencing depth, and the original article applied a >=25x threshold. Assay-specific complete cases among the 772 filtered rows were:

| Assay | Complete cases | PIN | LMD | Excluded for missing assay value |
| --- | ---: | ---: | ---: | ---: |
| DHA IC50 | 409 | 156 | 253 | 363 |
| Lumefantrine IC50 | 415 | 157 | 258 | 357 |
| Mefloquine IC50 | 408 | 154 | 254 | 364 |
| DHA ring-stage survival (RSA) | 179 | 115 | 64 | 593 |

The released workbook did not let us independently validate the full target-paper sample matching or inclusion logic. No genomic samples from Pf8 or the paper's SRA BioProject were joined to phenotypes, because a valid sample-level link was not established.

## Analysis

For each assay, PIN and LMD distributions were compared with a two-sided Mann–Whitney/Wilcoxon rank-sum test using a tie-corrected normal approximation and continuity correction. The effect summary is the difference in group medians (PIN minus LMD). A deterministic, within-group nonparametric bootstrap (4,000 resamples; seed `20261029`) gives percentile 95% intervals for the median difference. Benjamini–Hochberg q-values are calculated within the four-assay family. K13-wild-type comparisons are reported as a secondary stratum with their own four-test adjustment.

This is unadjusted. It does not reproduce the paper's mixed-effects regression controlling for site, parasitemia, COI, year and K13. The required sample-matched COI values and numeric coverage values were not in the released table.

## Results

In the filtered released table, PIN had higher median IC50 values than LMD for:

| Assay | PIN median (n) | LMD median (n) | Difference, PIN−LMD | Bootstrap 95% interval | Approx. p | BH q |
| --- | ---: | ---: | ---: | ---: | ---: | ---: |
| DHA IC50 (nM) | 3.71 (156) | 1.80 (253) | 1.91 | 1.59 to 2.46 | 2.14e-19 | 4.28e-19 |
| Lumefantrine IC50 (nM) | 15.14 (157) | 6.25 (258) | 8.89 | 7.40 to 11.61 | 1.05e-20 | 4.21e-20 |
| Mefloquine IC50 (nM) | 16.95 (154) | 11.13 (254) | 5.83 | 3.50 to 8.97 | 9.82e-11 | 1.31e-10 |
| RSA survival (%) | 4.07 (115) | 3.10 (64) | 0.97 | −1.08 to 2.72 | 0.263 | 0.263 |

For K13-wild-type records, the DHA medians were 3.70 nM for PIN (n=115) and 1.77 nM for LMD (n=236), difference 1.93 nM (bootstrap interval 1.56 to 2.53; approximate p=5.60e-16). Lumefantrine medians were 14.46 nM (n=115) and 6.10 nM (n=241), difference 8.36 nM (6.86 to 11.43; p=1.04e-16). RSA was not statistically distinguishable in this stratum (difference 0.72 percentage points; interval −1.68 to 2.89; p=0.572). See all assay-stratum rows and q-values in [`phenotype-comparisons.csv`](phenotype-comparisons.csv) and the machine-readable [`summary.json`](summary.json).

The direction for DHA, lumefantrine and mefloquine matches the article's reported direction; RSA remains non-significant, also in line with the article. But counts differ: for K13-wild-type DHA, this table-filtered result is PIN n=115/LMD n=236, versus n=110/240 in the paper; the paper reports 465 samples meeting its depth and allele-fraction criteria overall, which this workbook cannot reconstruct. Matching medians and direction do not demonstrate that we reproduced the same sample set.

The [WHO compendium](https://www.who.int/tools/compendium-of-molecular-markers-for-antimalarial-drug-resistance) lists K13 C469Y and A675V as validated markers in the downloaded Version 2.0 sheet. It does not classify PX1/PIN as a validated molecular marker. PIN should not be conflated with WHO's K13 partial-artemisinin-resistance markers.

## Interpretation and limitations

These results support an **association in the released ex-vivo dataset** between the source-labelled PIN haplotype and higher IC50 values for DHA, lumefantrine and mefloquine. RSA did not show a statistically clear difference. These laboratory measurements are not patient outcomes; this analysis cannot establish causation, predict treatment failure, or estimate treatment efficacy.

The result is partial because we could not apply the paper's numeric >=25x depth filter, reconstruct its exact 465-sample cohort, or fit its adjusted mixed-effects model without the sample-matched COI and full source/join inputs. It is a reanalysis of published measurements, not independent biological replication or a new discovery. The paper's article page states CC BY-NC-ND terms; dataset-specific and commercial reuse permissions should be checked before redistribution.

## Reproduction

From the repository root, with Node.js 20+ and the repository's `fflate` dependency:

```powershell
node experiments/malaria-resistance/scripts/fetch-sources.mjs
node experiments/malaria-resistance/scripts/analyze.mjs
node --test experiments/malaria-resistance/tests/analyze.test.mjs
```

The fetcher enforces file-size limits and stores source workbooks only in ignored `.cache`. An executable Colab version is in [`notebooks/px1_malaria_reanalysis.ipynb`](../notebooks/px1_malaria_reanalysis.ipynb); it has not been run remotely from this environment. The local Node analysis and tests were run successfully. No Supabase, cloud compute, author R script, or live experiment runner was used.

## Sources

- Niaré K et al. [Emergence and spread of *Plasmodium falciparum* PX1 polymorphisms associated with decreased susceptibility to antimalarials in Uganda](https://www.nature.com/articles/s41591-026-04590-5). *Nature Medicine* (2026), DOI [10.1038/s41591-026-04590-5](https://doi.org/10.1038/s41591-026-04590-5). The paper reports the 465 depth-qualified samples, per-assay comparisons, K13-WT results and adjusted model covariates.
- World Health Organization. [Compendium of molecular markers for antimalarial drug resistance](https://www.who.int/tools/compendium-of-molecular-markers-for-antimalarial-drug-resistance), Version 2.0 workbook (July 2026; downloaded 2026-10-29).
- MalariaGEN. [Pf8 data access guide](https://malariagen.github.io/parasite-data/pf8/Data_access.html). Consulted but not used for this phenotype analysis; the current guide describes authenticated Google Cloud access.