Strong isolation by distance and evidence of population microstructure reflect ongoing Plasmodium falciparum transmission in Zanzibar

Reviewer #1 (Public Review):

Summary:

Zanzibar archipelago is close to achieving malaria elimination, but despite the implementation of effective control measures, there is still a low-level seasonal malaria transmission. This could be due to the frequent importation of malaria from mainland Tanzania and Kenya, reservoirs of asymptomatic infections, and competent vectors. To investigate population structure and gene flow of P. falciparum in Zanzibar and mainland Tanzania, they used 178 samples from mainland Tanzania and 213 from Zanzibar that were previously sequenced using molecular inversion probes (MIPs) panels targeting single nucleotide polymorphisms (SNPs). They performed Principal Component Analysis (PCA) and identity by descent (IBD) analysis to assess genetic relatedness between isolates. Parasites from coastal mainland Tanzania contribute to the genetic diversity in the parasite population in Zanzibar. Despite this, there is a pattern of isolation by distance and microstructure within the archipelago, and evidence of local sharing of highly related strains sustaining malaria transmission in Zanzibar that are important targets for interventions such as mass drug administration and vector control, in addition to measures against imported malaria.

Strengths:

This study presents important samples to understand population structure and gene flow between mainland Tanzania and Zanzibar, especially from the rural Bagamoyo District, where malaria transmission persists and there is a major port of entry to Zanzibar. In addition, this study includes a larger set of SNPs, providing more robustness for analyses such as PCA and IBD. Therefore, the conclusions of this paper are well supported by data.

Weaknesses:

Some points need to be clarified:

1. SNPs in linkage disequilibrium (LD) can introduce bias in PCA and IBD analysis. Were SNPs in LD filtered out prior to these analyses?
2. Many IBD algorithms do not handle polyclonal infections well, despite an increasing number of algorithms that are able to handle polyclonal infections and multiallelic SNPs. How polyclonal samples were handled for IBD analysis?

Reviewer #2 (Public Review):

Summary:

This manuscript describes P. falciparum population structure in Zanzibar and mainland Tanzania. 282 samples were typed using molecular inversion probes. The manuscript is overall well-written and shows a clear population structure. It follows a similar manuscript published earlier this year, which typed a similar number of samples collected mostly in the same sites around the same time. The current manuscript extends this work by including a large number of samples from coastal Tanzania, and by including clinical samples, allowing for a comparison with asymptomatic samples.

The two studies made overall very similar findings, including strong small-scale population structure, related infections on Zanzibar and the mainland, near-clonal expansion on Pemba, and frequency of markers of drug resistance. Despite these similarities, the previous study is mentioned a single time in the discussion (in contrast, the previous research from the authors of the current study is more thoroughly discussed). The authors missed an opportunity here to highlight the similar findings of the two studies.

Strengths:

The overall results show a clear pattern of population structure. The finding of highly related infections detected in close proximity shows local transmission and can possibly be leveraged for targeted control.

Weaknesses:

A number of points need clarification:

It is overall quite challenging to keep track of the number of samples analyzed. I believe the number of samples used to study population structure was 282 (line 141), thus this number should be included in the abstract rather than 391. It is unclear where the number 232 on line 205 comes from, I failed to deduct this number from supplementary table 1.

Also, Table 1 and Supplementary Table 1 should be swapped. It is more important for the reader to know the number of samples included in the analysis (as given in Supplementary Table 1) than the number collected. Possibly, the two tables could be combined in a clever way.

Methods
The authors took the somewhat unusual decision to apply K-means clustering to GPS coordinates to determine how to combine their data into a cluster. There is an obvious cluster on Pemba islands and three clusters on Unguja. Based on the map, I assume that one of these three clusters is mostly urban, while the other two are more rural. It would be helpful to have a bit more information about that in the methods. See also comments on maps in Figures 1 and 2 below.

Following this point, in Supplemental Figure 5 I fail to see an inflection point at K=4. If there is one, it will be so weak that it is hardly informative. I think selecting 4 clusters in Zanzibar is fine, but the justification based on this figure is unclear.

For the drug resistance loci, it is stated that "we further removed SNPs with less than 0.005 population frequency." Was the denominator for this analysis the entire population, or were Zanzibar and mainland samples assessed separately? If the latter, as for all markers <200 samples were typed per site, there could not be a meaningful way of applying this threshold. Given data were available for 200-300 samples for each marker, does this simply mean that each SNP needed to be present twice?

Discussion:
I was a bit surprised to read the following statement, given Zanzibar is one of the few places that has an effective reactive case detection program in place: "Thus, directly targeting local malaria transmission, including the asymptomatic reservoir which contributes to sustained transmission (Barry et al., 2021; Sumner et al., 2021), may be an important focus for ultimately achieving malaria control in the archipelago (Björkman & Morris, 2020)." I think the current RACD program should be mentioned and referenced. A number of studies have investigated this program.

The discussion states that "In Zanzibar, we see this both within and between shehias, suggesting that parasite gene flow occurs over both short and long distances." I think the term 'long distances' should be better defined. Figure 4 shows that highly related infections rarely span beyond 20-30 km. In many epidemiological studies, this would still be considered short distances.

Lines 330-331: "Polymorphisms associated with artemisinin resistance did not appear in this population." Do you refer to background mutations here? Otherwise, the sentence seems to repeat lines 324. Please clarify.

Line 344: The opinion paper by Bousema et al. in 2012 was followed by a field trial in Kenya (Bousema et al, 2016) that found that targeting hotspots did NOT have an impact beyond the actual hotspot. This (and other) more recent finding needs to be considered when arguing for hotspot-targeted interventions in Zanzibar.

Figures and Tables:
Table 2: Why not enter '0' if a mutation was not detected? 'ND' is somewhat confusing, as the prevalence is indeed 0%.

Figure 1: Panel A is very hard to read. I don't think there is a meaningful way to display a 3D-panel in 2D. Two panels showing PC1 vs. PC2 and PC1 vs. PC3 would be better. I also believe the legend 'PC2' is placed in the wrong position (along the Y-axis of panel 2).

Supplementary Figure 2B suffers from the same issue.

The maps for Figures 1 and 2 don't correspond. Assuming Kati represents cluster 4 in Figure 2, the name is put in the wrong position. If the grouping of shehias is different between the Figures, please add an explanation of why this is.

Figure 2: In the main panel, please clarify what the lines indicate (median and quartiles?). It is very difficult to see anything except the outliers. I wonder whether another way of displaying these data would be clearer. Maybe a table with medians and confidence intervals would be better (or that data could be added to the plots). The current plots might be misleading as they are dominated by outliers.

In the insert, the cluster number should not only be given as a color code but also added to the map. The current version will be impossible to read for people with color vision impairment, and it is confusing for any reader as the numbers don't appear to follow any logic (e.g. north to south).

The legend for Figure 3 is difficult to follow. I do not understand what the difference in binning was in panels A and B compared to C.

Font sizes for panel C differ, and it is not aligned with the other panels.

Why is Kusini included in Supplemental Figure 4, but not in Figure 1?

Supplemental Figures 6 and 7: What does the width of the line indicate?

What was the motivation not to put these lines on the map, as in Figure 4A? This might make it easier to interpret the data.