We screened 873 febrile children <5 years of age across the 8 health facilities. Of those, 423 met the inclusion criteria and completed clinical follow-up through day 3, corresponding to 1 day after completion of artemether/lumefantrine treatment. Of the 423 samples collected, 365 (86.3%) day 0 samples were successfully confirmed as P. falciparum infections and subsequently analyzed by molecular sequencing to address the secondary objective: detection of pfkelch13 mutations, notably R561H, a validated marker of ART-R. Sequencing success varied by district: 100% in Kiremba and Ngozi, 91% in Kirundo, 75% in Buye, and 76% in Vumbi. We excluded the other 58 samples (13.7%) because of insufficient DNA quality.

The median age of enrolled children was 30 (IQR 17–45) months, and the sex ratio was balanced (1.0). Baseline characteristics showed variation across districts (Table). Median axillary temperatures ranged from 37.8°C (Buye) to 39.0°C (Kirundo). Pretreatment asexual P. falciparum parasite density varied widely; median values ranged from 10,543/µL (Vumbi) to 42,400/µL (Kirundo). Gametocyte carriage was detected in 5.4% of children overall, with district-specific prevalence ranging from 1.2% (Kirundo) to 22.0% (Ngozi). Mixed-species infections were uncommon (1.2% P. malariae in Ngozi Province), confirming P. falciparum as the predominant species.

Among the 423 children enrolled, 19 (4.5%; 95% CI 2.7%–6.9%) remained parasitemic on day 3 after artemether/lumefantrine treatment. That overall prevalence was below the WHO 10% threshold, but we observed marked heterogeneity across districts. Vumbi (13.6%) exceeded the WHO 10% threshold and neighboring districts, identifying a localized hotspot of delayed clearance. Those differences were statistically significant within Kirundo Province districts (p = 0.0002), but we observed no significant variation between provinces (p = 0.06).

The median asexual parasite density on day 3 was 260 parasites/µL (IQR 141–1,659 parasites/µL), with no significant variation across districts (p = 0.96). Nevertheless, district-specific values showed marked variation, from very low medians in Buye and Kiremba (≈100 parasites/µL) to >1,800 parasites/µL in Ngozi district. Gametocytes were detected microscopically in 11 (2.6%) children on day 3. Their prevalence varied geographically: uncommon in Kirundo Province (1.2%) but more frequent in Ngozi Province (4.8%) and peaked at 10.0% in Kiremba district (Table).

For day 0 isolates, of the 423 children enrolled, 365 (86.3%) samples yielded high-quality sequences and were included in the molecular analysis (Appendix Tables 1, 2). Most (97.5%, 356/365), isolates were wild-type at pfkelch13. We detected no validated or candidate WHO-associated mutations linked to ART-R. We did observe nonsynonymous variants in 9 (2.5%) isolates. The most frequent variant was A578S (4 cases, 1.1%), whereas R513S, A626S, H644R, and N672H were each detected in 1 or 2 isolates. None of those polymorphisms are currently associated with delayed parasite clearance, but their detection might indicate ongoing low-level diversification of the pfkelch13 locus in this setting.

Mutant pfcrt haplotypes were nearly fixed. The CVIET triple mutant, defined by mutations at pfcrt codons 72–76 (C72/V73/I74/E75/T76), predominated (84.4%, 308/365), whereas the CVIET+356T haplotype was detected in 9.3% (34/365) of isolates. The CVMNK wild-type was uncommon (4.9%, 18/365). Minor variants included SVMNT (72S/76T, 1.1%) and 356T alone (0.3%). We observed no significant difference in overall pfcrt haplotype distribution between provinces (p = 0.25). However, intraprovincial heterogeneity was evident, particularly for codon 356 in Ngozi Province (p = 0.01), suggesting microgeographic clustering of resistant variants. The predominance of CVIET confirms that chloroquine resistance remains highly fixed in northern Burundi, with additional mutations such as 356T emerging at low frequency. Of note, we detected no pfcrt mutations associated with piperaquine resistance. All isolates were wild-type at positions H97, F145, I218, M343, C350, and G353. Although the 356T variant occurred in 9.6% (35/365) of isolates, that mutation alone does not confer piperaquine resistance.

Mutant haplotypes were common at the pfmdr1 locus. We identified the wild-type NYSND haplotype, defined by pfmdr1 codons 86-184-1034-1042-1246 (N86/Y184/S1034/N1042/D1246), in 31.2% (114/365) of isolates. The most frequent mutant was the single 184F allele, detected in 57.3% (209/365) of isolates, followed by 86Y (7.9%). Additional combinations involving 86Y, 1034C, 1042D, or 1246Y remained uncommon (<1.6% each). We noted significant differences in haplotype distribution between provinces (p = 0.03), suggesting that selective pressures on pfmdr1 might vary geographically, potentially reflecting differences in drug use or drug pressure intensity.

At the dhfr locus, resistance alleles were nearly fixed. The triple mutant 51I/59R/108N predominated (92.1%, 336/365), confirming widespread pyrimethamine resistance. Other haplotypes were uncommon, including the double mutant 51I/108N (1.1%) and the single 108N variant (0.5%), and we detected the quadruple mutant 51I/59R/108N/164L in 12 (3.3%) isolates. Although infrequent, the quadruple mutant genotype is clinically relevant because the addition of 164L confers high-grade pyrimethamine resistance and has been associated with sulfadoxine/pyrimethamine treatment failure in East Africa. We detected the wild-type haplotype in only 12 (3.3%) isolates. Variation between provinces was not statistically significant (p = 0.2).

Mutations at the dhps locus were also widespread. Only 21/365 (5.8%) isolates were wild-type. The double mutant 437G/540E was detected in 148/365 (40.5%) isolates, whereas the triple mutant 437G/540E/581G was even more frequent (172/365 isolates, 47.1%). Minor haplotypes included 436A (3.8%) and 613S (0.3%). Although we observed no significant difference in overall haplotype proportions between provinces (p = 0.6), district-level variation within Kirundo Province was significant for codons 436A (p = 0.0001), 437G (p = 0.001), and 540E (p = 0.02).

Analysis of combined genotypes enabled inference of multilocus dhfr/dhps haplotypes, highlighting the extensive accumulation of antifolate resistance alleles. The fully wild-type dhfr/dhps combination was nearly absent (1/365 isolates, 0.3%). The most prevalent inferred haplotype was the sextuple mutant (dhfr triple 51I/59R/108N + dhps triple 437G/540E/581G), detected in 161/365 (44.1%) isolates. That genotype has been associated with markedly reduced efficacy of IPTp in East Africa. The classic quintuple mutant (dhfr triple 51I/59R/108N + dhps double 437G/540E) was also common, found in 134/365 (36.7%) isolates. Together, those 2 high-resistance genotypes accounted for >80% of infections. Less frequent combinations included quadruple or septuple haplotypes, each detected in <3% of isolates. Although overall haplotype distribution did not differ significantly between provinces (p = 0.2), variation occurred at the district level.

We performed molecular analysis on isolates from the 19 children who remained parasitemic on day 3. All isolates were wild-type at pfkelch13, confirming the absence of validated or candidate ART-R mutations. At pfcrt, the CVIET haplotype predominated (84.2%), and additional variants included CVIET+356T (10.5%) and SVMNT (5.3%). At pfmdr1, the 184F allele occurred in 57.9% of isolates, consistent with findings from day 0. Mutations at the dhfr locus were nearly fixed, the triple mutant 51I/59R/108N was present in 94.7% of isolates. At dhps, the double (437G/540E) and triple (437G/540E/581G) mutants together accounted for >85% of samples. Inferred dhfr/dhps haplotypes showed a high prevalence of quintuple and sextuple genotypes, mirroring the overall population.

Paired analysis of baseline (day 0) and follow-up (day 3) samples revealed that most isolates retained identical molecular profiles across time points. Nonetheless, we observed a few discordances, particularly in pfmdr1 (184F/Y) and dhps alleles (e.g., detection of 581G or 164L in some day 3 samples). Those differences likely reflect within-host selection of minority clones under drug pressure rather than de novo mutations.

ART-R is a growing threat to malaria control in sub-Saharan Africa, and validated pfkelch13 mutations are already established in Rwanda, Uganda, and Tanzania (8,9,13,17–20). Of particular concern is the detection of the pfkelch13 R561H mutation, a validated marker of delayed parasite clearance, in the Huye district of Rwanda and the Kagera region of Tanzania (8–10,12–14,21,22). Those findings raise the alarm over possible spread into neighboring countries, including Burundi, underscoring the urgent need for localized therapeutic and molecular surveillance.

We adopted a simplified surveillance design to detect ART-R, combining day 3 parasitemia assessment with pfkelch13 molecular screening. That approach provides complementary clinical and genetic indicators without replacing traditional therapeutic efficacy studies, which remain essential for evaluating ACT efficacy and informing treatment policy. The limited follow-up period and dried blood spot collection enable rapid geographic mapping of ART-R risk across multiple sites in resource-constrained contexts (23).

In 5 districts across northern Burundi, 4.5% of children treated with artemether/lumefantrine remained parasitemic on day 3, below the 10% WHO threshold for confirmed ART-R (24), but the Vumbi district exceeded the WHO threshold. However, the sample size was calculated for surveillance purposes and was not powered to detect statistically significant differences between districts. Consequently, the geographic comparisons we describe should be interpreted as exploratory findings, rather than definitive evidence of spatial heterogeneity.

None of the day 3 positivity cases harbored validated or candidate pfkelch13 mutations. That discordance suggests either false-positive results driven by imperfect adherence to the 6-dose artemether/lumefantrine regimen (25,26), variability in lumefantrine exposure (food intake, vomiting, drug–drug interactions), age-dependent immunity in young children (27), or genuine delayed clearance mediated by polygenic architecture of artemisinin response (e.g., coronin, ap2µ, or ubp-1) (28,29). Pharmacokinetics/adherence verification and broader genomic profiling are needed to resolve those alternatives.

The overall clinical response to artemether/lumefantrine treatment appeared satisfactory despite 4.5% day 3 positivity, likely because lumefantrine, the partner drug with a longer half-life, remains fully effective at eliminating residual parasites. The pfmdr1 mutations observed (low N86, high 184F) indicate drug pressure but are not validated markers of lumefantrine failure. Those findings suggest lumefantrine efficacy is currently maintaining artemether/lumefantrine effectiveness in Burundi, underscoring the importance of monitoring both artemisinin and partner drug susceptibility.

The absence of pfkelch13 mutants in Burundi, despite their presence in neighboring countries, might reflect intense transmission (5,7), where parasite diversity and immunity limit resistant lineage expansion. However, pfkelch13 mutations have been documented in high-transmission areas of Uganda (20), demonstrating that transmission intensity alone does not prevent resistance selection or spread. The absence of validated mutations should not be interpreted as protection. Importation risk remains high, given porous borders and regional connectivity, underscoring the urgency of cross-border surveillance (24).

Beyond ART-R, this study revealed a concerning accumulation of antifolate resistance. The dhfr triple mutant was nearly fixed, and >80% of infections carried highly resistant multilocus haplotypes. At the dhps locus, mutations were also highly prevalent. The 437G/540E/581G triple mutant was present in 47.1% of isolates, whereas only 5.8% of parasites remained wild-type. That pattern indicates widespread sulfadoxine resistance, with direct implications for malaria prevention in Burundi. Sulfadoxine/pyrimethamine is used for IPTp and more recently in PMC. The high prevalence (44%) of the sextuple haplotype raises serious concerns because that genotype is associated with reduced IPTp effectiveness and adverse birth outcomes and likely compromises PMC efficacy. Those findings warrant urgent reassessment of sulfadoxine/pyrimethamine-based prevention strategies in Burundi (30–32).

ACT partner-drug resistance markers were entrenched. The pfcrt CVIET haplotype (74I/75E/76T) predominated (84.4%), with an additional subset carrying 356T (9.3%). Wild-type CVMNK was infrequent (4.9%). Those results confirm that chloroquine resistance remains essentially fixed in the parasite population, consistent with observations across East Africa, despite chloroquine withdrawal (33,34). That pattern contrasts with observations from several other countries in Africa, where chloroquine-sensitive parasites have reexpanded after drug withdrawal (35). The persistence of CVIET in Burundi likely reflects the use of artesunate/amodiaquine as first-line treatment until 2019, which maintained selective pressure favoring pfcrt mutant parasites. Similar persistence of CVIET has been reported in other countries that used artesunate/amodiaquine, such as South Sudan. A decline in CVIET prevalence might require several more years to become apparent (36–39).

We detected no pfcrt mutations associated with piperaquine resistance (e.g., F145I, H97Y, M343L, G353V) in this study (40–42). Those mutations have been identified as key drivers of dihydroartemisinin/piperaquine treatment failures in Southeast Asia, but they appear absent in East Africa parasite populations.

At the pfmdr1 locus, the Y184F mutation was the dominant variant (57.3%), whereas the wild-type NYSND was found in 31.2%. Other variants involving 86Y, 1034C, 1042D, or 1246Y were uncommon (<1.6%). The predominance of 184F is consistent with reports from Zanzibar (Tanzania) and Senegal (43,44). The Y184F mutation is frequently selected by artemether/lumefantrine treatment and can modulate parasite responses in vitro, but it should be regarded as a marker of drug pressure rather than a validated clinical marker of lumefantrine resistance (45).

The limitations of this study include partially supervised drug treatment, unmeasured drug concentrations, day 3 follow-up only, and no assessment of copy number variation in pfmdr1 and plasmepsin 2/3, the established markers associated with reduced lumefantrine (46–48) and piperaquine (40,49,50) susceptibility. Study strengths include multiple sentinel sites in a high-risk border region, combined clinical and molecular surveillance, and an integrated baseline dataset informing national and regional strategies.

WHO recommends deploying multiple first-line ACTs to delay malaria drug resistance emergence. In Burundi, where artemether/lumefantrine is the sole first-line treatment, introducing an alternative such as dihydroartemisinin/piperaquine could reduce selective pressure. The absence of pfcrt mutations associated with piperaquine resistance suggests dihydroartemisinin/piperaquine could retain full efficacy and serve as a second first-line option.

In conclusion, our findings highlight a critical juncture for Burundi. We detected no pfkelch13 mutants, but excess localized day 3 positivity and entrenched sulfadoxine/pyrimethamine resistance represent clear warning signals. High-transmission dynamics may currently protect Burundi against pfkelch13-mediated ART-R, but that protection is fragile. Strengthened therapeutic efficacy monitoring, expanded molecular surveillance, including genomic approaches, and cross-border collaboration are urgently required to prevent resistant P. falciparum establishment and spread in the Great Lakes region of Burundi.