Results

Bioassay findings

The dose–response relationship of the B. germanica populations collected from four different locations was evaluated using probit regression. In all locations, there was a statistically significant positive association between the log-transformed dose and mortality (p < 0.001), indicating that the probability of death increased with increasing doses of the insecticide (Fig. 1). The estimated coefficients of the probit models are presented in Table 1**.** The Pearson chi-square goodness-of-fit test yielded non-significant P-value for all populations (P > 0.05, ranging 0.0598 to 0.392) indicates that the model adequately describes the observed dose-mortality relationship (Table 1). The intercept values differed among locations, suggesting variability in baseline susceptibility, while the slopes were all positive and significant, reflecting a consistent dose-dependent increase in mortality.

Fig. 1: Probit model predicts mortality in [location 1: Dormitory, Location 2: Apartment, Location 3: Hospital, Location 4: Fast-food restaurant and Location 5: laboratory standard strain]. LD50 (solid), LD90 (dashed), LD95 (dash-dot), LD99 (dotted).

Strains | Estimate | St. Error | P-value | 95% confidence intervals | P-value for Pearson chi-square goodness-of-fit statistics
Dormitory | (Intercept) | −0.11 | 0.06 | <0.0001 | −0.247, 0.010 | 0.392
log dose | 0.49 | 0.03 | 0.428, 0.557
Apartment | (Intercept) | −0.59 | 0.08 | <0.0001 | −0.766, −0.443 | 0.0843
log dose | 0.55 | 0.04 | 0.475, 0.636
Hospital | (Intercept) | −2.02 | 0.20 | <0.0001 | −2.417, −1.667 | 0.121
log dose | 0.91 | 0.08 | 0.751, 1.093
Fast-food restaurant | (Intercept) | −2.30 | 0.22 | <0.0001 | −2.745, −1.901 | 0.176
log dose | 1.07 | 0.10 | 0.889, 1.274
Susceptible strain | (Intercept) | 0.75 | 0.10 | <0.0001 | 0.607, 0.916 | 0.0598
log dose | 0.55 | 0.03 | 0.481, 0.627

The probit models were used to calculate the lethal doses for 50%, 90%, 95% and 99% mortality (LD50, LD90, LD95 and LD99) for each population (Table 2).

Strains | N | LD50 (50% CL) | LD90 (90% CL) | LD95(95% CL) | LD99 (99% CL) | Resistance ratio based on LD50 | Resistance Status (Standard Criteria)
Dormitory | 540 | 1.21 | 17.17 | 36.03 | 144.32 | 1.77 | R > 1/low resistance
Apartment | 540 | 2.91 | 30.04 | 58.02 | 169.28 | 4.27 | R < 5/ moderate resistance
Hospital | 540 | 9.08 | 36.91 | 54.88 | 115.46 | 13.35 | R > 10/ high resistance
Fast-food restaurant | 540 | 8.49 | 28.10 | 39.42 | 74.35 | 12.48 | R > 10/ high resistance
Susceptible Strain | 180 | 0.68 | 2.53 | 4.93 | 17.08 | – | –

Susceptible populations had the lowest LD50, LD90, LD95 and LD99 values, indicating the greatest insecticide sensitivity. Hospital and Fast-food restaurant strains showed the highest resistance ratios (12.48 and 13.35), indicating relative resistance, while dormitory and apartment populations exhibited intermediate susceptibility with resistance ratios of 1.77 and 3.36, respectively. These findings demonstrate substantial variation in insecticide sensitivity across populations from different locations. Such differences may reflect prior exposure to insecticides, environmental conditions, or genetic variation among populations.

Comparison of mortality between treatment groups and control

To determine whether fipronil concentrations caused significantly higher mortality than the control, a chi-square test of independence was performed for each concentration within each population. In the susceptible strain, all tested concentrations (0.001–25 ppm) resulted in significantly higher mortality than the control (all p < 0.001). In the low-resistant Dormitory strain, the lowest concentration (0.001 ppm) already showed a significant difference (χ2 = 9.730, df = 1, p = 0.0018), and all higher concentrations were also significant (p < 0.001).

In the high resistant Apartment strain the concentration of 0.001 ppm did not differ significantly from the control (χ2 = 2.034, df = 1, p = 0.1538), but concentrations ≥0.1 ppm were all significantly different (p ≤ 0.0064). In the highly resistant Hospital strain, no mortality was observed at 0.001 and 0.1 ppm, and the first significant difference appeared at 1 ppm (χ2 = 5.217, df = 1, p = 0.0224). In the Fast-food restaurant strain, also highly resistant, concentrations of 0.001 and 0.1 ppm yielded no mortality, and even at 1 ppm the difference was not significant (χ2 = 3.077, df = 1, p = 0.0794); significant differences started at 2.5 ppm (χ2 = 7.434, df = 1, p = 0.0064). These chi-square results are fully consistent with the resistance ratios derived from probit analysis, confirming that higher fipronil resistance is associated with a higher threshold dose required to achieve a significant increase in mortality over the control.

Correlation analysis between dose and mortality

In addition to probit regression, Spearman correlation analyses were performed between log-transformed dose and mortality proportion for each population (using concentrations >0 ppm). Positive correlations were observed in all populations. Spearman correlation coefficients ranged from 0.815 (Hospital strain, p = 0.0074) to 0.979 (susceptible strain, p < 0.001). These results independently confirm a dose-dependent increase in fipronil-induced mortality, supporting the validity of the probit model.

Mutation finding

We analyzed a 245-bp fragment of the Rdl (GABA receptor) gene encompassing the A302S, A301S and A299S mutation sites previously associated with phenylpyrazole (fipronil) resistance (Fig. 2). Samples included a susceptible strain and four field-collected B. germanica populations (n = 5 per population). Based on earlier reports linking fipronil resistance to the A301S and A299S substitutions, we screened these loci to assess their occurrence in the studied populations. The A302S mutation was not detected in any of the field populations, whereas the A299S mutation was identified in individuals from the Fast-food restaurant and hospital populations (Fig. 3**,** Table 3).

Fig. 2: Schematic diagram of polymorphic sites in the Rdl monomer. The phenylpyrazole resistance-associated substitution A299S is indicated by a blue dot, the A301S substitution by a green dot, and the well-characterized A302S resistance-associated substitution by a red dot. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3: A: Nucleotide sequences of the TM2 region of the B. germanica Rdl gene, which includes the point mutation that results in the A302S, A301S and A299S substitution. Representative sequences from four field-collected German cockroach populations were aligned against the reference sequence (MW267921.1), with the A299S A302S and A301S, regions highlighted. B. Direct sequencing chromatograph showing two individual genotypes of the phenyl pyrazole insecticide resistance-related sites.

Populations | n | A302S(S/S) | A299S(A/S) | A301S(A/S)
Hospital | 5 | 5 | 1 | 0
Fast-food restaurant | 5 | 5 | 1 | 1
Apartment | 5 | 5 | 0 | 0
Dormitory | 5 | 5 | 0 | 0
Susceptible strain | 1 | 1 | 0 | 0

Haplotype network finding

Sequencing of a 245-bp fragment of the Rdl gene from 21 B. germanica individuals revealed 17 unique haplotypes, indicating exceptionally high genetic diversity (Hd = 1.000 ± 0.020) and substantial nucleotide diversity (π = 0.0405 ± 0.007), with 24 polymorphic sites and 27 mutations detected.

Haplotype network analysis identified 18 haplotypes across populations collected from dormitories, apartments, hospitals, fast-food restaurants, and a laboratory strain. Hap_18 (accession MW267921.1, USA) was positioned at the center of the network, suggesting it represents a common ancestral haplotype. Populations from hospitals and fast-food restaurants exhibited greater mutational distances and higher genetic divergence, whereas the laboratory strain showed reduced diversity and clustered closely around the central haplotype.

Two haplotypes (H1 and H13) carried the G → T substitution at position A299S and A → S in position 301, previously linked to phenylpyrazole insecticide resistance. Tajima's D was negative (D = −0.88357) but not statistically significant (p > 0.1), and the network displayed a star-shaped pattern (Fig. 4).

Fig. 4: Genealogy of Rdl haplotypes is represented as a network. Circle size corresponds to haplotype frequency across populations.